RESULTS OF THE CITIZEN DELIBERATIONS
The results of the citizens' deliberations were unveiled on June 14, 2018. You can now consult the summary and read the report in French. English version to come!
ACCOUNT OF THE DELIBERATIONS
of the Montreal Declaration AI responsible
RESULTS OF THE CITIZEN DELIBERATIONS
The results of the citizens' deliberations were unveiled on June 14, 2018. You can now consult the summary and read the report in French. English version to come!
On November 3, 2017, Université de Montréal launched the co-construction process for the Montreal Declaration for a Responsible Development of Artificial Intelligence (Montreal Declaration).
We had no idea of the interest the initiative would capture, nor of the size of the task that lay ahead. Eight months later, we present the ﬁrst results of the citizen deliberations, halfway through the process.It is a very successful one: dozens of events were held to discuss the social issue surrounding AI, and ﬁfteen deliberation workshops took place over three months, involving over 500 citizens, experts and stakeholders of all professional horizons.
The halfway report we are presenting must be taken as a temporary, non-exhaustive summary of a democratic deliberation process to enlighten public policy decisions regarding artiﬁcial intelligence. The work around what we call the Montreal Declaration was led by a multidisciplinary and inter-university team of researchers, mainly in Quebec but also across the world. Awareness of the social issues around artiﬁcial intelligence is shared by this research community, but also by society as a whole. We therefore suggested a citizen co-construction process because we believe everyone has a right to be heard about how our society should be. This approach is innovative in both content and form:ﬁrst, because it carries out a prospective design of applied ethics, because it’s a matter of anticipating ethical controversies around future artiﬁcial intelligence technologies or social situations where the use of these technologies is pushed to the limit of what we can anticipate. Then, we carried out this consultation process on an unheard-of scale. The numbers mentioned above paint a clear picture.This process, to be clear, will continue, and as the Montreal Declaration remains open to review, the co- construction will not end when this ﬁrst deliberative endeavour does.
We called the public around the drafting of the Declaration, and were called upon in return: what will the Declaration change? Who is writing? Isn’t this just a vain university professor thing? Aren’t there already too many manifestos, professions of faith on the ethical values of artiﬁcial intelligence? Isn’t surrounding the development of artiﬁcial intelligence with ethical principles and recommendations a way of condoning it? Isn’t that approving a technocratic vision of society? Why not devote our energy to criticizing this development? None of these interrogations are without merit, and because we are committed to increased transparency around artiﬁcial intelligence, we are also committed to increased transparency around human and collective intelligence. This halfway report will, we hope, provide a few answers.
The ethics of artiﬁcial intelligence have been a hot topic in many countries over the last two years. Every actor in its development, researchers, businesses, citizens, political representatives, all recognize the urgency of establishing an ethical, political and legal framework to guide the research and use of artiﬁcial intelligence. Because there is no doubt that we are at the dawn of a new industrial revolution with the rise of artiﬁcial intelligence technologies. The impacts of this revolution on the production of goods, the delivery of services, the organization of work and the workforce, or even on family and personal relationships are still unknown but will be major, possibly disruptive in certain ﬁelds. Indeed,the societal changes brought about by artiﬁcial intelligence are surprising in their suddenness and spark varied reactions, from enthusiasm to disapproval and scepticism. We could ignore them and launch into speculations around the existence or not of that we call artiﬁcial intelligence, but we’d only be pushing back the problem to a time when it will no longer be possible to inﬂuence its development.
Many objections and fears were raised during this ﬁrst co-construction process. Many workshop participants and observers in the Declaration project questioned the technocratic ideology that sees in technology a way of rationally organizing all of society, and that reduces social issues to technical problems. Others question the ability and the will of public institutions to regulate lucrative technologies. These objections must not be casually dismissed, because they are based on historical precedents that shook their faith in technological innovations, and even more so in the people promoting them.
But it is also important that those raising objections don’t undermine every effort to positively inﬂuence the future of society and support them by getting involved in the democratic deliberations that allow us to keep control. We can complain about the effects of new information technologies and artiﬁcial intelligence on social relationships, we can criticize the reduction of social life to a series of lifestyles, this will not prevent technological innovation, nor will it inﬂuence it. Yet that is the entire purpose of the Montreal Declaration: guide the development of artiﬁcial intelligence in order to promote or preserve fundamental ethical and societal interests.
In conclusion, we will not settle the unrelenting question regarding the use of the term “artiﬁcial intelligence”: is it appropriate to refer to data processing, recognition and decision-making algorithms? Its use can be contested by opposing the fact that artiﬁcial intelligence refers to very limited knowledge processes when compared with human intelligence, or even the intelligence of pigeons. It’s undeniable. But with that reasoning, paramecia offer complexity that surpasses that of any algorithm, even a learning one. By going down that path, you come across a deadlock of intelligence as a whole. What is human intelligence? The hundreds of thousands of pages that have been written to answer that question still doesn’t sufﬁce.
However, a few statements can help avoid misunderstandings that are at the root of the controversy: ﬁrstly, people often confuse intelligence and thought. Intelligence is a property of thought, it is not thought as a whole. Then, intelligence is particular in that it reduces the complexity of the world in which the intelligent being evolves to allow him to better master his environment. We give ourselves rules to analyze reality, calculate it, evaluate it and make decisions. A long philosophical tradition of thinkers that did not lack intelligence have claimed it from Socrates to Rusell, along with Leibniz. I a certain way, intelligence reduces reality to better act on it. Finally, stemming from the above, intelligence, even human, is largely algorithmic: it analyzes data and makes calculations according to procedures. Sometimes these procedures are inadequate, the analysis is wrong. But to establish it in the ﬁrst place, you must ﬁrst analyze and use procedures.
Reﬂecting on the goals we wish to pursue is not strictly a matter of calculations. Directing your personal and social life towards certain worthwhile goals does not depend on an algorithmic procedure. Knowing if we must use nuclear weapons to kill the greatest number of people and weaken an enemy country cannot be solely determined by a calculation of the consequences. There’s something tragic about avoiding reﬂection on moral consequences by seeking only a calculation of the means. That being said, it’s true that artiﬁcial intelligence cannot do it, and if it could do it we’d have another set of problems facing humanity’s future, much more pressing than those we are faced with today. In the world we know and can anticipate in the near and mid-term future, the reﬂection on the ﬁnality of social life and existence in general is still a product of human intelligence.
The Montreal Declaration rests entirely on this statement: it is up to human and collective intelligence to deﬁne the purposes of social life and, accordingly, the direction of artiﬁcial intelligence development so that it is socially responsible and acceptable, even desirable.
1. WHY HAVE A MONTREAL DECLARATION RESPONSIBLE AI?
The Montreal Declaration is a collective work that aims to put the development of artiﬁcial intelligence to work for the good of everyone, and orient social change by developing recommendations with a strong democratic legitimacy.
The selected method of citizen co-construction rests on a preliminary declaration of general ethical principles that state FUNDAMENTAL VALUES.
The initial work of identifying these values and principles allows us to launch a citizen involvement process that will deﬁne the ethical principles of responsible AI development and the recommendations to put into place to ensure that AI is promoting fundamental human interests.
1.1 THE INTELLECTUAL ORIGINS OF THIS PROJECT
The artiﬁcial intelligence (AI), and more speciﬁcally deep learning, revolution opens perspectives to unimagined technological developments that will help improve decision-making, reduce certain risks and offer assistance to those who are most vulnerable. This revolution is remarkable in many ways, although it also revives questions that were ﬁrst raised in the 18th century, at the time of the Industrial Revolution. It would be unwise to ignore the unique aspect of this revolution by hiding behind platitudes that aren’t preparing us to face current challenges. Of course, human beings are gifted beings with great technical abilities—human history is itself a history of technical transformations of nature, and artiﬁcial intelligence extends the trend to automation—but upon closer inspection nothing is similar to what’s in play with the arrival of artiﬁcial intelligence technologies. The cognitive skills we believed unique to humans can now be exercised by algorithms, machines that must be recognized as, in a certain sense, intelligent.
The social impacts of these new technologies, although very diverse, are still somewhat unknown. They could prove brutal if we don’t take the time now to have an ethical, political, judicial, sociological or psychological reﬂection on the type of society and human relationships we want to promote or protect while still using the advantages of the information technologies and algorithm calculations.
The use of algorithms to make technical or administrative decisions isn’t new. The rise of decision-making algorithms truly begins in the 1950s, especially in the healthcare ﬁeld: emergency room triage in hospitals, detection of sudden infant death syndrome risks, heart attack prediction (1).
All these algorithm techniques, “the procedures” already raise a certain number of ethical and social issues: those of social acceptability of an “automatic” decision, of the ﬁnal decision (is a human being at the end of the decision-making chain?), or of responsibility in case of a mistake.
And it is clear that these issues are being raised again with the latest algorithmic innovations.
What is different, then, about the latest technologies that full under the AI acronym? From an objective standpoint, what changes is the quantity of information that can be handled by computers (massive data) and the complexity of learning algorithms that, by feeding off of massive data, can accomplish perceptive and cognitive tasks allowing visual or audio recognition, and make decisions indeﬁned contexts. By combining different features (facial recognition, behaviour analysis, decision- making), AI raises extremely important ethical problems. From a subjective point of view, what’s new is the citizen wake-up call, as late as it was sudden, around the issues of algorithmic governance, the treatment of personal data and the social impact that some professional sectors are already experiencing.
If the progress of AI can surprise and fascinate, it also awakens the fear that using machines, namely robots, will considerably reduce the human relationship component when it comes to medical treatment, elderly care, legal representation, or even teaching. The reactions to the development of artiﬁcial intelligence can even prove to be hostile when AI is used for increased control of individuals and society, a loss of independence and a curtailing of civil liberties. This is why the hope of artiﬁcial intelligence being the bringer of social progress, always holds a dark shadow: placed into the wrong hands, AI could become a weapon of mass domination (control of private life, concentration of capital, new discrimination). Many people also question the intentions of the researchers, the developers, the entrepreneurs and the policymakers.
The development of AI and its applications therefore involves fundamental ethical values than can come into conﬂict and create serious moral dilemmas and deep social and political controversies: must we prefer public safety by increasing smart surveillance (facial recognition, anticipating violent behaviour) at the expense of individual freedoms? Objectively improving the well-being of individuals, namely by encouraging people to adopt behaviours normalized by smart devices (nutritional behaviour, work management, day planner) can it be done while still respecting people’s independence? Should the economic performance targets take priority over a concern for an equitable share of the beneﬁts of the AI market?
These dilemmas or tensions cannot be overcome simply by ranking fundamental values and interests. To put it another way, it’s not about classing the values in order of importance a priori, or building a simple and unequivocal scale of values, let alone favouring some while ignoring others (security at the expense of liberty, efﬁciency without social justice, well-being at the expense of independence). We also can’t hope to ﬁnd unique and permanent solutions.
It’s better to take the moral dilemmas caused by the development of AI seriously and collectively build an ethical, political and legal framework that will allow us to fact it while respecting the different fundamental values that we legitimately hold as members of a democratic society.
(1) Paul Meehl, Clinical versus Statistical Prediction, University of Minnesota, 1954.
1.2 FORUM ON THE SOCIALLY REPONSIBLE DEVELOPMENT OF ARTIFICIAL INTELLIGENCE
These reﬂections were the starting point for the initiative by the Fonds de recherche du Québec et de l’Université de Montréal to organize an international meeting to discuss the social impacts of AI and start the work around the Montreal Declaration for a Responsible Development of Artiﬁcial Intelligence (2).
On November 2 and 3, 2017, at the Palais des congrès de Montréal, a forum bringing together the greatest experts in the ﬁelds concerned by a reﬂection on AI, from pure science to social sciences and humanities. The Forum suggested setting the guidelines fora collective reﬂection on the ethical land socially responsible development of artiﬁcial intelligence, by pursuing the following three objectives:
offer a public reﬂection space around AI development issues and its social impacts
raise interest and notoriety among decision makers, industrial partners, politicians and the general community interested in AI, making them aware of the social questions raised by the sudden growth and numerous uses of AI;
privilege an interdisciplinary and intersectorial approach as a key factor to successful ethical and sustainable AI.
Thus were deﬁned the guidelines on an inclusive approach (interdisciplinary and intersectorial) which is at the heart of the elaboration of the Montreal Declaration for a Responsible Development of Artiﬁcial Intelligence that is not only responsible, but socially progressive, guaranteeing equality and justice. The preliminary version of the Montreal Declaration was presented at the end of the Forum. It was then a matter of launching the citizen co-construction process around AI ethics, a process we will expand upon in section 4.
(2) The Forum’s scientiﬁc committee was made up of Louise Béliveau (Université de Montréal, Vice-rectorat aux affaires étudiantes et aux études), Yoshua Bengio (Université de Montréal, Département d’informatique, MILA, IVADO), David Décary-Hétu (Université de Montréal, École de criminologie), Nathalie De Marcellis-Warin (École Polytechnique, Département de mathématiques et de génie industriel, CIRANO – Centre interuniversitaire de recherche en analyse des organisation), Marc-Antoine Dilhac (Université de Montréal, Département de philosophie, CRÉ Centre de recherche en éthique), Marie-Josée Hébert (Université de Montréal, Vice-rectorat à la recherche, à la découverte, à la création et à l’innovation), Gregor Murray (Université de Montréal, École de relations industrielles et CRIMT – Centre de recherche interuniversitaire sur la mondialisation et le travail), Doina Precup (Université McGill, School of Computer Science; MILA), Catherine Régis (Université de Montréal, Faculté de droit, CRDP – Centre de recherche en droit public), Christine Tappolet (Université de Montréal, Département de philosophie et CRÉ – Centre de recherche en éthique).
1.3 TOWARDS THE MONTREAL DECLARATION
The initial identiﬁcation of these values and corresponding principles was only designed to launch the citizen participation process that will help ﬁne-tune the ethical principles of responsible AI development, add to them and complete them.It should therefore come as no surprise that the Montreal Declaration is schematic and that the statement of principles is willingly very simple and consensual, leaving the interpretation and completion to the public deliberations (3).
If one of the goals of the co-construction process is to ﬁne-tune the ethical principles suggested in the preliminary version of the Montreal Declaration, another goal, just as important, was developing recommendations to oversee AI research and its industrial and technological development. However, it is too frequent to see analysis reports and recommendations forgotten as soon as they’re published: this is why it’s crucial to keep up the momentum built in the co-construction period.
Once the co-construction process is complete, it is necessary to open a public debate in the arenas where political, legal and policy decisions are made, in order to concretely implement the recommendations that came out of the citizen deliberation. These recommendations are not only legal in nature and, when they are, they don’t necessarily involve modifying a law. They could, however, request a modiﬁcation to the legal framework and, in certain ﬁelds, they have to. In other cases, the purpose of the recommendations is to nourish and guide the reﬂection of professional organizations so they modify their code of ethics or adopt a new ethical framework.
This step is therefore the ultimate goal of the co-construction process. We must, however, immediately inform you that when faced with a technology that has never ceased to evolve over the last 70 years and whose major innovations now come every 2 to 5 years, on average, it would be unreasonable to present the Montreal Declaration as deﬁnitive and complete. It is essential to think of co-construction as an open process, with successive and cyclical deliberation, production and recommendation production stages, and think of the Declaration itself as a guiding document that can be reviewed and adapted according to the evolution of artiﬁcial intelligence knowledge and techniques. This process of knowledge production, citizen deliberation and ethical framework and public policy recommendations, will have to be extended into a perennial institutional structure that allows it to remain reactive to AI evolution.
(3) The scientiﬁc committee in charge of writing this preliminary version was made up of Yoshua Bengio (Université de Montréal,Département d’informatique, MILA, IVADO), Guillaume Chicoisne (IVADO), Marc-Antoine Dilhac (Université de Montréal, Département de philosophie, CRÉ Centre de recherche en éthique), Vincent Gautrais (Université de Montréal, Faculté de droit, CRDP – Centre de recherche en droit public), Martin Gibert (CRÉ – Centre de recherche en éthique, IVADO), Pascale Lehoux (Université de Montréal, ESPUM – Ecole de santé publique), Joëlle Pineau (Université McGill, School of Computer Science; MILA), Peter Railton (Université du Michigan, , Académie américaine des arts et des sciences, philosophie), Christine Tappolet (Université de Montréal, Département de philosophie et CRÉ – Centre de recherche en éthique).
1.3 MONTREAL AND THE INTERNATIONAL CONTEXT
The Montreal Declaration initiative is part of a favourable scientiﬁc, social and industrial context. Montreal has become a major artiﬁcial intelligence research and development hub, with a community of researchers (Yoshua Bengio at Université deMontréal, a pioneer in the ﬁeld of deep learning, Joëlle Pineau at McGill, and so many others), world-renowned university labs (MILA, IVADO) andan incubator full of thriving start-ups and businesses
(Element AI, Imagia to name just a few). This scientiﬁc, technological and industrial development is at the heart of a revolution transforming social practices, business models and lifestyles, affecting all sectors of society. The City of Montreal is also this living lab of social and technological change. With fundamental scientiﬁc research come social and ethical responsibilities that the Montreal AI community fully accepts.
But outside of Montreal, it’s all of Quebec, and all of Canada, that offers a favourable social context to engage in a reﬂection on the social impacts of AI. Like MILA in Montreal, Vector in Toronto, AMII (Alberta Machine Intelligence Institute) in Edmonton, and the CRDM (Centre de recherche en données massives) in Québec make up hubs of excellence in fundamental research that have brought about extremely quick and robust industrial growth.
The Canadian Institute for Advanced Research (CIFAR, or ICRA) played a lead role in this Canadian development of AI by supporting fundamental research when AI was going through its “winter”.The Montreal Declaration initiative is supported by various players in Quebec and Canada outside of Montreal: the Fonds de recherche du Québec,
the CRDM de l’Université Laval à Québec, the Canadian Institute For Advanced Research.
Many international commentators have also shown their support for the Montreal Declaration, namely for its elaboration method. The Declaration team was able to establish a dialogue with institutions such as the Royal Society du Royaume-Uni (4) and the EGE (European Group on Ethics in Science and New Technologies (5)) that had their own study program and recommendations on AI. We ﬁrst note a convergence in the guidelines for ethical AI development as well as a shared intent to promote a democratic conception of AI use for the common good.
The Montreal Declaration initiative must also be viewed through the international context of an AI spring. It is preceded by many initiatives that must be recognized because they catalyzed the reﬂection around responsible AI. We must ﬁrst recall the creation, in 2014, of the Future of Life Institute that produced the Asilomar Declaration in 2017: after a 3-day conference, a declaration containing 23 fundamental principles surrounding AI research and its uses was signed by more than 1200 researchers. Professor Yoshua Bengio took part in the event at the time and brought attention to the risks of irresponsible and malicious AI use (6).
Since the Asilomar Conference, many reports on AI ethics have been published. The report from the Association internationale des ingénieurs électriciens et électroniciens (IEEE), Ethically aligned design. V2, was made public at the end of 2017 and gathered several hundred AI researchers and engineers. The AI Now Institute based in New York University has also produced several reports, the latest of which deals with evaluating the impacts of AI (7). Two ambitious strategic reports were published in March and April of 2018: the Mission Villani report in France and the one from the United Kingdom House of Lords “AI in the UK: ready, willing, and able?” Without claiming completeness, let us at least mention the participative approach of the CNIL (Commission nationale de l’informatique et des libertés) in France that led to the publication of a report with an evocative title: “Comment permettreà l’homme de garder la main? – Les enjeux éthiques des algorithmes et de l’intelligence artiﬁcielle”, in December 2017.
How does the Montreal Declaration position itself in this concert of independent initiatives? And what to think about the ethical inﬂation around AI? This last question is all the more important that we share the same warning as the EGE in its report Artiﬁcial Intelligence, Robotics and “Autonomous” Systems (March 2018) that in the absence of a coordinated reﬂection on the ethical and social issues of AI, there exists a risk of “ethics shopping” (8). The immediate consequence is a form of delocalization of ethical costs in areas of the world where ethical criteria are low priorities. Another risk is a form of trivialization of ethical discourse.
The speciﬁcity of the Montreal Declaration initiative is that it is essentially participative. From February to April 2018, the co-construction process brought together over 500 citizens, experts and stakeholders over ﬁfteen workshops and co-construction days. Although other participative initiatives have been led elsewhere, namely in France, the Montreal Declaration stands out by its size and its prospective methods.
The Montreal Declaration’s vocation is to open dialogue space in Quebec and Canada and offers a collective thinking platform that extends beyond the Canadian borders. The goal is to identify socially acceptable and innovative AI trends using informed citizen reﬂection in the different concerned democracies as a reference point. This dialogue space must also be accessible to citizens in non-democratic societies that wish to take part in a global debate on the future of human societies.
(4) We wish to thank UK Science and Innovation Network in Canada who facilitated the dialogue
(5) The European Group on Ethics in Science and New Technologies (EGE) is an independent advisory body of the President of the European Commissions.
(6) Yoshua Bengio interview during the Asilomar conference: futureoﬂife.org/2017/01/18/yoshua-bengio-interview/
(7) AI Now Institute, “Algorithmic Impact Assessments: A Practical Framework for Public Agency Accountability”, April 2018.
(8) EGE, Artiﬁcial Intelligence, Robotics and ‘Autonomous’ Systems (March 2018), p. 14
2. THE PRELIMINARY VERSION OF
THE MONTREAL DECLARATION FOR
A RESPONSIBLE DEVELOPMENT OF AI
To read the preliminary version, click here.
3. THE ETHICAL AND SOCIETAL ISSUES OF AI
The collective reﬂection process at the heart of the development of the Montreal Declaration rests on a preliminary version of the Ethical principles Declaration itself and informative exposes on AI and the ethics of AI.
3.1 WHAT IS AI?
The idea of AI is not a new one. You have to go back to at least the 17th century and the idea of a universal characteristic and combinatorial art from philosopher and mathematician Leibniz: reasoning comes down to calculating, and thought is conceived in algorithmic fashion (9). The notion of calculus ratiocination (logical calculation) predates the idea of an intelligent machine as it will be developed three centuries later, in the 1940s, by Alan Turing. In 1948, in a report entitled “Intelligent Machinery” and in 1950, in his famous article “Computing Machinery and Intelligence” (10), Alan Turing brings up a machine’s intelligence and develops the imitation game to deﬁne the conditions in which a machine can be said to think. The term artiﬁcial intelligence appears for the ﬁrst time in 1955 in the description of a workshop offered by John McCarthy (Dartmouth College), “2 months, (10) -man study of artiﬁcial intelligence”. But the uses and the development possibilities seemed very limited then, and so began the winter of AI, with minimal interest from the scientiﬁc community. Yet, if the discipline’s development paled in comparison to the philosophical and cultural fervour it inspired (one need only recall 2001: A Space Odyssey, Blade Runner or Terminator, to merely quote hit movies), research in the ﬁeld never ceased, and the dawn of the 21st century brought springtime for AI.
AI consists of a certain way of simulating human intelligence (11), taking inspiration from it and reproducing it. But foremost it is the brain, the human intelligence headquarters, which was designed as a machine capable of gathering, spotting and collecting data from its environment that it will then analyze, interpret and understand, using this experience to establish connections. The ﬁeld of AI research consists of producing mathematical tools to formalize the operations of the mind and thereby create machines that can accomplish more or less general cognitive tasks, associated with natural human intelligence. For example, discovering complex patterns among a large quantity of data, or reason in probabilistic fashion, in order to classify according to information categories, predict quantitative data or group data together. These cognitive skills are the basis for other skills such as deciding among many possible actions to achieve a goal, interpret an image or sound, predict behaviour, anticipate an event, diagnose a condition, etc.
But these cognitive skills are only possible if the machine is also capable of perceiving sensitive shapes such as images and sounds, which has been made possible by recent computer innovations. The notion of AI therefore also covers visual or sound recognition technologies that allow the machine to perceive its environment and elaborate a rendering of this environment.
These AI accomplishments rest on two elements:data and algorithms, meaning series of instructions that perform a complex action. Simply put, if you want to cook a new dish, you need to know the ingredients (the data) and follow a recipe that provides instructions how to use them correctly (the algorithm). Up until now, data processing capacities (quantity of data and processing algorithms) were too limited to consider a useful development for AI technologies. Things changed with the use of materials that made building very small and very fast calculators (computer chips) and storing massive amounts of data possible, as well as with the dawn of the information era thanks to the Internet.
What changed is the gigantic amount of data we can not only generate and transmit, but also process. If big data existed in the past, for example in the ﬁnancial industry, nowadays it’s a multitude of inanimate objects, spaces or receivers that are constantly producing unstructured data, meaning coming from disorganized information supports, which must be manipulated and transformed before it can be used. It can be millions of messages published on social media, all the words contained in a library full of thousands of books, or content from a huge number of images.
But what changed is also the type of algorithm developed by AI researchers. Determinist algorithms, which are a determined set of instructions like a cooking recipe, are being replaced by learning algorithms which rely on increasingly complex neural networks as the calculating power of machines increases. In computing, we talk about machine learning and the progress of this ﬁeld of study was reinforced by the development of deep learning.
At the heart of the notion of AI itself is the ability to adapt and learn. Indeed, for a machine to be considered intelligent, it must be able to learn by itself from the data it receives, as a human being does. And just like with humans, machine learning can be supervised, or not, by human beings that train machines on data.
It is these deep learning techniques that allowed machines to surpass human beings in complex games such as chess with AlphaZero, who also beats any other machine that doesn’t use deep learning, and the game of Go, which was reputed unmasterable for algorithms, but which saw AlphaGo triumph over the pest players in the world in 2015.
Although these examples are telling, AI can also serve other purposes such as automating tasks that required human intervention, especially perception and recognition duties. For example: processing speech; recognizing objects, words, shapes, and text; interpreting scenes, colours, similarities or differences in large sets, and by extension data analysis and decision-making- or help with decision-making. The possibilities are incredibly vast, and increase tenfold every time engineers and programmers combine them to create new uses.
(9) Leibniz (1666), De Arte combinatoria.
(10) A. M. Turing (1950), « Computing Machinery and Intelligence ». Mind 49, p. 433-460.
(11) Alan Turing begins his « Intelligent Machinery » (1948) report as follows: « I propose to investigate the question as to whether it is possible for machinery to show intelligent behaviour. »
3.2 AI IN EVERYDAY LIFE AND PHILOSOPHICAL QUESTIONING
AI engages us in an ethical reﬂection that, unlike one concerning nuclear or genomics, deals with everyday objects and technologies. AI is all around us and shapes our lives more than ever. We’re used to wearing small connected objects (phones, watches) and we’re preparing for the arrival of self-driving vehicles, cars and buses, but already we take trains and subways that operate independently, and planes, on autopilot, can take off, steer and land without human intervention. We use ranking algorithms for our Internet searches, autocorrect built into our messaging apps, curation apps for music or meetups, and we know that companies use sorting algorithms, banks use management and ﬁnancial investment algorithms, and that certain medical diagnoses can now be very exactly made by algorithms, etc.
These technologies are so seamlessly integrated into our everyday life that we no longer really think about them. When we talk about AI, most people still associate it with menacing, multifunctional machines that have some sort of consciousness, able to formulate a plan to destroy all humans (12). Yet the AI experience is a thoroughly banal one nowadays, with recommendation algorithms ﬂooding the Internet (Google, Amazon, Facebook). If you’re shopping online, there’s a good chance a pop-up window will open and that Inès will start up a conversation with:
“Hi, my name is Inès. How can I help you shop today?”
For a few moments, you get the impression that there’s someone, named Inès, is behind the screen talking to you; for a few moments, it’s okay to doubt. Inès asks you questions, answers yours, provides the important information you need to continue shopping. But after a little back-and-forth, you realize that although Inès provides relevant information, she replies in mechanical fashion, she doesn’t understand the way you write, doesn’t get jokes or open-ended questions, in other words, she doesn’t interact with you in a natural fashion. Inès is a conversational agent, a chatbot, AI. It’s become commonplace to chat online with chatbots to get more information about your health plan or new bank account, or even ask for fashion advice.
For now, chatbots can be spotted within a few minutes of conversation, usually much sooner. If a chatbot could go undetected by a human being for a reasonable amount of time, it could be considered that the machine successfully passed the Turingtest, and we would then be faced with, according to this test, a case of artiﬁcial intelligence, meaning a machine that thinks.
In his famous article, “Computing Machinery and Intelligence”, the father of modern computing, Alan Turing, proposes an answer to the question: “Can a machine think?” (13) And yet, in the introduction of his article, he changes the problem he feels he can provide an answer to: can a machine act in such a way that it is indistinguishable from a human being?He then offers the famous “imitation game” which consists of putting a human being asking questions (the interrogator) in contact with another human being and a machine answering his questions. If the machine can imitate a human being to the point that the interrogator can’t tell whether the human being or the machine replied, we can consider that the machine thinks. This is what is meant by
This imitation game caused a lot of controversy and saw philosophers ﬁercely clash with one another over whether a machine could be said to think. An experiment known as the “the Chinese chamber was made popular in the 1980s by philosopher John Searle (14). According to Searle, a machine that outwardly acts in the same fashion as a human being cannot be considered to have intelligence in the strong sense of the word. To illustrate this point, Searle asks us to imagine a room in which a person who, knowing nothing of Chinese, will try to pass fora Chinese speaker. It’s a variation of the imitation game: the person in the Chinese room, let’s call him John, receives messages written in Chinese that Chinese speakers outside the room hand him. John doesn’t understand a word of the messages he receives, but he possesses a very complex instruction manual which allows him to manipulate the Chinese characters and compose replies that are understood by Chinese speakers outside the room, so that they believe that the reply was written by someone who speaks Chinese. Searle deducts that in this case John simulated language skills but doesn’t possess them; he made people believe he understood Chinese, but he didn’t understand what he was writing. According to Searle, the same conclusion goes for AI: an intelligent machine manipulates characters, it follows an algorithm, meaning a series of instructions to accomplish a task (in this case, write), but doesn’t understand what it’s doing.
The debate is a fascinating one and is far from being settled, but we don’t really need to answer Turing’s question to wonder about the place AI holds in our lives and in our societies. For now well-trained chatbots can be as good as humans within a very limited framework of conversations, but leave no one guessing once that framework changes. And even if AI is ushering in an era where it is harder and harder to tell a naturally intelligent being from an artiﬁcially intelligent one, intelligent machines remain tools developed to accomplish well-deﬁned tasks. We
can therefore leave it up to cognitive philosophy metaphysics, psychology and neuroscience to debate the concept of artiﬁcial intelligence and discuss the possibility of robots developing emotions.
(12) Stanley Kubrick masterfully captured (and helped craft) this fantasy with the very human computer HAL 9000, in his ﬁlm 2001: A Space Odyssey (1968).
(13) A. M. Turing (1950).
(14) Searle (1980), ‘Minds, Brains and Programs’. Behavioral and Brain Sciences 3, p. 417–57.
3.3 THE ETHICAL ISSUES OF AI
Why introduce ethics when we can discuss the societal, social and economic impacts of AI?
Can we afford the luxury of an ethical reﬂection?
And isn’t it a bit naive to want to provide an ethical framework for AI development, which generates colossal proﬁts? These are questions ethicists hear on a regular basis among sceptical citizens, as well as decision makers who experience the limit of their ﬁeld of intervention. To answer it, we must ﬁrst very brieﬂy present the ﬁeld of ethics when discussing the societal issues of AI.
To keep it simple, ethics is a reﬂection on the values and principles that underlie our actions and decisions, when they affect the legitimate interests of other people. This supposes that all can agree on a person’s legitimate interests, and this is precisely what feeds the debate in ethics. The ﬁeld of ethics is therefore not concerned with what can be done, but generally what must, or should be done: we can kill a million people with a single nuclear bomb, but must it be done to impress an enemy country and demoralizing a population already suffering from
war? Take a less tragic example: you can lie to a friend about their new haircut, but is it moral in order to save them from deception? What must be done in that case? To answer that question, we must examine the available options: tell the truth, or not tell it, or tell only part of it, or tell it in a certain way. We must also examine the consequences of each option, question if they are important, and if so, why. We must also reﬂect on the objectives which are valorous (doing good unto others, respecting others). Finally, we must give ourselves a rule, a moral principle: for example, the categorical principle according to which it is always wrong to lie, regardless of the consequences;or the hypothetical principle according to which it is not morally right to lie unless…
The ﬁeld of ethics that applies to AI issues is public ethics. If we use the same type of reﬂection as public ethics, the object isn’t the same, nor is the reﬂection context. Public ethics is concerned with all the questions that involve difﬁcult collective choices on controversial institutional and social practices that affect all individuals as members of society, and not as members of a particular group:should a doctor tell his patient the truth about his health condition even if it will depress him and speed up the disease’s progress? This question doesn’t concern the doctor’s personal morality, but the types of behaviour we can rightfully expect from someone who holds the social role of doctor. This question is of a public nature and should be the subject of a public debate to deﬁne, using social values, best practices in terms of the patient-doctor relationship. By public debate, we mean all types of discussions which can take many various forms of consultations, deliberations or democratic participation, and that is open to a diversity of individual and institutional players such as professionals in the ﬁeld, association or union representatives, experts, policymakers and citizens. Public ethics call for a collective reﬂection to establish the principles of the best practices and demands that the players justify their suggestions on the basis of acceptable arguments in a pluralism context. In the case of the medical lie, you can appeal to shared values such as independence, respect of people, dignity, the patient’s health or well-being, etc. From these values, it is possible to establish principles that guide the practice of medicine and provide paths to regulation through the implementation of a code of ethics, by modifying a law or enacting a new law.
Public ethics is not besides or above the law, which has its own logic, but it helps clarify the issues of social life that various actors must keep in mind to respond to a citizen’s standard expectations and ensure equitable social cooperation. In this sense, public ethics shape public policies, and can lead to legislation, regulation, a code of ethics, an audit mechanism, etc.
In the ﬁeld of AI, it’s this type of ethical reﬂection that we implement. Let’s take the example of Melody, a medical conversation agent. Melody makes online diagnoses, accessible on your cellphone, according to the symptoms you describe. In a certain way, it acts as a doctor. This can be very practical in a society where the healthcare system is either inaccessible or underdeveloped. But the fact that it is practical is not sufﬁcient to authorize the public release of an app like Melody. Indeed, this app raises ethical questions that weren’t readily apparent with Inès, the shopping advisor chatbot. For example, we should question if Melody must give its user every possible prognosis, even if he is not equipped to understand the information. This problem is a simple transposition of a medical ethic questioning which has already received a normative response for which there is widespread consensus.
The notion of informed consent, of a patient’s free and enlightened decision helped clarify a doctor’s obligations. Does this solve the problem that Melody and its sister applications that often multiply unchecked? (15)
Generally speaking, probably, but speciﬁc attention paid to this technology reveals that it’s not that simple. The context does not allow Melody to ensure that the patient understands the diagnosis, or the urgency or not of treating the diagnosed condition. What rules must be invented to guarantee a patient’s autonomy and well-being? That is the issue of collective deliberation on AI’s ethical issues.
Other issues are speciﬁc to AI and have yet to ﬁnd ethical solutions. For example, if Melody makes a wrong diagnosis and the condition of the user who followed her advice goes seriously downhill, who is responsible? In the case of a medical consultation with a human doctor, it’s very easy to determine who is responsible for a medical error, but that’s not the case with decision-making algorithms. Do you hold the algorithm responsible? The developer, or rather the company that developed the algorithm and that makes money from its use? And if the product is certiﬁed, isn’t it rather the certifying body that should be blamed and legally sanctioned?
Public ethics questioning clearly introduces a reﬂection on the institutions that allow credible responses to be offered to a moral dilemma. It also deals with the type of society we want and the principles of its organization. By pursuing the reﬂection on medical chatbots, we cannot elude the question of the use of developing such intelligent machines, from a social and human standpoint. We must indeed question whether it is acceptable that smart apps replace medical doctors, even accepting the hypothesis that they can make a precise diagnosis, even more precise than a human. What does a patient-doctor relationship look like when the doctor is a chatbot? What essential elements are gained and which are lost? It is not a “utilitarian” type of question, but a question on the signiﬁcance of our social relationships, on the recognition of our vulnerability as patients, on human identity. Let’s go one step further: investing in the development of this type of AI rests on an eminently arguable social choice, which requires a collective discussion on the type of society we wish to build. We can indeed consider that we should improve access to an efﬁcient public healthcare system and therefore further invest in the training of doctors and an equitable health organization.
(15) The British public health service, the NHS (National Health Service) recently created a library of trustworthy apps (NHS Apps Library). Apps that do not offer sufﬁcient guarantees can be deleted from the library, which brings serious commercial repercussions for the company selling the app.
3.4 AI ETHICS AND THE MONTREAL DECLARATION
The development of AI and its uses therefore involves fundamental moral values that can come into conﬂict and provoke serious ethical, social and political controversies: should we develop apps like Melody to diagnose isolated people more quickly, or improve the healthcare system for all so everyone can see a doctor? There is no simple answer, but choices must be made.
The Montreal Declaration is currently in a preliminary version that serves as a starting point to the ethical reﬂection. The values presented in this version, although incomplete, supply a basic moral vocabulary to begin the ethical analysis of everyday situations and facilitate deliberations. The analysis of the Melody chatbot case illustrates this purpose of the Declaration. To understand the issue of enlightened patient understanding of a diagnosis, of attributing fault in the case of an erroneous diagnosis or of accessing health services, the Montreal Declaration offers a list of values you can immediately refer to:autonomy, responsibility, justice. It would be easy to demonstrate that the privacy value helps frame the problem of patient data conﬁdentiality.
The previous sections were presented in the various co-construction days and workshops (16).
They served as a starting point for the deliberations on prospective scenarios and the elaboration of ethical, political and legal solutions to AI societal issues. These informative presentations ended with the following general question:
How must we organize society to make an ethical use of intelligent machines, compatible with our fundamental moral and social interests? Which rules must be followed to make the best use of these machines while protecting our autonomy, ensuring social equality and equitable distribution of the fruits of the AI economy?
The co-construction process helped bring different credible responses to this question.
(16) Other analyses were presented, namely in the case of self-driving vehicles. We will expose them in more in-depth fashion in the ﬁnal version of this report.
4. THE CO-CONSTRUCTION APPROACH
4.1 THE CO-CONSTRUCTION APPROACH PRINCIPLES
To answer the many interrogations raised by the use of intelligent machines and ensure that AI develops “in good intelligence” with democracy, it is necessary to use an “excess” of democracy and involve the greatest number of citizens in the reﬂection process around the social issues of AI. The goal of the co-construction approach is to open a democratic discussion on the way society must be organized to responsibly use AI.
It’s not only a question of knowing what people think of a certain innovation and surveying their “intuitive” preferences; co-construction is not a public opinion survey around questions such as:
“Are you scared that AI will replace judges?”, “Would you prefer that a human or a robot operate on you?”
This type of question is not without interest, and the survey method provides important information do policymakers, as well as important working material for social sciences. However, although co-construction invites collective reﬂection around democratic issues, it also requires the development of documented, credible answers to pressing questions and the formulation of political and legal recommendations beneﬁtting from a strong democratic legitimacy.
This is the entire reasoning behind the approach initiated by the Montreal Declaration: giving back to democracy the ability to settle moral and political questions that concern society as a whole. The future of AI is not only written in algorithms, it resides foremost in collective human intelligence.
4.1.1 The principles of good citizen involvement
From the moment you involve the public in a consultation and participation process (co- construction) on controversial social questions, you must ensure that the process avoids the risks usuall associated with a democratic exercise. And yet, two objections are traditionally brought up to disqualify public involvement:
1. Ignorance: according to this objection, which is the most common, the public is ignorant and does not possess the ability to understand complex issues that require scientiﬁc knowledge, mastery of logical forms of argument and knowledge of political and legal processes.
2. Populism: according to this objection, which is a frightening one, the involvement of the unqualiﬁed public can be an opportunity for the demagogic manipulation that stokes popular stereotypes and can lead to the passing of unreasonable propositions, hostile to social progress, or even tyrannical towards minorities.
We do not share the belief that the public is so ignorant that they must not be consulted. We do not subscribe to the idea that non-expert members of our society have unsurmountable prejudices and their alleged irrationality leads them towards systematic errors. Ignorance is certainly an important problem, but we believe instead that they can shed light on neglected aspects of social controversies, because they are concerned by the issues discussed, and they can contribute to signiﬁcant solutions that experts haven’t thought of, or were unable to support publicly.
If, for certain individuals, prejudices and a tendency towards irrationality cannot be completely eliminated, it is possible to overcome these biases collectively.In favourable conditions, non-expert individuals can take part in complex debates surrounding social problems, such as those presented today by artiﬁcial intelligence research and its industrial applications. Experts in various matters relevant to citizen involvement on artiﬁcial intelligence can help implement these favourable conditions.
We have identiﬁed 4 conditions necessary for the co-construction process: epistemic diversity, access to relevant information, moderation and iteration.
We must ﬁrst ensure that the deliberating groups are as diverse as possible, in terms of social environment, sex, generation or ethnic origin. This diversity is not only required by the idea we have of an inclusive democracy, but is also necessary to increase the epistemic quality of the debates. This simply means that every person brings a different perspective to the subject being debated, and thus enriches the discussion.
Access to relevant information
But we know, however, that epistemic diversity is not enough and that if the participants have no skills or knowledge in the ﬁeld being discussed, they cannot produce new knowledge, or ﬁnd their way in the discussion. They are then collectively susceptible to amplify individual mistakes. We must therefore prepare the participants by providing relevant, quality information, both accessible and reliable.The deliberations must therefore be preceded by an information phase.
Other than having quality information, it is necessary that the participants reason freely, which is to say without being impeded by cognitive biases. We deﬁne cognitive bias as a distortion of rational thought by intuitive mechanisms. One of the most common, and most problematic in a deliberation is the conﬁrmation bias: we have a tendance to only accept opinions that conﬁrm our own beliefs, and to reject those that go against what we already believe. There are dozens of cognitive biases that can deform the logic course of our reﬂections.
But there are also biases that apply to the deliberation itself, such as the tendency to adopt more and more radical positions: if the group that is deliberating is initially distrustful of artiﬁcial intelligence innovations, it is quite probable that they will be entirely hostile towards them at the end of the deliberation process. To avoid this type of knee-jerk reaction, we believe it’s important to ensure epistemic diversity in the deliberating group and to put a moderation body in place.
This does not necessarily have to take the shape of a personal intervention from a moderator. Although we don’t reject personal moderation, we believe we can overcome deliberation biases through other means, such as introducing unexpected events in the scenarios that sparked the discussions.
Ideally, we should be able to bring together the population as a whole to take part in a reﬂection on the responsible development of artiﬁcial intelligence. But the conditions we just described cannot be applied to very large groups, let alone a society of many millions of people. It is therefore important to conduct citizen involvement in smaller groups and increase the number of meetings. This is the iteration phase of co-construction.
The reasons to proceed this way are technical, but easily understood. The mathematician and player in the French Revolution, the marquis de Condorcet, had demonstrated that the judgment of groups is always right more often than each person individually, and that this increases as the group grows larger. For this to be the case, however, two conditions must be met: the individuals in the group must have more than a ﬁfty/ﬁfty (50/50) chance of being right, and they must not communicate with one another (Condorcet rightly feared the risks of manipulation).
Yet we cannot ensure that for very large groups that the individuals have the required skills and that each individualhas more than a ﬁfty-ﬁfty chance of having an appropriate opinion. Allowing deliberation (communication between one another) is one way to increase the skill of the participants, as long as it is in the framework we are suggesting. Of course, that does not satisfy Condorcet’s second condition, but it does guarantee the ﬁrst. And to increase the quality of opinions, it is necessary to multiply the deliberating groups: since we cannot increase the size of the group, we must increase the number of participants by proceeding with an iteration of participation sessions.
For all of these reasons, we have selected the structure of a co-construction workshop that brings together non-expert citizens, experts, stakeholders (associations, unions, professional representatives, businesses), as well as political players. These workshops are organized in different formats adapted to the deliberation spaces and satisfy the conditions for fruitful, solid citizen involvement.
4.1.2 Experts and citizens
Why allow citizens to be heard on complex ethical and political questions that require a good knowledge of the technologies being discussed? Why not only consult the experts? There are many reasons, but the easiest is that AI affects everyone’s lives, therefore it concerns everyone and everyone must have a say in the socially desirable orientations of its development.
Even when we are not in the presence of a dilemma, strictly speaking, public ethics questions cannot be solved without making choices that favour certain moral interests over other moral interests, without neglecting them. This is the result of value pluralism which deﬁnes the moral and political context of modern democratic societies. It is therefore possible to favour well-being by challenging the priority of consent: think of a medical app that could access personal data without our consent, but that would help better heal serious diseases thanks to the data.
This type of ethical and social choice should be in the hands of all members of our democratic society, and not just a part, a minority, even if they are experts.
The experts’ role is not to solve the ethical dilemmas brought on by artiﬁcial intelligence themselves, nor become legislators. What are the experts doing then?The experts involved in the Montreal Declaration co-construction process don’t intend to think for the citizens and suggest a legal and ethical framework that the citizens would merely rubberstamp. Expertise must be at the service of citizen reﬂection when considering complex social and ethical AI issues.
Sometimes ethicists can give off the impression of looking to preach, of knowing the answers to the touchy questions that the public is asking themselves, and even of being able to preemptively solve tomorrow’s problems. It’s important to specify their role. Ethicists play three modest but crucial roles:
They must ensure that favourable conditions are in place for citizen involvement;
They must clarify the ethical issues that underlie the controversies surrounding artiﬁcial intelligence;
They must rationalize the arguments being defended by the participants by showing them the arguments we know to be wrong or biased and explaining the reasons why they are wrong.
The role of ethicists is therefore that of informed guidance. Experts in other research ﬁelds (computer sciences, health, safety, etc.) also play a guidance role by providing participants with the most useful and reliable information regarding the object of controversy (How does an algorithm that learns to make a diagnosis work? Can a doctor be replaced by a robot programmed for the diagnosis? What are the protections we can put into place to defend against attempts to hack our medical data? etc.)
And yet, it must be recognized that the experts themselves often show important cognitive biases. They can be too optimistic or pessimistic towards new technologies they know well; they also tend to put too much weight into their own opinion, especially when they believe they can predict the evolution of their ﬁeld of society, of social trends,etc. It’s by involving them as citizens in the co- construction workshops that we reduce the biases linked to expertize, as well as the authority effect caused by the knowledge imbalance with the other participants.
The co-construction workshops are participation spaces that help give direction to the socially desirable development of AI and innovate through proposals that shake up the recognized analysis framework. This essential contribution from citizen deliberations is then analyzed and expanded upon by work committees made up of experts from different ﬁelds (researchers, professionals). This work of expanding and drafting recommendations follows the direction deﬁned by the deliberation and remains faithful to the proposals issued at the co-construction workshops.
4.2 THE CO-CONSTRUCTION WORKSHOP METHODOLOGY
The ﬁrst version of the Montreal declaration on Responsible AI, presented November 3, 2017, during the Responsable AI Forum, is the foundation of the co-construction process. Schematically, after having deﬁned the “what”? (“which desirable ethical principles should be gathered in a declaration on the ethics of artiﬁcial intelligence”), it’s a matter in this new phase of anticipating with citizens and stakeholders how ethical controversies surrounding AI could arise in the next few years (in the ﬁelds of health, law, smart cities, education and culture, the workplace, public services) to then imagine how they could be solved (for example, with a device such as sectorial certiﬁcation, a new actor mediator, a form or a standard, a public policy or a research program).
The goal of the co-construction approach and its workshops is namely to exemplify and test the principles of the Montreal Declaration for Responsible AI thanks to potential scenarios. Ultimately, the process will help specify sectorial ethical issues, and then formulate priority recommendations to the AI community.
More than ten co-construction workshops took place between February and May: 3-hour citizen cafés in public libraries, and two big co-construction days with various citizens, experts and stakeholders
(at the SAT in Montreal, then at the Musée de la civilisation in Quebec City).
The choice of organizing citizen cafés in public libraries is directly tied to the current reinvention dynamic of these public services in Quebec and Canada. By going from a lending space to that of an inclusive “third space” library that seeks to strengthen the capacities of all its citizens (ex. with digital literacy services, citizen support, cultural mediation and discussion areas, the lending of tools and the creation of Fab Labs), public libraries will most certainly have a key role in the responsible deployment of AI in Quebec and Canada (17).
The co-construction days were held in symbolic spaces (Société des arts technologiques in Montréal, Musée de la civilisation in Québec) and namely focused on the meeting between stakeholders and the very diverse disciplines that must work together to imagine a responsible deployment of AI in society.
(17) Christophe Abrassart, Philippe Gauthier, Sébastien Proulx and Marie D. Martel, « Le design social : une sociologie des associations par le design?
Le cas de deux démarches de codesign dans des projets de rénovation des bibliothèques de la Ville de Montréal »,
4.3 ORIGINALITY OF THE CO-CONSTRUCTION APPROACH
When compared with other AI ethics initiatives currently underway in the world, this co-construction approach will present four particularly original and innovative dimensions :
The use of foresight methods, with sectoral scenarios set in 2025 exemplifying through short tales how ethical controversies about AI could appear or increase in the next few years (in the ﬁelds of health, law, smart cities, education and culture, the workplace). These 2025 scenarios, which present a variety of possible situations in the face of a wide-open future, will be used to spark the debate, to identify, specify or anticipate sectorial ethical issues on the deployment of AI in the coming years. These discussions with a 2025 horizon can then help retroactively formulate concrete recommendations for 2018–2020, to guide us towards collectively desirable situations.
The use of participative design facilitation methods in multidisciplinary “hybrid forums” (18), including citizens and stakeholders, in a context of shared uncertainty in the face of possible futures (to ﬂesh out a scenario, design ways to respond to an ethical risk, suggest an addition to the Declaration in the case of an orphan issue, i.e. without a corresponding principle).
Lastly, paying attention to the “paradigm biases” that have very powerful reframing effects in the way they position problems (ex. tackling the ethical issues of self-driving cars strictly from the tramway dilemma angle [ex. MIT’s Moral Machine site] and in the context of the “speed-distance” paradigm in transport design), in order to ensure a plurality of issues and draw attention to still unknown or very emerging situations in a rapid change context.
This co-construction approach aims to create a learning trajectory to develop, throughout the events, a facilitating kit that’s reproducible and user-friendly adaptable and open, that could be published in “open source” at the end of the co-construction approach. The detail of the world cafés and co-construction days can be found appended to the report.
4.4 WORLD CAFÉS OUTSIDE LIBRARIES
We must also mention the involvement of two philosophy students at Université de Montréal, Pauline Noiseau and Xavier Boileau, who organized many world cafés in spaces other than libraries, and whose formula was more focused on organic discussions about an AI issue. Moderators used very short scenarios, and hosted 2-hour sessions. These sessions were strong deliberation moments with citizens that wanted nothing more than to
be involved in public debates, but that were rarely called upon. That’s how a world café at the Maison d’Haïti, on April 25, 2018, allowed high school youth and retirees from the Saint-Michel neighbourhood in Montréal-Nord to trade opinions around AI issues. From an AI scenario on household connected objects (a smart refrigerator), this session namely sparked original reﬂections on cooking as a relational human activity, raising issues of authenticity, of affection (“a touch of love”), and of social ability, issues that hadn’t come up in other consultations based on the same scenario.
(18) Callon, Lacoumes, Barthe, Agir dans un monde incertain. Essai sur la démocratie technique, Paris, Le Seuil, 2001
4.5 PORTRAITS OF THE PARTICIPANTS
Recruiting citizens, experts and professionals from different ﬁelds of work helped reach a diversity of participants for the co-construction. University faculties, as well as inter-university research centres and their networks helped reach an important number of players involved in the development and use of AI.
To reach the general public, the websites and social media of our different partners played an important role, although the local recruitment efforts from each library involved in the project proved to be the most efﬁcient.
Notable fact, there was practically the same number of men and women in all workshops. A strong majority of participants had post-secondary education and were in the 19-34 age group.
5. DELIBERATION PATHS IN THE WORKSHOPS
EXAMPLES FROM TWO ELEMENTS : SMART CITIES AND THE WORKPLACE
5. DELIBERATION PATHS IN THE WORKSHOPS
EXAMPLES FROM TWO ELEMENTS : SMART CITIES AND THE WORKPLACE
5. DELIBERATION PATHS IN THE WORKSHOPS
EXAMPLES FROM TWO ELEMENTS : SMART CITIES AND THE WORKPLACE
5.1 THE DELIBERATION PATHS
How did the discussions and deliberation in the co-construction workshops unfold? What kinds of reactions did they provoke? What were the main points of discussion that led to recommendations for an AI framework? This section of the document details certain highlights from the deliberations between participants, where each person took care to specify the reasons, principles and values justifying their position on the prospective scenario suggested as a starting point, whether it was to agree, disagree, nuance or question something.
In a word, to do what pragmatic sociology has deﬁned as justiﬁcation.
To illustrate this work, the paths of two teams representing two sectors among the ﬁve discussed in the co-construction were selected:
a table of citizens that discussed the self-driving car (smart city sector) and a table of researchers and experts dealing with the impact of AI on jobs in businesses (workplace sector).
To formulate these recommendations, each team underwent three steps where ideas were generated, then deliberated:
First step: formulating sectorial ethical and social issues in 2025 (by cross-referencing the general principles of the Montreal Declaration with the 2025 user situations described in the debate-provoking scenarios): the formulating of individual issues (on Post-its) was then expanded upon in a collective discussion from which came a selection of three priorities.
Second step: the formulating of recommendations to be implemented in 2018-2020 to prepare for a responsible rollout of AI in Quebec: from the formulating of recommendations to the choice of a few newspaper headlines.
Third step: he storytelling of the launch of a ﬁrst recommendation in 2020 (the newspaper headline)to take stock of the “time for collective action” with its organizational constraints: from formulating ideas to synthesizing them in orderly fashion within a narrative.
We must mention that between these steps and micro-steps of the deliberative path, the “nature”
of the ideas generated varies: some are individual intuitions (when, at the start of the exercise, participants write down many sectorial issues on Post-its), others stem from a collective discussion (where each person justiﬁes their point of view), and others yet are the result of a hierarchy determined by the group (when selecting three key issues to write on the summary poster).
We therefore ﬁnd in these prospective workshops three properties of the deliberative devices highlighted by Blondiaux and Sintomer in their article L’impératif délibératif (19) (Politix, 2002, pp. 25-26): allow the imagination of new solutions in an uncertain world; allow a rise in generality and aim for consensus or “deliberative disagreements” in a society marked by the pluralism of values;and ﬁnally, provide a factual and normative source
of legitimacy through the inclusion of everyone in these deliberations.
(19) Blondiaux L. et Sintomer Y., « L’impératif délibératif », Politix, 2002, p. 25-26.
5.1.1 Smart city sector : Self-drinving car (SDC) and sharing the road equitably
Summary of the initial 2025 scenario.
In 2025, the ﬁrst SDC are circulating in Montreal and controversy arises around sharing the road and public spaces. Some lanes are now reserved for SDC and are protected by barriers, so that they can driveat a moderate, but ﬂuid speed (50 km/h)without risking accidents. SDC can also drive elsewhere, but at very slow speeds (25 km/h). Protesters for active mobility (walking, biking)disturb the operation of these protected lanes, knowing that the algorithms of the SDC areset to “altruistic” mode to protect outside people.
The goal of this scenario was to open a discussion on the ethical issues of SDC through a situation recreating the density and complexity of a city: low and different speeds, ﬂuidity as a priority criteria for speed, protection barriers for safety, the road as a shared space for competing uses.
The deliberative path presented is the result of a 3h table in a Montreal public library, with eight citizens interested in new technologies and other wisemaking active family mobility habits (walking, biking). From this scenario set in 2025, the discussion led to formulating an initiative presented as a headline in the March 13, 2020, edition of the Responsible AI Gazette: “First autonomous mobility literacy workshop held.” What was the deliberative path that led to this original proposition? What were the deﬁning moments? How did the ideas grow at every step? We present and comment certain signiﬁcant moments of the path taken by this team.
First deliberative moment : Formulating ethical issues in 2025
Many interrogations drawn up on Post-its were submitted by the participants in relation to different principles of the Montreal Declaration:
THE AUTONOMY PRINCIPLE
“Will humans become too dependent when it comes to moving?”, “Will the freedom of movement be impeded by AI?”, “We’re giving up a lot of
micro-decisions to AI and interconnected systems, at the expense of humans.”
THE WELL-BEING PRINCIPLE
“A lot less room for spontaneity with SDC”, “What will the neighbourhood development look like in regards to the road axis of SDC?”, “Will transportation datainﬂuence city urbanization?”
THE DEMOCRACY ANDJUSTICE PRINCIPLE
“What is the difference in installing transportation axis in working-class neighbourhoods as opposed to afﬂuent neighbourhoods?”, “Will only those who are well-located get to enjoy the ﬂuidity of trafﬁc?”
THE PRINCIPLE OF PRIVACY
“Will we be able to replace anyone’s movements?” of responsibility: “Who will be responsible for an accident?” or of security: “Possibility of hacking ﬂeets of vehicles?” this last principle coming from the participants, in addition to those found in the declaration.
Many in-depth discussions then occurred, participants bouncing off of the ﬁrst ideas to generate new ones on spontaneity and freedom to travel, on the safety of personal data and its management by a central organization, on the question of algorithm settings and the possibility of manipulating them.
Then, after a nearly 45 minute-long discussion, the participants used coloured stickers to select 2025 ethical issue groupings that seemed a priority to them. The participants voting with coloured stickers on the wall with Post-its and discussed ideas associated with four principles of the Montreal Declaration, two of which were regrouped : safety, justice, and well-being and autonomy.
This selection of priority issues by the team is an original one: although the issues of security, responsibility and privacy are often raised in studies and debates on SDC, those of justice, well-being and autonomy are much less discussed.
Second deliberative moment : Recommendations for an AI framework in 2018-2020
To respond to these issues, the team chose to continue its discussions by trying to think about the four associated principles. Many AI framework recommendations were formulated by the participants. We present three (out of six) here, which allow you to follow the path of an idea all the way to the headline of a newspaper.
These recommendations, which show true institutional creativity (beyond the very broad examples of tools provided in the participant booklet), are in line with the issues identiﬁed at the previous step, but also present an enrichment of ideas (they are not simple deductions from tools adapted from an identiﬁed ethical case).
The idea of training for vigilance and of participating in a collective decision (through an all-party committee and open planning) do indeed lead to recommendations for capacity building and local forms of democracy.
Third deliberative moment : Writing a headline for a 2020 newspaper
These measures were then storyboarded in the following way on the poster. The headline of the March 13, 2020, edition of the Responsible AI Gazette designed by the team read as follows:
“FIRST AUTONOMOUS MOBILITY LITERACY WORKSHOP HELD”
“The Quebec public library network has established a training program on the use of self-driving cars. On the curriculum:collective vigilance; the code of ethics; how to get involved in the city’s decision-making process; sharing the rod between pedestrians, bicycles, SDCs, trucks; explaining the rules; trial sessions; the question of algorithm settings.”
This newspaper headline, which was formulated after a discussion among the participants, contributes again to the progression of ideas. Indeed, the principle of a workshop on “autonomous mobility literacy” allows the creation of new meaning by integrating the various recommendations formulated in the previous step widening the scope to take about autonomous mobility and not simply SDCs (thus allowing for the possibility of autonomous multimodal transportation). This headline also presents a collective action device with a progress target (the training and abilities of citizens, the possibility to participate in the city’s decision- making committees regarding SDC rollout) and an organization (a rollout in public libraries across Quebec, which are currently transforming into cultural services third parties for all citizens.
The result of this table is particularly interesting because it helps consider the ethical question of self-driving vehicles from the perspective of autonomy and social justice in the city, and not strictly from a responsibility in case of accident scenario, as MIT’s Moral Machine initiative does, for example, from the moral dilemma of the tramway (20).
(20) MIT site: moralmachine.mit.edu
5.1.2 Workplace sector : Socially responsible restructuring?
Summary of the initial 2025 scenario.
In 2025, many businesses use AI in their management tools. Such is the case for an eco-friendly logistics company that must make a massive investment in AI and robotics in order to remain competitive. Parcel sorting, routing, administrative follow-up, calculating the carbon footprint of the trips, self-driving electric trucks: in total, up to one third of the company’s positions could be cut. The company, which is very socially involved, wants to proceed with this restructuring in socially responsible fashion, for instance by creating a data processing coop to rehire as many salaried employees as possible, independently from the big corporations. Will it be able to do so in time?
The goal of this scenario was to spark a discussion on the ethical and social issues concerning the change in the process caused by AI that thousands of SMEs and big businesses in Quebec will be faced with between 2020-2030.
The deliberation path presented in this section comes from a table eld over an entire day in Montreal bringing together nearly ten researchers and experts working on workplace mutations, social and the social responsibility of businesses and unions.
A citizen that had previously attended a workshop in a public library was also at this table.
Starting from the 2025 scenario, this team’s work led to the formulating of an initiative that made the headline of the February 18, 2020, Responsible AI Gazette: “First measures of the mixed interdepartmental committee on responsible digital transition.” Like in the previous case about self-driving cars, what was the deliberative path
that led to this original proposition? What were the deﬁning moments? How did the ideas grow at every step?
We present and comment certain signiﬁcant moments of the path taken by this team.
First deliberative moment : Formulating ethical issues in 2025
Many Post-its were drafted by the participants in the ﬁrst part of the morning workshop. Here are a few of them and an overview through a few formulas taken from Post-its and the table of grouping by Montreal Declaration principles.
Certain formulated issues were associated with different Montreal Declaration principles:
THE WELL-BEING PRINCIPLE
“What do we favour?
The company or society?”,
“Adopting different perspectives on well-being:individual (employee), social and collective development, economic development (SME)”,
“What do performance ideals look like when robots or co-bots never get tired, unlike humans?”, “What are the possible positive aspects: professional reinforcement, for ex. in medicine, less drudgery for certain positions”, “What are the new forms of work and protection with work/leisure?”
“What professional and life paths? Can you choose not to change careers because of AI? What are the consequences?”, “collective autonomy: for the collective and critical anticipation of discussion
on the urgency of adaptation”
THE RESPONSIBILITY PRINCIPLE
“Who is held responsible for these changes?”,
“Is the social and ethical responsibility of the transition individual — each company – or collective —society, the government?”, “What funding for the transition?”; “How to align the cost-effective directive and the responsibility in an emergency context?”
THE KNOWLEDGE PRINCIPLE
“What collaboration between humans and robots? Workload, health and safety, training, acceptability, cybersecurity,” “How is data collected in a context where this type of work is mainly carried out by private corporations (GAFAM)?”, “How to prevent people getting stuck in classes?”, “What are the possibilities of data being shared?”, “What is the impact on the educational system?”
THE JUSTICE PRINCIPLE
“What independence in the face of power being concentrated among GAFAM?”, “What social redistribution of the social beneﬁts of AI?”, “Will the productivity gains created by AI and industry 4.0 be sufﬁcient to fund the social transition if companies practise tax evasion?”, “What equity in case of sharing and coding an employee’s implicit knowledge to transform it into data or feed the automation?”, “Do we have a choice, as employees, not to reveal this information?”, “On what criteria will we choose those who are replaced and those who are trained?”, “What access to the social protection of tomorrow?”, “What access to rights, such as the right of association, in this new workplace reorganization?”
THE DEMOCRACY PRINCIPLE
“Is precariousness a fatality when the transition can be anticipated?”, “the politicized short-term vision rather than a long-term vision”,
“the obscuring of decision-making processes”, “risks of biases in the algorithm training sets”, “the need for a democratic debate”.
We can mention here that the typology of the Montreal Declaration on responsible AI principles worked well to provide benchmarks for the discussion, and that the participants even suggested original problems concerning certain principles: the necessity of addressing well-being and responsibility for the transition from different points of view (individual and collective); the relationship with social time, with an opposition between the collective anticipation and the opaque language of urgency, as a condition of our collective autonomy and exercising our democracy (the lack of time preventing well-informed democratic work); a strong requirement for justice in the social redistribution of AI beneﬁts, namely in terms of equity accompanying the codiﬁcation, and therefore possible automation, of employee skill sets.
After a good hour of discussion, the participants used coloured stickers to select groupings of 2025 ethical issues they deemed priorities. The votes being spread out pretty evenly over the various issues, all deemed equally important by the group, the formulation of three priorities for the poster was also a synthesis exercise of the ideas discussed in the ﬁrst part of the workshop (see table below).
Second deliberative moment :Recommandations of an AI framework in 2018-2020
To respond to these issues, the team continued talks in the afternoon by going around the table once more leading to the drafting of AI framework recommendations by the participants, which then led to numerous recommendations that were collectively discussed one by one. The table below presents an excerpt (six propositions out of the more than 10 that were formulated by the group), in order to follow the path of an idea up to the formulation of a newspaper headline.
As in the previous case about self-driving cars, these recommendations, which show true institutional creativity (beyond the very broad examples of tools provided in the participant booklet), are inline with the issues identiﬁed at the previous step, but also present an enrichment of ideas. If digital literacy is indeed a goal in the policy’s agenda (ex. Stratégie numérique du Québec), it’s the necessity that it expands that was highlighted. The other recommended measures are innovative and invite the creation of a new public, all-party or collective devices to ensure Quebec society’s true autonomy when faced with AI issues in the workplace. In that sense, the group has chosen collective responsibility towards AI in its transition into society.
Third deliberative moment :Writing a headline for a 2020 newspaper
These measures were then storiﬁed for the poster. The headline formulated by the team for the February 18, 2020, Responsible AI Gazette reads as follows:
First measures of the mixed interdepartmental committee on responsible digital transition
The new committee, created on March 14, 2018, after the co-construction day for the Montreal Declaration Responsible AI, quickly got to work and developed a coherent strategy integrated with all stakeholders. In early 2020, the committee was proud to announce the launch of 4 programs:
1. A new digital insurance fund worth 2 billion dollars (funded by productivity gains attributed to AI).
2. An agreement with all cégeps and universities to accelerate the renewal of training programs.
3. A support program to create self-employed worker cooperatives (against precariousness).
4. A literacy fund worth 10 billion dollars over 5 years, on the basis of a new skill set inventory.
This newspaper headline, which was formulated after a discussion among the participants, contributes again to the progression of ideas. Indeed, the mixed interdepartmental committee on responsible digital transition would be a creation. This new institutional player, born from a reﬂection on a 2025 scenario concerning the impact of AI on the Quebec workplace, could represent a new common step for many public policies that successfully address the digital transition and the issue of digital literacy, but don’t raise the question of AI’s social impact: the Stratégie numérique du Québec du ministère de l’Économie, de la Science et de l’Innovation (MESI), the Stratégie nationale sur la main-d’œuvre 2018- 2023 du ministère du Travail, de l’Emploi et de la Solidarité sociale (MTESS), the Plan stratégique2017-2022 du ministère de l’Éducation et de l’Enseignement supérieur (MEES). This new player, which could be the result of a cross collaboration between the Commission des partenaires du marché du travail (CPMT), the Comité consultatif sur le numérique and the Commission mixte de l’enseignement supérieur, would speciﬁcally anticipate workplace transformations and new training and adaptation issues created by the rollout of AI in Quebec’s public and private organizations.
6. THE FIRST CO-CONSTRUCTION RESULTS
Citizens gathered around 45 tables to discuss their perception of issues tied to applying the Declaration’s principles. (Table 5)
They identiﬁed different potential solution categories to respond to these issues. (Table 6)
6.2 CO-CONSTRUCTION DATA : EXPLANATORY NOTES
The current section relates the results collected during the co-construction tables held in winter 2018 for the Montreal Declaration, 45 tables in all that brought together hundreds of citizens. Discussions were held around 5 major sectors of AI development: the education sector (9 tables); the justice and predictive police sector (8 tables); the healthcare sector (12 tables), the workplace sector (5 tables), and the smart city and connected object sector (11 tables).
These results stem from a preliminary and non- exhaustive analysis of the two main axis discussed at each table: the various issues raised by AI development, as well as the potential solutions identiﬁed that came out of citizen discussions based on provoking scenarios. At this level, the analysis remains descriptive and as close as possible to
the citizen’s words. For the purposes of this report, emphasis was placed on:
1) the great directions expected in terms of responsible AI development;
2) the presentation of issues citizens determined to be priorities;
3) the issues that could lead to the creation of new principles in the Declaration;
4) the potential solutions identiﬁed by citizens to respond to these issues.
The great directions expected in terms of responsible AI development refers to citizen recommendations that are not speciﬁed in concrete potential solutions. They nonetheless allow the main positions and standard expectations citizens have towards AI development.
Each co-construction table was invited to choose 2 or 3 issues to be treated as priorities before 2025. Only issues that were considered priorities by citizens were analyzed for the purposes of this report. These priority issues were therefore described on the basis of citizen formulations and classiﬁed, for each sector, according to the Declaration principles they are linked to. However, it’s worth noting that just because certain issues weren’t considered priorities that they weren’t discussed, that they’re less important, or that the principles weren’t discussed for each sector. One single principle for each sector is detailed in this progress report.
Different issues that could lead to the creation of new principles in the Declaration were identiﬁed on the general basis of the discussions that took place. In this report we present, in non-exhaustive fashion, those that proved to be particularly relevant.
Finally, the potential solutions identiﬁed by citizens to respond to these issues have been classed in 11 main categories. These categories will be speciﬁed in subsequent steps of the analysis. The category that seems most relevant to bring up for each sector is presented in greater detail.
Regarding the quantitative data in this report, the number of occurrences corresponds with the number of tables where each issue/potential solution was formulated in consensual fashion, in conformity with the co-construction process.
The total number of potential solutions (n=190) corresponds to those identiﬁed as priorities by citizens (since they were invited to clearly formulate them on posters). However, potential solutions mentioned during the discussions but not explicitly appearing on the posters are also taken into consideration.
Quotes from the report are presented in such a way as to reference the co-construction table when they come from a group formulation (consensus). Other quotes correspond to individual formulations (written on Post-its by participants or copied verbatim by members of the group).
6.3 RESPONSIBLE DEVELOPMENT OF AI: THE GREAT DIRECTIONS EXPECTED BY CITIZENS
Generally speaking, the participants recognized that the arrival of AI came with important potential beneﬁts. Namely, when it came to work and legal matters, participants recognized the time savings that AI devices could bring:
“It would help reduce wait times to treat cases.” - A participant
However, it was also mentioned that AI development had to be done with caution and right now to prevent abuse, although some consider the possibilities brought on by AI to still be limited.
The implementation of a framework is therefore recognized as necessary to prevent risks rather than determining who is to blame when they occur:
“You don’t care so much about knowing who to sue when things go wrong, you want to ﬁnd ways to make sure things don’t go wrong in the ﬁrst place.”
- A participant
The citizens highlighted the need to implement different mechanisms to ensure the quality, intelligibility, transparency and relevance of the information being communicated. They also discussed the difﬁculty of guaranteeing truly enlightened consent.
The great majority of the participants recognized the necessity to align public interests with private ones and prevent the apparition of monopolies, or limit the inﬂuence of corporations (which are seen as ungovernable) through more cohesive and legal measures. These mechanisms should be, as much as possible, simple and changing so they can adapt to the rhythm of AI development and allow its steady control. In the legal sector, certain participants mentioned a “gap” separating technology (deﬁned as quick, innovative, even abstract) and our institutions (often too stiff in their integration of technology) that aren’t able to deal with these changes in society. Some tables went as far as suggesting “nationalizing AI”, which would then “become a public service, and programmers would be public servants”. (Smart City and connected objects table, INM, Montreal, February 18, 2018, Connected refrigerator scenario).
The participants also recommended guaranteeing a contextual approach to AI, which must take different parameters into account (ex. mandatory or optional collection of data the algorithm learns from). These mechanisms should come from and involve independent, trained people to favour the diversity and integration of those who are most vulnerable, and protect the mixed aspect of the lifestyles.
Whatever the use, the majority of the participants insisted on the fact that AI must remain a tool, and that the ﬁnal decision must come from a human being (whether it’s a legal ruling, a hiring decision or a health diagnosis), which implies recognizing its limitations.
“AI proposes, mankind disposes.” - A participant
The protection of an individual’s privacy and the management of personal data were heavily discussed. For example, processing healthcare data should be managed in a special way, given the highly sensitive nature of the information. It should therefore both favour methods of control ranked according to the type of use and adopt security as an operational mode. Regarding the workplace sector, the participants recommended the obligation to inform users of how their data is processed.
Aware that these recommendations involve important institutional changes, participants highlighted the need to keep in mind that AI is not necessarily desirable to begin with.
“Just because you can, doesn’t mean you should.” - A participant
The citizens generally agreed that the consequences of AI use in the different sectors—for both the individual and society as a whole — must clearly be measured to establish benchmarks without unduly hindering progress.
6.4 RESULTS: CITIZEN PERCEPTION OF THE ISSUES
The great issues of responsible AI development
Citizens that took part in the co-construction days were invited to select 2 or 3 issues to address as priorities before 2025 regarding the responsible development of artiﬁcial intelligence.
The responsibility principle was the one most often deemed a priority, followed by the autonomy principle, privacy, then well-being (individual and collective), knowledge and justice. It’s worth noting, however, that they are all closely intertwined.
The principles of knowledge, responsibility, privacy, justice and democracy are presented below by sector. As for the autonomy principle, often selected as a priority, it concerns the preservation, even encouragement of individual autonomy opposite risks of technological determinism and reliance on tools. It also raises the issue of double freedom of choice: being able to make your own choice when faced with a decision guided by AI, but also being able to choose not to use these tools without risking social exclusion. The freedom included in this autonomy principle regarding AI systems would involve any person’s capacity for self-determination.
“Develop technologies that favour human autonomy and freedom of choice.”
(Education table, Bibliothèque de Laval, March 24, 2018, Hyperpersonnalisation of education scenario).
The well-being principle also holds an important place for participants. It’s there, pervasive at every table, exhibiting a collective desire to move towards a society that is fair, equitable and that favours the development of all. Well-being is therefore both a collective (tied to equity and accessibility issues comprised in the justice principle) and an individual issue, aiming for the fulﬁllment without impeding on autonomy and privacy. Participants showed a preference for AI development “that would allow any individual to access personal and social fulﬁllment”. (Education table, Bibliothèque Père Ambroise,Montréal, March 3,2018, AlterEgo scenario).
Broadly speaking, the well-being principle was also behind a call to maintain quality human and emotional relationships between experts and users in all ﬁelds.
6.4.1 Key Issues by sectors
As for the education sector, issues regarding privacy, responsibility, well-being and knowledge were considered priorities by 6 tables out of 9. Discussions about issues dealing with the knowledge principle were especially relevant to introduce the question of transforming human skill sets in an AI era:
ISSUES DEALING WITH THE KNOWLEDGE PRINCIPLE
(6 tables out of 9)
The issues dealing with the knowledge principle for the education theme stem from skill transformation issues in a context where both the role of a teacher and the methods of developing and accessing knowledge are rapidly changing. This principle was mostly discussed under the optics of transforming the learning relationship, which would then become an issue of a teacher’s expertise whose work would have to be modified.
It was also mentioned in relation to the diversity principle to discuss the need to foster a variety of intelligence and relationships to knowledge.
“Redefining/transforming the nature of the relationship between teachers and students in the classroom and modifying relationships to knowledge.”
(SAT Table, Montreal, March 13, 2018, Nao scenario).
“Human skills and abilities: importance of developing many learning environments.”
(Musée de la civilisation Table, Quebec City, April 6, 2018, AlterEgo scenario).
LEGAL SYSTEM AND PREDICTIVE POLICE
As for the justice and predictive police sector, issues regarding privacy, responsibility and knowledge principles were considered priorities by 5 tables out of 8. Discussions about the issues concerning the responsibility principle allowed us to clarify the principle’s scope:
ISSUES CONCERNING THE RESPONSIBILITY PRINCIPLE
(5 tables out of 8)
The responsibility principle was formulated in two main ways: as a demand for human accountability in legal rulings, and by concern for who is responsible for the decision (and any potential error). The algorithm’s lack of transparency goes against accountability, in the eyes of the citizens, since it’s hard to retrace what is considered in the decision. The responsibility principle is therefore tied to the knowledge and transparency principles in regards to the claim to make decisions explainable and preserving a place for human players and their skill Sets in the legal system.
“[Justice] must remain a tool whose sole purpose is to protect individuals.
Promoting compassionate and equitable justice that accounts for singularities and past experiences.
Artificial intelligence must not have the right to judge human behaviour.
The final decision must always require human intervention.”
(SAT Table, Montreal, March 13, 2018, Preventive arrest scenario).
“Transparency, accountability and responsibility in regards to creating the tool, to the data being used, and to the tool’s consequences.”
(SAT Table, Montreal, March 13, 2018, Conditional release scenario).
In regards to healthcare, the issues concerning the privacy and responsibility principles were considered
priorities, by 9 and 10 tables out of 12, respectively. The issues concerning privacy hold particular significance for the sector given the relatively sensitive quality and near-always personal character of health data.
ISSUES CONCERNING THE PRIVACY PRINCIPLE
(9 tables out of 10)
Participants identified different issues related to confidentiality and invasion of privacy.
These issues concern a potential invasion of privacy that can be linked to the development and configuration of
AI systems (ex. which should help avoid pirating, shortages and abuse). They also deal with what the citizens called “rétroaction” (use of data previously collected for another purpose) and accessing this data through private companies. Faced with these issues, citizens worried about how to make sure the data isn’t sold, and how to guarantee that the patient keeps control over their data (especially when it’s private data), and hold imperative rights to them.
“How far are we willing to share our personal data (information) as individuals in order to feed healthcare
services?” (Musée de la civilisation Table, Quebec City, April 6, 2018, Digital Twins scenario).
As for the workplace sector, the issues concerning the justice and knowledge principles were considered priorities (respectively 5 and 4 tables out 5). All the tables that gathered around the development of AI in the workplace therefore considered that the issues concerning justice, equity and diversity should be addressed separately.
ISSUES CONCERNING THE JUSTICE PRINCIPLE
(5 tables out of 5)
he justice principle raises two main concerns:
ensuring an equitable sharing of AI benefits among all players, social groups and territories, and “installing nondiscriminatory algorithms that favour diversity, inclusion and social justice”.
(Musée de la civilisation Table, Quebec City, April 6, 2018, A socially responsible restructuration scenario).
“Sharing AI benefits (productivity gains); equity among social groups, territories (cities and regions), taking
vulnerabilities into consideration; the meaning of work in society and in the construction of our identities.”
(Musée de la civilisation Table, Quebec City, April 6, 2018, A socially responsible restructuration scenario).
SMART CITY AND CONNECTED OBJECTS
As for the smart city and connected objects sector, the issues concerning the autonomy and democracy
principles were considered priorities by 9 and 7 tables out of 11. Many issues seemed to potentially infringe on the democracy principle according to citizens:
ISSUES CONCERNING THE DEMOCRACY PRINCIPLE
(7 tables out of 11)
Participants discussed issues tied to the balance between collective interests and individual needs; to managing access to public spaces and sharing those spaces, or even sharing the benefits stemming from the development of AI technologies (namely, between individuals, the public sector and the private sector). They insisted on the necessity and the difficulty of ensuring a collective (involving citizens) and enlightened (which implies a certain
transparency regarding the development of AI systems) decision-making process to define guidelines around connected objects. Citizens also questioned the true independence of public authorities in regards to AI development, and out forward the risk of normalizing behaviour that could lead to marginalization, thereby running the risk of infringing on the democracy principle.
“How can we manage an intelligent transportation system in democratic fashion?”
(Du Boisé Library Table, Montreal, March 17, 2018, Self-driving car scenario).
6.4.2 Issues that could lead to the creation of new principles
Different issues identified or discussed by citizens seem particularly interesting and could lead to the
eventual creation of new principles in the Montreal Declaration, namely for their transversal aspect (for both the sectors and the principles).
For example, the impact of the responsible development and use of AI on the environment.
These issues ask how to guarantee the responsible and equitable use of material and natural resources.
They also raise the matter of ensuring a positive energy balance when it comes to the polluting effects of AI and the technologies associated with its use.
“We forgot to talk about the environmental aspect: the stocking of data, the problem of an outrageous
accumulation of data and the costs in terms of energy (or room) that involves.” - A participant
SPECIFY THE JUSTICE PRINCIPLE: DIVERSITY AND EQUITY
The justice principle was discussed according two types of issues: sometimes in terms of diversity, sometimes in terms of equity and social justice. This principle could therefore be split in two to put forward each of these aspects:
A diversity principle could therefore aim to prevent discrimination by finding mechanisms free of biases tied to sex, age, mental and physical capacity, sexual orientation, social and ethnic origins and religious beliefs, without creating any new ones. The diversity principle also calls upon favouring a multitude of perspectives and intelligence rather than standardizing individual profiles according to a limited number of categories and criteria.
“A loss of diversity brings extreme standardization. It comes back to the need to maintain complexity when dealing with human questions.” - A participant
A social justice and equity principle would involve making AI benefits available to all, and that AI development will not contribute ton increasing social and economic inequalities, but rather reduce them.
“Accessing new technologies: a privilege only the rich can afford?
Will this type of technology widen inequalities?” - A participant
ISSUES OF TRUST, RELIABILITY, SECURITY: A CAUTION PRINCIPLE?
Furthermore, the issues concerning trust in the development of AI technologies were regularly brought up. The issue of trust in AI and its algorithms in different sectors mainly presents itself as a certain suspicion towards these techniques as well as how representative the selected data and the validity of the interpretations made really are, suggesting a caution principle:
“Since it’s scientific, a person could tend to forget that an algorithm can be wrong: caution.”
- A participant
This trust issue is also closely tied to the question of the reliability of AI systems. Paying close attention is paramount to ensuring the quality of the collected data and the correlations that can be made as well as their purpose in order to avoid blind faith and prevent potential manipulation. Along the same lines, the participants raised issues tied to the security of AI devices, namely the risk of potential abuse, pirating and cyberattacks on
the systems and the data they hold, as well as the validity of the recommendations and decisions made
by algorithms. These issues are accompanied by a dilemma between “collective fluidity” and “system
vulnerability”, meaning AI that must be both flexible and solid (ex. in the smart city and connected objects sector).
TOWARDS A TRANSPARENCY PRINCIPLE?
At the crossroads of the knowledge, responsibility and justice principles lies a transparency principle that implies being able to understand an algorithmic decision and react to it. For that to happen, citizens must insist on algorithmic procedures being explainable so that anyone can understand and verify the criteria that was taken into consideration when making the decision:
“Transparency in the variables used, the data, the parameters. Explaining a decision in clear, natural language.” (Workplace Table, Bibliothèque Mordecai-Richler, Montreal, March 10, 2018, AI as mandatory path to the workforce scenario).
This explainability issue implies the necessity of finding a way to simplify these algorithmic procedures so anyone can make sense of them; this goes hand-in-hand with the development of digital literacy, which will enable enlightened consent and a critical mind towards the system. The explainability of these algorithms accompanies the issue of being able to verify algorithmic decisions, hold someone responsible for those decisions and
eventually correcting certain negative effects such as discriminatory biases. It’s also about making these algorithm explanations accessible, for example through open development (open source, licence free, open data), namely out of a concern for feedback to understand why a decision was made and manage eventual feelings of injustice after being refused (when applying for a job, social assistance or insurance, for example).
A warning was made on two occasions, however, regarding this transparency principle: This transparency could have a potential effect on the security of the algorithmic systems (risk of hacking). This transparency principle would tie into the issue of trust in AI technologies.
“If AI analyzes things that are too complex for the human brain, who’s keeping an eye on what’s going on behind the curtain?” - A participant.
6.4.1 Issues regarding the relationship between humans and AI
Regardless of the sector, the citizens identified many issues regarding the relationship between humans
and AI. Participants namely worried about the place left for humans in such systems, which could lead to
For instance, citizens are concerned with the respect of “human nature”, across all sectors. It is confronted with the place of the object in society and its relationship to the human for the smart city and connected objects sector: will reduce the status of the human being and grant more importance to protecting objects? Respecting “human nature” also means, in the eyes of the citizens, guaranteeing to take into consideration some singularity, some complexity, some human messiness according to numerous parameters that are hard to quantify, such as what the participants called “individual charisma” in the legal sector. Not taking into account human dynamics and its possibilities for change shows a concern with the “static” vision of a human being provided by the algorithm, which would make its decisions problematic and unreliable. In healthcare, this reliability is also questioned when it comes to a diagnosis or suggestions provided by algorithms that have no holistic visions of the individuals, who can’t be reduced to their biological data.
Participants are worried about an eventual dehumanizing of services which could appear if AI is granted too much space. In healthcare, it’s a certain dehumanizing of care and the loss of the doctor-patient relationship. In the smart city and connected objects sector, participants are concerned about striking a balance for a harmonious development of society and human beings while implementing AI and connected objects. In the workplace sector, this dehumanizing can be perceived as the automation of tasks. In the legal sector, it could come from a potential lack of “empathy”, “instinct”, “wisdom” in AI systems, which raises concerns about prosecuting cases rather than treating them “humanly”:
“Cases will become standardized and the person themselves won’t be considered enough.”
- A participant
Participants are worried about a loss of emotional and relational quality, sometimes seen as a potential
“denaturalizing”, even an “alienation” (from social life in favour of digital life), this across all fields.
These concerns namely refer to the transformation in the relationship to care, knowledge, wisdom, work,
but also the skills of individuals.
“The challenge isn’t making machines more intelligent, it’s making humans more intelligent.”
- A participant
Will doctors still have the same expertise if they’re constantly relying on expert systems? What effect will that have on the trust placed in their expertise as opposed to the AI’s? A similar reflection was brought up in the education sector regarding replacing teachers with AI:
“If there’s too much AI equipment in schools, teachers will become useless.”
- A participant
Still in the education sector, the citizens reminded everyone that human intervention is necessary:
“We can’t rely solely on a machine."
- A participant
Finding a path to complementarity between humans and AI therefore seems very important. This complementarity was discussed as a “good balance” for sharing tasks (for example, between the “objective” and “subjective” for the workplace sector, between an infinitely patient AI as a “learning assistant” and a teacher with emotional and relational capacities for the education sector). In the workplace sector, citizens suggested implementing a watch to preserve “human primacy”, which should guarantee that technology is only a support:
“A guarantee that the system is not an end in itself, but that it is focused around the human."
- A participant
Workplace Table, Musée de la civilisation, Quebec City, April 6, 2018, AI as mandatory path to the workforce scenario).
6.5 RESULTS: POTENTIAL SOLUTIONS
6.5.1 The general potential solutions suggested by citizens
Citizens who took part in the co-construction days were invited to suggest potential solutions to the previously identified issues. 190 potential solutions were formulated and adopted through consensus during these activities (although other suggestions may have been discussed during the tables.)
All co-construction tables agreed on 3 general potential solutions to guarantee socially responsible AI development, regardless of sector:
1. Legal dispositions
2. Putting training in place for all
3. Identifying independent key players for AI management.
Regardless of the sector, all tables agreed on recommending implementing a legal framework adapted to the reality of AI development and personal data management (especially massive data). For example, participants recommended implementing specific rules and laws, new types of contracts, even putting a moratorium in place. Implementing training that is accessible to all was also strongly recommended, both for professionals of the affected sectors (to guarantee adequate use of AI systems in their work) and the general population (to guarantee everyone can participate in the debate and gain basic digital literacy).
Citizens also identified the institutional players and the key independent and competent players (existing or to be created) who would oversee the responsible development of AI. The players identified are people (ex. ombudsman, auditor, life and well-being commissioner) or groups of people (ex. setting up an artificial intelligence centre for civilian security, a 1–800 number against connected objects discrimination or a Ministry odf data ethics and digital protection).
In all sectors as well, citizens suggested creating technical and ethical evaluation mechanisms for AI. Namely, establishing a certification (or label) system as an ethical guarantee was suggested on many occasions. Different tables also recommended implementing a code of ethics (whether it’s a matter of updating the existing code or creating new ones); and participatory mechanisms (ex. co-constructions, public consultations or an AI summit) in order to guarantee a democratic development of AI and its management. The importance of implementing research programs in various disciplines (ex. philosophy, social sciences, bioethics) was also raised. The creation of digital tools (ex. digital and interactive healthcare forms, individual digital file in the workplace sector) was also suggested.
Developing incentives that aim to encourage responsible development – was agreed upon at different tables, as was implementing diversity quotas (which reward companies that guarantee not to exclude or discriminate against certain minorities through AI biases) or funding companies that establish transitions for employees whose job is being replaced by AI. Finally, establishing professional frameworks (and different internal procedures for companies) and the creation of public policies that could lead, for example, to the creation of a digital citizenship, were all put forward.
6.5.2 Potential solutions by sector
Citizens gathered around 9 co-construction tables in which the theme of AI development in the education sector was discussed. Participants formulated 27 potential solutions or general AI framework guidelines during these activities.
7 POTENTIAL SOLUTIONS CONCERNING TRAINING WERE FORMULATED BY 4 TABLES OUT OF 9
In regards to education, participants recognized the need to be proactive in setting up training for the entire community affected by AI development in that sector. This training should cover digital literacy, media literacy, as well as ethics and the issues tied to integrating AI in an educational environment. This training could, for example, take the form of digital literacy accompaniment for both parents and students, or be directly integrated into the initial citizen training.
The citizens also recommenced training education professionals more specifically, for instance by
including the development of work skills “teamed up” with AI devices in the curriculum for the initial
and university training of teachers (ex. a certification for the B.Sc. or an accreditation system). This training
will have to be both technological (how to use AI), but also geared towards teaching techniques with AI
(how to organize teaching sequences and insisting on the fact that knowledgeable professionals
orchestrate AI, not the other way around.
“Accrediting agents of change (both psychoeducators and active teachers) by teaching establishment to gradually integrate AI in an academic environment.”
(SAT Table, Montreal, March 13, 2018, AlterEgo scenario).
The importance of establishing adequate training was also raised. The training’s purpose would be to provide the appropriate information allowing stakeholders to accept their responsibility towards AI, in order to avoid teachers putting blind faith in educational AI devices. This training would accelerate the understanding of actors in the field of education and favour their mobilization to develop AI so it serves the autonomy of the learners while preparing them to deal with these realities. This training will help develop human skill sets and
provide power to guide and even redefine future AI development.
“Raise awareness around responsible use of AI and promote a diversity of relationships to knowledge.”
(SAT Table, Montreal, March 13, 2018, Nao scenario).
LEGAL SYSTEM AND PREDICTIVE POLICE
Citizens gathered around
8 co-construction tables to discuss the theme of AI development in the legal sector.
Participants formulated 36 potential solutions or general AI framework guidelines during these activities.
10 OF THE POTENTIAL SOLUTIONS FORMULATED ARE LEGAL DISPOSITIONS AND ARE RECOMMENDED
BY 7 TABLES OUT OF 8
In regards to the legal system and predictive police, it is imperative to establish laws and regulations on transparency: it’s a matter of demanding transparency from private and public companies collecting criminal data, but also of laying bare the decision-making processes when these decisions are made by algorithms. Explaining the decision must come with measures allowing access to mobilized algorithms and ensuring they are explained in intelligible fashion.
As a first transparency mechanism, many participant tables suggested that the AI used in the legal sector—even all public sector AI—be developed in open code, under free licence. From a legal standpoint, it’s about guaranteeing “the right to a full answer and defence”, namely with the possibility to challenge a decision by raising procedural or formal deficiencies (Table Musée de la civilisation Table, Quebec, April 6, 2018, Parole scenario)
This transparency imperative goes hand-in-hand with establishing legal dispositions giving the right, believed to be fundamental, to be judged by a human being to preserve procedural justice and individualization of the sentence. Underlining the need for law to adapt to a new technological reality with AI in legal decision-making, many debates occurred around conciliating human and artificial
players in this process. The consensus was as follows:
“The right to appeal before a human judge: The appeal procedure for a decision made by a computer must
always be heard by a human judge.” (Musée de la civilisation Table, Quebec City, April 6, 2018, Parole scenario).
In the perspective of preventive AI used for police purposes, it is mentioned that there is a desire to establish a “framework that allows us to go beyond and eliminate biases, discrimination and abuse of power” (SAT Table, Montreal, March 13, 2018, Predictive Arrest scenario) as well as reinforce laws around consent to ensure it is truly an enlightened one. There’s also the idea of limiting public and private stakeholders access to private data such as “private conversations on digital platforms” (Du Boisé Library Table, March 17, 2018, Preventive Arrest scenario) and enforcing a “right to be forgotten, to modify and correct data as well as a right to personal access to the data gathered” (Père Ambroise Library Table, March 3, 2018, Predictive Arrest scenario).
Citizens gathered around 12 co-construction tables to discuss the theme of AI development in the healthcare sector.
Participants formulated 46 potential solutions or general AI framework guidelines during these activities.
8 FORMULATED POTENTIAL SOLUTIONS ARE AI EVALUATION DEVICES (IN HEALTHCARE, CERTIFICATIONS), AND ARE RECOMMENDED BY 8 TABLES OUT OF 12:
AI EVALUATION DEVICES
Citizens recommended establishing AI ethical certification in healthcare, meaning the development of a certification (or label) for algorithms and robots, on the database from research projects (participative study on the context that influences AI development) to determine the criteria for this certification and its various levels. These criteria should include transparency, security and relevance of the tool. For example, these certifications would be designed to standardize access to the decision-making process of the algorithms, or to validate the tools of healthcare robots. These certifications should be issued by the government or independent, multiparty organizations to protect public interest and patient well-being, would mainly target private companies developing AI healthcare.
“Upfront certification for healthcare robots and their toolbox (namely, to protect public interests)”
(Mordecai-Richler Library Table, Montreal, March 10, 2018, Helper robots for the elderly scenario).
Citizens gathered around
5 tables to formulate
32 potential solutions
regarding AI development
in the workplace
THE SUGGESTIONS CONCERNING PARTICULARLY STOOD OUT. THEY WERE RECOMMENDED BY 3 OUT
OF 5 TABLES
Participants suggested creating a multi-sectorial “permanent consultation space” within the government, to respond to the division of powers (tied to the democracy principle). The information gathered digitally could then be more accessible and that space would be responsible for structuring sectors emerging in the field of employment.
Citizens also mentioned the importance of user participation in designing the interface of AI tools, which could take the form of “design thinking” with different partners and would allow them to review the work of the programmers:
“Allowing user input in machine learning through open AI (based on the Wikipedia model) to correct and review biases by and for society.” (Musée de la civilisation Table, Quebec City, April 6, 2018, AI as mandatory pathway to employment scenario).
User feedback should help follow data collection and algorithm development, and reduce the “gaps” that could lead to prejudice towards individuals from competent authorities (ex. ethics committees, corporations) to adapt the system.
SMART CITY AND CONNECTED OBJETS
Citizens gathered around
11 tables on the theme of AI development in the smart city and connected objects
These 11 tables formulated 51 potential solutions.
THE SUGGESTIONS CONCERNING PARTICULARLY STOOD OUT. THEY WERE RECOMMENDED BY 3 OUT OF 5 TABLES
INSTITUTIONAL PLAYERS AND OTHER PLAYERS
Table participants discussing the theme of smart city and connected objects suggested many ideas for the creation of institutional players, whether independent societies or advisory committees. The democratic ideal of committees or assemblies allowing citizen participation was recalled many times. For the control of connected objects, 2 models were therefore suggested, including a mechanism forcing the self-regulation of private players:
Based on the model of the Régie du logement du Québec, a Régie des objets connectés (connected object management) would help set prices for connected objects (such as refrigerators) and would set forward social assistance to facilitate their acquisition. It would also issue ownership certificates when purchasing a connected object to establish that the data generated by this object belongs to the user. This person can then choose to give their consent or not for the data to be communicated to the company commercializing the object as well as their insurance, without risking any penalties
An independent authority on data management could allow citizens to conduct a class action when there are abusive uses. It could also manage a digital platform where users can speak freely and publicly about the advantages and disadvantages of AI devices and thereby have an impact on the branding of private players commercializing these devices. The private players would then be forced to self-regulate through the pressure users place on their image (Musée de la civilisation Table, Quebec City, April 6, 2018, Connected refrigerator scenario).
To respond to an equity issue and thereby ensure an equitable sharing of AI, an advocate could be reached at
“1–800 discrimination of connected objects” (INM Table, Montreal, February 18, 2018, Connected refrigerator scenario). It could then be a part of a “multiparty committee that democratically manages incidents, injustices and other issues” (Mordecai-Richler Library Table, Montreal, March 10, 2018, Self-driving car scenario). Furthermore, an independent auditor could be mandated to lead an accounting audit to ensure an equitable sharing of AI benefits (INM Table, Montreal, February 18, 2018, Connected refrigerator scenario).
For self-driving car regulation, the creation of the SAIAQ (Société de l’Assurance de l’Intelligence Artificielle du Québec) would bring modifications to road safety laws to adapt them to autonomous driving. It would also include auto insurance 2.0 that would suggest new kinds of contracts for this type of driving (Bibliothèque du Boisé Table, Montreal, March 17, Self-driving car scenario).
7. A CONTINUOUS CO-CONSTRUCTION PROCESS
7.1 CONTINUE THE DELIBERATION
The Montreal Declaration project concentrated its ﬁrst phase on ﬁve key sectors: education, health, work, smart city and predictive police. An entire year of co-construction wouldn’t even cover all the reﬂection themes. The co-construction initiative will therefore continue in September 2018, allowing for discussions about new themes that had barely been touched upon in the scenarios used in the co-construction phase.
Co-construction workshops on these themes will produce a series of analyses and suggestions that will complete those that were produced in the ﬁrst co-construction phase. They will also feed the reﬂection on the ethical principles of the Montreal Declaration and the recommendations for a public policy on AI which will be developed in their extension.
We will present public policy recommendations around priority ﬁelds of action. The priority ﬁelds of action are transversal recommendation axis with sectors and themes. We will only reveal the priority ﬁelds, and the recommendations, once the deliberation process is complete, but we can already say that three ﬁelds of action have established themselves:
1) Digital literacy
2) Diversity and inclusion
3) Transition and social mutations
7.2 AN INSPIRING INITIATIVE
Among the initiatives inspired by the Montreal Declaration, we must ﬁrst mention the work of the Montreal AI Ethics Meetup group, founded and coordinated by Abhishek Gupta (McGill). This group, which brings together over one hundred multidisciplinary researchers and concerned citizens concerned by AI developments, devoted many 2h sessions, between December 2017 and March 2018, to the Declaration principles. Although they are not citizen deliberations, the Montreal AI Ethics Meetup sessions are nonetheless true collective intelligence exercises that involve a variety of high- level researchers. A detailed report of their critical reﬂections was submitted to the Declaration team and can be read online. The authors of the report are : Stephanie Dyke, Paule-J Toussaint, Abhishek Gupta, Gregory Caicos, Marc Daher, Peter Chen. This initiative is especially encouraging because it comes from the heart of Montreal’s AI community.
Another important initiative was the evening of reﬂection organized by ESG UQÀM (École des Sciences de la Gestion de l’UQAM) on February 15, 2018, entitled: “Vers un développement responsable de l’IA : Soirée de réﬂexion autour de la Déclaration de Montréal pour un développement responsable de l’ia” (Towards a responsible development of AI: Evening of reﬂection around the Montreal Declaration). This evening brought together seven researchers from UQÀM around the seven values of the Declaration. After a general introduction by Yoshua Bengio and Martin Gibert, the audience had a chance to hear the thoughts of professors Marie-Jean Meurs (well-being), Christophe Malaterre (autonomy), Hugo Cyr (justice), Sébastien Gambs (privacy), Étienne Harnad (knowledge), Dominic Martin (democracy) and Maude Bonenfant
(responsibility). The summary of the exchanges that took place during this university meeting can be read online (21).
At Université de Montréal, the Faculty of Arts and Sciences created Perspective in February 2018, an interdisciplinary lab of ideas whose ﬁrst is explicitly aligned “in the tracks of the Montreal Declaration”. This lab brings together a group of graduate students tasked with producing reports with the intention
of enlightening public policymakers of the social impacts of AI.
Finally, many organizations (businesses, development organizations or associations) showed their interest in the Montreal Declaration, hosted presentations of the Declaration or organized discussions about its principles. This is the case for IBM, Montréal InVivo, l’ACFAS, Printemps numérique, or C2 Montréal.
The citizen involvement and consultation process initiated by the Montreal Declaration is growing and is now operating outside Quebec, in Toronto under the impulse of the ICRA, and also soon in Europe, Brussels and London.