The unveiling of the Montreal Declaration for Responsible Development of Artificial Intelligence marks the culmination of a year of intense work. Work that, in many ways, is just beginning, says Maxime Johnson.
“An ethical compass that will guide the development of artificial intelligence to morally and socially desirable ends” is the goal of the Montreal Declaration, a document presented this week in the city by researchers and academics from around the world. varied horizons, from artificial intelligence to philosophy to public health and criminology.
After a long reflection and public consultations conducted in front of hundreds of people in Quebec and abroad, the committees responsible for the creation of the document propose 10 main principles to frame the artificial intelligence.
These principles ensure, for example, that artificial intelligence systems (AIS) improve the living conditions, health and working conditions of people (welfare principle), guarantee the confidentiality of data and the anonymization of personal profiles (principle of protection of privacy and privacy), and that they should not lead to a standardization of society through the normalization of behaviors and opinions (principle of inclusion diversity).
These principles are also accompanied by recommendations to help policymakers develop modern public policies adapted to artificial intelligence. The risks associated with artificial intelligence are great, but the Montreal Declaration is a thoughtful and comprehensive guide, which clearly explains the guidelines that universities, businesses, and governments must put in place so that it has a positive impact. on society.
The authors of the Declaration, however, are not fooled. “I’m very proud of what we did, but it’s far from enough. We can pronounce a lot of beautiful words, but if it does not translate into action, it will not help much, “warned Yoshua Bengio, professor in the Department of Computer Science and Operations Research (DIRO) of the Université de Montréal and member of the orientation committee of the Declaration, during the presentation of the document.
Here are five challenges that will need to be addressed if the Declaration is to become more than a set of good intentions.
1. Rallying big companies to the cause
The Montreal Declaration is not binding. Rather, it invites interested parties to sign it voluntarily, without any assurance that its principles will be respected. As Yoshua Bengio pointed out at the presentation, “some principles are pretty radical and require asking questions that change the way things work.” Some may even conflict with other business goals.
In the principle of democratic participation, the Declaration states, for example, that “the code of algorithms, public or private, must always be accessible to the competent public authorities and relevant stakeholders for verification and control purposes”.
The Declaration does not ask to open its code to the public, but the majority of companies, both Silicon Valley giants and local companies, will probably be reluctant to voluntarily submit their algorithms to the authorities.
The document obviously has an interest even if it is not followed in its entirety. A report released in September by SAS, Accenture, Intel and Forbes Insights, for example, indicated that 63% of companies that have adopted AI have ethics committees in place to examine the use of this technology (a proportion that is climbing 73% in Canada). The Montreal Declaration risks becoming a powerful tool for these committees, whether signed or not. But for that, it will convince companies to use it.
2. Accompany the Declaration with a political approach
Reflection on the Montreal Declaration has highlighted several elements that governments should adapt to the advent of artificial intelligence. “The legal framework does not go far enough in taking into account new digital realities,” said Marc-Antoine Dilhac, professor in the Department of Philosophy at the University of Montreal and co-scientific director of the co-construction of the Declaration.
For the latter, the laws surrounding personal data should, for example, adapt to the objectives of their collection. People are more likely to communicate their data for the creation of algorithms for the diagnosis of rare diseases than for political targeting in the manner of Cambridge Analytica, at the heart of the Facebook user data scandal.
The creation of a “data ministry,” an idea raised during the consultations surrounding the creation of the Montreal Declaration, could be a good way to modernize the machinery of government.
3. See the heart of the black box
Some of the challenges of the Montreal Declaration are also more technical. The reasoning behind the decisions made by an artificial intelligence is not always obvious, even for those who created the system. Some SIAs, especially those using artificial intelligence techniques such as deep learning, make decisions based on the data transmitted to them, but without providing any justification. This phenomenon is also known as the “black box”.
The industry is currently working on methods to make it clearer (IBM has presented a research on the subject this week in Montreal at the NeurIPS 2018 scientific conference), but as long as the models of deep learning will not explain better what they do, some principles of the Declaration may be difficult to
For Marc-Antoine Dilhac, the black box is not just a technical issue. “It’s often rhetorical. This is something that suits companies well, “he says. Some would hide themselves a little too quickly behind the black box to wash their hands of an ethical problem, instead of thinking about the discrimination that could result from the data they use to train their artificial intelligence models, for example.
4.Develop the Declaration
The Declaration is conceived as “an open, revisable and adaptable guidance document based on the evolution of knowledge and techniques, and feedback on the use of AI in society”. With the place that artificial intelligence is called to occupy in our lives, this constant update will be more than necessary.
The good news is that the International Observatory on the Societal Impacts of Artificial Intelligence and Digital Technology, whose creation was confirmed on Monday, will be in a good position to ensure the evolution of the Declaration.
It will, however, be important to give it more than a new layer of annual polish, especially if the sector finds after a few years that the approach favored by this first edition was not the right one and that a more restrictive solution should rather be considered, for example.
5.Ensure citizen participation
The people played a role throughout the co-creation of the Montreal Declaration. People’s participation, however, should not stop there. “Citizens are called to show their vigilance,” said Marc-Antoine Dilhac. People working in the field of artificial intelligence, but also companies and authorities who use these services, must become whistleblowers, believes the scientific co-director of the Declaration.
“Those who notice biased or unreliable results have to say it,” he adds.
The concept may seem far away for many, but with the expected advances in artificial intelligence in the coming years, people in touch with technology should multiply. Reading the Montreal Declaration for the Responsible Development of Artificial Intelligence is therefore advisable now, whether one is working in the field or not.