Artificial intelligence (AI) is still in its infancy. “Its level of reasoning is very superficial, not even equivalent to that of a frog,” says Yoshua Bengio, AI pioneer and an expert on deep learning. However, it already poses serious problems of monopolization and inequitable distribution, which can only be resolved on a global scale. International coordination is imperative in the development of AI, he cautions.
Yoshua Bengio, interviewed by Jasmina Šopova
Over the last five years or so, basic AI research has been all the rage with some information technology giants, who are investing considerable sums of money in the field. Could you explain this phenomenon?
The answer is very simple. Science in AI has reached a level of maturity that makes it very useful for companies. The accumulation of big data and the increased computing power available, facilitate the development of new AI products, which will be even more profitable in the future than they are today.
Today, when we search the internet, we are constantly solicited by targeted advertising – these ads allow companies like Facebook, Amazon, YouTube, etc., to thrive. Currently, AI products have only a small share of the market. But economists predict that they will account for up to 15 per cent of the total production of goods within a decade. That is enormous.
AI will then allow these companies to sell more, to get rich and to be able to pay the researchers they recruit even more than they do now. By increasing their customer base, they will increase the amount of data they have access to – and that data is a gold-mine that makes the system even more powerful.
All this creates a virtuous cycle, which is good for these companies but unhealthy for society. Such a concentration of power can have a negative impact on both democracy and the economy. It favours large companies and slows down the ability of small new companies to enter the market, even if they have better products to offer.
We must encourage greater diversity in the business world associated with AI and avoid a monopoly situation.
But the monopoly is already being established. How can this be remedied?
With anti-monopoly laws. History teaches us that they can be effective against the excessive power of some companies. Remember Standard Oil in the United States, which bought its competitors to monopolize the oil market? Or Hollywood, which until the middle of the twentieth century, controlled seventy per cent of film theatres and imposed its rule on the distribution of films? The legal decisions against these companies and some others, helped to rebalance the markets.
I believe that judicious advertising regulations can go a long way towards preventing the establishment of monopolies in AI research. We are all, in a way, prisoners of advertising and we often forget that we have the option of making a collective decision to regulate it, so that it is not harmful to society.
Besides, the services provided by large private companies like Google and Facebook could very well be made public − in the same way that television, which provides a similar service, is.
You have decided not to work in the private sector, haven’t you?
Yes, I want to remain neutral. My project is to develop a science that is accessible to everyone, and not only to a few shareholders. I want research to develop in a way that it targets the most useful applications for humanity − and not necessarily the most profitable for the economy.
That said, I have tried to create a common ecosystem that is mutually beneficial to research and industry at the University of Montreal, where I work. Several private laboratories have been set up in Quebec’s capital, and they collaborate with us. Researchers from industry are employed as associate professors at the university and help train students. Companies make donations to universities and give them complete freedom to choose which areas of research they will invest in.
What is the proportion of researchers working in the academic field today?
If I base my answer on the people I meet at major international conferences, I would say that it is about half. Five years ago, virtually all AI researchers worked in the academic field.
Private companies recruit talent from around the world. Does this contribute to brain drain in less developed countries?
Inevitably. That’s why we must think collectively about how the poorest countries can benefit from the most recent research results − but also about how to create research centres within their universities. In Africa, for example, more and more academic institutions are offering courses in AI and summer schools are being organized, which are proving very useful.
In addition, there are a large number of courses, tutorials and codes available online for free. I meet many young people who have been trained through the internet. We must also look for the best ways to help these students train themselves.
Some countries, including Canada, are investing heavily in AI research.
Yes, Canada has decided to fund not only basic research and to help startups, but also to invest in collective thinking and research in the social sciences and humanities, in order to assess the social impact of AI.
At the initiative of the University of Montreal, a debate was started on 3 November 2017 to help develop the Montreal Declaration for a Responsible Development of Artificial Intelligence. This approach essentially aims to establish ethical guidelines for the development of AI at the national level.
In the first phase of this long-term participatory process, the general public is invited to debate with experts and policymakers. Seven values have been identified: well-being, autonomy, justice, privacy, knowledge, democracy and responsibility.
At what stage is this reflection, at the international level?
To my knowledge, there is no international treaty governing AI research. Yet, these are international issues and without international coordination, we will not be able to move forward in the right direction.
First and foremost, the general public and policymakers must be made aware of the concerns about AI. In some parts of the world, researchers have already issued warnings about major problems, and the media and general public have responded. These are the first steps that will lead us towards a broader global political dialogue on the problems posed by this discipline, particularly in the areas of ethics, the environment and security.