EPFL computer science professor tells conference that AI has no legitimate roll in defining, implementing, or enforcing public policy.
Artificial intelligence shapes our modern lives. It will be one of the defining technologies of the future, with its influence and application expected to accelerate as we go through the 2020s. Yet, the stakes are high; with the countless benefits that AI brings, there is also growing academic and public concern around a lack of transparency, and its misuse, in many areas of life.
It’s in this environment that the European Commission has become one of the first political institutions in the world to release a white paper that could be a game-changer towards a regulatory framework for AI. In addition, this year the European Parliament adopted proposals on how the EU can best regulate artificial intelligence to boost innovation, ethical standards and trust in technology.
Recently, an all-virtual conference on the ’ Governance Of and By Digital Technology’ hosted by EPFL’s International Risk Governance Center (IRGC) and the European Union’s Horizon 2020 TRIGGER Project explored the principles needed to govern existing and emerging digital technologies, as well as the potential danger of decision-making algorithms and how to prevent these from causing harm.
Stuart Russell, Professor of Computer Science at the University of California, Berkeley and author of the popular textbook, Artificial Intelligence: A Modern Approach, proposed that there is huge upside potential in AI, but we are already seeing the risks from the poor design of AI systems, including the impacts of online misinformation, impersonation and deception.
"I believe that if we don’t move quickly, human beings will just be losing their rights, their powers, their individuality and becoming more and more the subject of digital technology rather than the owners of it. For example, there is already AI from 50 different corporate representatives sitting in your pocket stealing your information, and your money, as fast as it can, and there’s nobody in your phone who actually works for you. Could we rearrange that so that the software in your phone actually works for you and negotiates with these other entities to keep all of your data private?" he asked.
Reinforcement learning algorithms, that select the content people see on their phones or other devices, are a major problem he continued, "they currently have more power than Hitler or Stalin ever had in their wildest dreams over what billions of people see and read for most of their waking lives. We might argue that running these kinds of experiments without informed consent is a bad idea and, just as we have with pharmaceutical products, we need to have stage 1, 2, and 3 trials on human subjects and look at what effect these algorithms have on people’s minds and behavior."
Beyond regulating artificial intelligence aimed at individual use, one of the conference debates focused on how governments might use AI in developing and implementing public policy in areas such as healthcare, urban development or education. Bryan Ford, an Associate Professor at EPFL and head of the Decentralized and Distributed Systems Laboratory (DEDIS) in the School of Communication and Computer Sciences , argued that while the cautious use of powerful AI technologies can play many useful roles in low-level mechanisms used in many application domains, it has no legitimate role to play in defining, implementing, or enforcing public policy.
"Matters of policy in governing humans must remain a domain reserved strictly for humans. For example, AI may have many justifiable uses in electric sensors to detect the presence of a car - how fast it is going or whether it stopped at an intersection, but I would claim AI does not belong anywhere near the policy decision of whether a car’s driver warrants suspicion and should be stopped by Highway Patrol."
"Because machine learning algorithms learn from data sets that represent historical experience, AI driven policy is fundamentally constrained by the assumption that our past represents the right, best, or only viable basis on which to make decisions about the future. Yet we know that all past and present societies are highly imperfect so to have any hope of genuinely improving our societies, governance must be visionary and forward looking," Professor Ford continued.
Artificial intelligence is heterogeneous and complex. When we talk about the governance of, and by, AI are we talking about machine learning, neural networks or autonomous agents, or the different applications of any of these in different areas? Likely, all the above in many different applications. We are only at the beginning of the journey when it comes to regulating artificial intelligence, one that most participants agreed has geopolitical implications.
"These issues may lead directly to a set of trade and geostrategic conflicts that will make them all the more difficult to resolve and all the more crucial. The question is not only to avoid them but to avoid the decoupling of the US from Europe, and Europe and the US from China, and that is going to be a significant challenge economically and geo-strategically," suggested John Zysman, Professor of Political Science at the University of California, Berkeley and co-Director of the Berkeley Roundtable on the International Economy.
"Ultimately, there is a thin red line that AI should not cross and some regulation, that balances the benefits and risks from AI applications, is needed. The IRGC is looking at some of the most challenging problems facing society today, and it’s great to have them as part of IC," said James Larus, Dean of the IC School and IRGC Academic Director.
Concluding the conference, Marie-Valentine Florin, Executive Director of the IRGC reminded participants that artificial intelligence is a means to an end, not the end, "as societies we need a goal. Maybe that could be something like the Green Deal around sustainability to perhaps give a sense to today’s digital transformation. Digital transformation is the tool, and I don’t think society has collectively decided a real objectivel for it yet. That’s what we need to figure out."