The emergence of artificial intelligence (AI) promises dramatic changes in our economic and social structures as well as everyday life in Europe and elsewhere; it has been compared to both electricity and the internet. Both are general, ubiquitous, and reshaped the world. But the internet analogy is more apt, for while electricity requires standards and regulation, it just works the way it works; but the functioning of the internet and its economy were largely shaped by key policy choices made along the way. We now sit in the early days of AI, and the choices we make over the next decade will crucially shape its place in and relation to society. We applaud the European Commission for tackling the challenge of determining the role that government can and should play and support meaningful regulations of AI systems in high-risk application areas. The stakes are high, and the potential ability of AI to remake institutions means that it is wise to consider novel approaches to governance and regulation, rather than assuming that existing structures will suffice.

The Commission will undoubtedly receive detailed feedback from many corporations, industry groups, and think tanks representing their own and others’ interests, which in some cases involve weakening regulation and downplaying potential risks related to AI. We hope that the Commission will stand firm in doing neither. Moreover, as experts who have been involved for years or decades in developing the core technologies, we would like to emphasize one central point: that while it is difficult to forecast exactly how or how fast technological progress will occur, it is easy to predict that it will occur. It is imperative, then, to consider AI not just as it is now, represented largely by a few particular classes of data-driven machine learning systems, but in the forms it is likely to take.

AI does and will come in many forms, including as intelligent software tools, as integrated into massive online systems, and as instantiated as software agents designed to substitute for humans. Each of these raises particular issues and challenges: how do we govern recommendation tools whose recommendations are difficult to predict or understand? How do we manage massive systems that mediate interactions between people, and in which people serve as part of the system? What do we do with software agents that replace people in their jobs or impersonate people in their interactions?
These and many other questions are challenging but largely addressable through proper governance for today’s AI systems. But in each case AI systems of the future will be more capable, more flexible, more general, more continually learning — in short, more intelligent! Laws and regulations can have a defining role in industries, set powerful precedents, and can sometimes hold sway long after their intended lifespan. It is important that in crafting legislation now, the Commission considers, in consultation with high-level experts, the many forms that AI is likely to take, and the capabilities that it will at least potentially have in years to come.

The EU has already shown foresight and clear leadership in adopting meaningful regulations in other technology issues. We, the co-signed experts, support the Commission in taking a meaningful, future-oriented approach regarding the effects of AI systems on the rights and safety of EU citizens.