AI safety first, through strong regulation, and massive research investments.

What seems clear, from that list, is that the consensual aspects call for a lot of attention and energy to be devoted towards AI safety. Beyond all the debates I and others take part in, I really hope this message comes out loud and clear:

- Some of us believe that keeping access to future AI to a restricted set of trustworthy people and regulated organizations, i.e., more “closed”, will lead to safer outcomes; while others believe it’s keeping them “open” that will usher in the best possible future.

- Some think that even talking about future risks is a bad idea, because it prevents us from focusing on existing harms; others think that these future harms are potentially so important that it would be irresponsible not to bring them up today ;

- We advocate for different communication approaches when it comes to exposing more people to these risks ;

Here’s where there are still ongoing debates: - We don’t all agree on what those future risks are, or even if we should invest in trying to study and anticipate them given how much harm may be caused by these future systems and how uncertain these scenarios may be ;

- There are unknown risks linked with the development of more advanced AI.

- More investment is needed in some of the « less hyped » aspects of AI research: the fundamental math behind it, and any other way that can help us better understand and possibly control these systems ;

What is mostly consensual: - Harms exist today and must be addressed through regulation ;

As many experts are thinking about the best possible course of action, I would like to highlight some points that are mostly consensual, and others where there are still ongoing debates. My hope is that this will help everyone get more clarity on that very important topic.