The urgent need for AI safety: three videos

Today I would like to recommend three videos highlighting the importance of AI safety from various perspectives.

First, Yoshua Bengio. He is a professor at the Université de Montréal and widely held as one of the world's two or three most respected AI researchers. Yesterday (September 2), he spoke at the Royal Swedish Acadamy of Engineering Sciences (IVA). Since early 2023, Bengio has been outspoken about the urgent need to address existential AI risk and AI safety, and this was also the focus of his talk yesterday:

The talk is nontechnical, very clearly laid out and quite crisp: it begins about 13:20 into the unedited video and goes on until about 36:10. After that follows a long and fairly enlightening discussion with Fredrik Heintz, who is a bit of a key player in the Swedish AI ecosystem, being a professor at Linköping University, a long-time preident of the Swedish AI Society, and a member of the AI commission launched by the Swedish government in December last year. I've had a number of interactions with Fredrik over the last few years, in media and elsewhere, and on these occasions he never came across as particularly interested in the need to save humanity from AI catastrophe. This time, however, he engaged so seriously with what Bengio had to say about the topic that I take it as a highly welcome shift in his position towards a better appreciation of AI safety concerns. Well done, Fredrik!

While Bengio's talk works well as a first introduction for a beginner to the fields of AI risk and AI safety, I feel that an even better such introduction may be Robert Miles' recent video AI ruined my year. Unlike Bengio, Miles is not primarily an AI researcher but a very skilled communicator and popularizer of some of the field's key ideas. The video is a summary of the past year's dramatic unfolding of some key AI events, and a touchingly personal recollection of how these have forced him into some pretty deep soul searching:¹

Finally, here's a third video - please bear with me, because it's just 11 seconds long - where a famous clip with Gary Oldman in the movie Léon is efficiently exploited in order to make a key point to US presidential candidate Kamala Harris:

Footnote

1) These are the key qualities for which I recommend Miles' video. The fact that my name is visible in it for a split second plays little or no role in this.