April 4, 2023

A Disturbing Conversation About AI

In the past six months, publicly-available artificial intelligence models have advanced from an interesting toy to, with the launch of ChatGPT-4, something altogether different. The new version of OpenAI's language model can write reasonably good code, pass standardized tests like the LSAT with flying colours, and understand subtle jokes—things that machines have never done.

As its capabilities have advanced, a growing number of people have expressed concern that AI could pose a threat to people—not just by wiping out jobs or being used to spread disinformation, but by acquiring goals of their own, and pursuing them at our expense.

Our guest today, AI safety researcher Jérémie Harris, has advised top security officials in the US and Canada on these risks, and he joins us today to explain why he is concerned about where artificial intelligence is heading, and what it means for us.