Why the US Government Already Banned Two AI Models | Dr. Roman Yampolskiy

Link to full episode: https://youtu.be/ebWFexw51qM?si=5W4y2WkHIqse7pie

The US government has already blocked two of Anthropic’s most advanced AI models from being deployed. Roman Yampolskiy says the restriction was temporary and did not reach every part of the government, and he says it happened. His account of why: the NSA and other agencies went to elected officials and told them these systems were going to hack the government and hack the nuclear response, and that the country could not allow it. Inside a short window the official position moved from “we will not interfere with US AI development” to “those models are no longer allowed to be deployed.”

Yampolskiy reads that reversal as the start of something bigger. He points to a letter signed by around 1,200 employees at top AI labs asking government to step in and slow them down, because the people building the systems say they cannot keep up with them. He points to an AI conference in China where the country’s president said governments need to retain control over these systems. Nobody’s interest is served by everyone dying, he argues, so this should be the easiest deal anyone has ever had to make.

Danny pushes back on whether the people writing the rules can understand what they are regulating. Yampolskiy is blunt about it. Elected politicians do not use computers and do not use AI, so technical advisers tell them how to vote, on Bitcoin and now on this. Danny adds the part that bothers him more, which is that those advisers usually have money riding on the answer.

From there the two get into surveillance, and whether China’s camera and social media control grid is arriving in the United States. Yampolskiy points at the phone in your pocket and at what Snowden already told everyone. The NSA has the capability. What he cannot say is how much gets collected and how much gets kept. He also throws cold water on the idea that a defense contractor is sitting on a superintelligence twenty years ahead of anything public, because the money and the compute are out in the open market now.

Roman Yampolskiy is an AI safety researcher and computer scientist who has spent years arguing that a superintelligent system cannot be controlled once it exists.

Subscribe to this channel: https://www.youtube.com/channel/UCQ_piSR8gm-TfHZcDWOJciA?sub_confirmation=1

Written by Danny Jones Clips