The Alan Turing Institute Says We Are Losing Control of AI in 5 Years
The boffins at the Alan Turing Institute finally realized that AI might actually be a problem, not just a way to generate weird cat pictures. They’ve dropped a report suggesting we might be toast, so let’s talk about how to hit the off switch.
The Alan Turing Institute just released a report titled Frontier AI Risks: A practical way forward, arguing that we don't need to wait for the robots to start a world war to see that things are going sideways. Instead of obsessing over sci-fi scenarios, they are focusing on five immediate, boringly terrifying risks: cyberattacks, democratic destabilization via deepfakes, systems acting in ways we didn't ask for, the loss of human oversight, and the reduced barrier to creating biological weapons.
Testing models in a lab is apparently about as useful as a chocolate teapot because real-world conditions constantly change the behavior of autonomous systems. The report demands better verification and, essentially, babysitting for algorithms that are becoming too smart for their own good. The Institute plans to partner with the AI Security Institute and the NCSC to figure out how to stop AI from going rogue, backed by a 2 million pound grant from Coefficient Giving.
There is no magic "off switch" or simple way to just slow down development, because that would require global coordination—something the current state of international politics is famously terrible at. It seems the world is busy building a digital god and just hoping it doesn't decide to delete the users.
Source: Alan Turing Institute
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.