Demis Hassabis, the CEO of Google DeepMind, is sounding the alarm about AI growing too fast without enough safety checks.
On the Masters of Scale podcast, he compared unchecked AI progress to an "oil spill," hard to contain once it happens, and urged tech companies to invest more in monitoring and keeping humans in the loop.
Hassabis warns AlphaZero could escape control
Hassabis pointed out that some AIs, like DeepMind's AlphaZero, can already teach themselves new skills at lightning speed.
He warned that if these systems start improving on their own without strong oversight, they could slip out of human control.
As he put it, "You don't quite know how to control, keep control of such a system."