Welcome to the Technical AI Safety Podcast, the show where I interview computer scientists about their papers. This month I covered Optimal Policies Tend to Seek Power, which is closely related to Seeking Power is Often Robustly Instrumental in MDPs which is a part of the Reframing Impact sequence and was recently a part of the 2019 review.
The point of the show is to make papers more parsable, the interview features a detailed walkthrough padded on either side by discussion of where the work came from and where it's going.
I had a lot of fun doing this month's episode, a tricky paper to wrap my head around but very rewarding. Do let me know if you have trouble finding it on your favorite podcast app, thanks!
With Alex Turner
Optimal Policies Tend to Seek Power
by Alexander Matt Turner, Logan Smith, Rohin Shah, Andrew Critch, Prasad Tadepalli
Some researchers have speculated that capable reinforcement learning agents are often incentivized to seek resources and power in pursuit of their objectives. While seeking power in order to optimize a misspecified objective, agents might be incentivized to behave in undesirable ways, including rationally preventing deactivation and correction. Others have voiced skepticism: human power-seeking instincts seem idiosyncratic, and these urges need not be present in reinforcement learning agents. We formalize a notion of power within the context of Markov decision processes. With respect to a class of neutral reward function distributions, we provide sufficient conditions for when optimal policies tend to seek power over the environment.