Comment by Rome Viharo (rome-viharo-1) on Why humans won't control superhuman AIs. · 2024-10-16T17:33:22.280Z · LW · GW

I too believe symbiotic self-alignment is the only possible approach. The same system that can bring out the best of AI aligned to humans must also be the same system that aligns humans to some commonly shared principle. I believe multi-agent game theory, human to human, ai to human, ai to ai, as a system has some strong potential to address this issue.