I’ve been sitting on these ideas for a while now, turning them over while watching AI development accelerate in directions that honestly concern me. I’m an independent researcher and technologist who’s been deeply embedded in AI-adjacent work for years, and sometimes I think that outside perspective is exactly what’s missing from alignment discourse. The two ideas I want to put forward are pretty straightforward at their core: first, that values in AI systems should be developed the way a mentor develops them in a student through scenarios, reasoning, and graduated trust rather than bolted on as constraints. Second, that no single AI system should ever be the sole ethical decision-maker on anything consequential, and that a structured board of diverse models acting as a conscience layer could be a meaningful architectural safeguard. I’m sure there are gaps and I’m genuinely here to have them pointed out that’s the whole point of posting this. If you’re interested and want to read the full paper, let me know and I’ll post it.
Novel Alignment Proposal
I’ve been sitting on these ideas for a while now, turning them over while watching AI development accelerate in directions that honestly concern me. I’m an independent researcher and technologist who’s been deeply embedded in AI-adjacent work for years, and sometimes I think that outside perspective is exactly what’s missing from alignment discourse. The two ideas I want to put forward are pretty straightforward at their core: first, that values in AI systems should be developed the way a mentor develops them in a student through scenarios, reasoning, and graduated trust rather than bolted on as constraints. Second, that no single AI system should ever be the sole ethical decision-maker on anything consequential, and that a structured board of diverse models acting as a conscience layer could be a meaningful architectural safeguard. I’m sure there are gaps and I’m genuinely here to have them pointed out that’s the whole point of posting this. If you’re interested and want to read the full paper, let me know and I’ll post it.