I have been exploring this question for quite some time now and recently published a preprint proposing a nested VSM architecture for the purpose. Coming from a systems/cybernetics background, honestly, it received a traction I haven't expected, so I pulled together the courage to share it with the "hardcore alignment...
I have been working on a control-theoretic approach to inner alignment, specifically looking at how biological systems handle safety. They rarely use hard constraints like "never do X" because those are brittle. Instead, they seem to use coherence as a depleted resource. I am trying to map this to AI...
Summary Current alignment techniques (RLHF, anti-scheming training) often rely on post-hoc reward shaping, which can be brittle under distributional shift. As a Systems Engineer, I propose a structural alternative: ConsciOS, a nested control architecture based on Stafford Beer’s Viable System Model (VSM) and Active Inference. The Architecture Instead of a...