Evaluating Stability of Unreflective Alignment

James Lucassen, Mark Henry, Philippa Wright +1

Many theoretical obstacles to AI alignment are consequences of reflective stability - the problem of designing alignment mechanisms that the AI would not disable if given the optio…