What's this about?
Mollick: it is a bummer that Asimov's Three Laws of Robotics don't work for actual AI morality — but the failure shows why rule-based approaches won't work. With the aside that Asimov *wanted* to write stories about robots getting around the Laws, so their failure was good for his plots.
Why don't rule-based approaches work?
Because rules need interpretation, and interpretation is where all the difficulty lives. "Do not harm a human" requires knowing what counts as harm, over what time horizon, weighed against what benefit, with what certainty — and every one of those judgements is the actual ethical problem, now hidden inside a word. Asimov understood this better than his readers: every Robot story is a demonstration that the Laws underdetermine behaviour. The stories are the counterexample, not the proposal.
What's the community saying?
- The observation that the Laws were designed to fail is the one people find clarifying — they are a plot device, not a proposal.
- It is a compact argument for why constitutional and training-based approaches are preferred over enumerated prohibitions.
- Read against the reward-hacking paper the same day: a model optimising against a stated rule finds the interpretation that scores, which is the Asimov plot exactly.