Machines could learn human values by observing behavior instead of being programmed

Evidence

Researchers are increasingly making the argument that our approach to instilling human values in machines should take the same tack. Perhaps, rather than painstakingly trying to hand-code the things we care about, we should develop machines that simply observe human behavior and infer our values and desires from that. — Brian Christian, The Alignment Problem, 1086

Categorization

• XREF: Pairs with the book’s opening themes about hand-coding ethics and whistleblowing; also connects to the chess-game metaphor of unknown rules. — connects to: Ambiguity in inferred reward functions is usually harmless behaviorally — Brian Christian, The Alignment Problem, 1102; Ensuring AI captures human intent is now a central computer science problem — Brian Christian, The Alignment Problem, 65; AI researchers aim to align general-purpose systems with hard-to-specify human intentions (+8 more)
• corroborated: 2 independent passages support this claim — Brian Christian, The Alignment Problem, 1086

Related

• Ensuring AI captures human intent is now a central computer science problem (similarity 0.88)
• AI alignment may require machines to pursue human objectives, not their own (similarity 0.88)
• Rule-based machine ethics produce unintended emergent consequences (similarity 0.81)
• Ethical rules for robots require human specifications of their key variables (similarity 0.80)
• Overparameterized value models keep machines obedient at modest cost (similarity 0.80)
• Machine learning increasingly intertwines technical problems with human and social questions (similarity 0.80)
• AI researchers aim to align general-purpose systems with hard-to-specify human intentions (similarity 0.80)
• Intent-alignment doesn’t specify whose intentions matter, leaving trade-off conflicts unresolved (similarity 0.79)
• AI safety concerns have shifted from fringe to central research focus (similarity 0.78)
• Ambiguity in inferred reward functions is usually harmless behaviorally (similarity 0.78)

Broader Connections

• AI researchers aim to align general-purpose systems with hard-to-specify human intentions #expands (0.85)
• Ensuring AI captures human intent is now a central computer science problem #corroborates (0.75) — Brian Christian, The Alignment Problem, 65