Systems can learn from shifting their own expectations before final outcomes arrive

Evidence

And so maybe we don’t need to wait until we get the eventual ground truth in order to learn something. Maybe we can learn from these fluctuations themselves. Any time our expectation changes can be treated as an error in our previous estimate, and, hence, an opportunity to learn: not from the ultimate truth, which has yet to arrive, but from the new estimate, made by our very slightly older and wiser self. As Sutton puts it: “We are learning a guess from a guess.” — Brian Christian, The Alignment Problem, 600

Categorization

• corroborated: 2 independent passages support this claim — Brian Christian, The Alignment Problem, 600
• XREF: connects to: Information can simulate and control the objects it moves ahead of. — James W. Carey, Communication as Culture- Essays on…, 819; Learning reduces uncertainty that is what motivated deference to humans — Brian Christian, The Alignment Problem, 1294; Deliberate action requires partial but not total foresight of consequences — Georg Simmel, on individuality and social forms_…, 785 (+1 more)

Broader Connections

• Deliberate action requires partial but not total foresight of consequences #expands (0.80) — Georg Simmel, on individuality and social forms_…, 785