Deep Blue vs Kasparov

Context: FIT1061_MOC · the W2 milestone and Kialo motion — the first machine win judged by result, argued over by method · the case that stresses Turing Test (Imitation Game) and The AI Effect (Defining AI) What it weighs: whether a victory counts as intelligence because of what it achieved or because of how it was produced.

Quick Revision

  • 🎯 Test question: When an AI wins, does it matter how? — the word reflective response, and the Kialo motion “Deep Blue’s win tells us nothing about intelligence.”
  • ⚠️ Key Constraint: a verdict without the machinery named earns nothing ➔ say brute-force tree search over positions/s with a hand-tuned evaluator, then judge that, not “a computer”.

📝 Core Commitments

  • The facts you argue from ➔ 1997-05-11, Game 6: Kasparov resigns after moves, match lost . IBM, over a decade, \approx\1030+480\approx2\times10^8$ positions/s.
  • The mechanism ➔ no intuition, no experience, no learning ➔ it searched (Search Problem Formulation), evaluated the leaves, picked the best reachable outcome. Same skeleton as Uninformed Search (BFS and DFS), plus a heuristic evaluator and pruning.
  • Behavioural rule ➔ judge only the observable output; the machinery is off-limits by design — Turing’s whole move was to replace “can it think” with a performance test.
  • Mechanist rule ➔ judge the process; a result produced by exhaustive enumeration is a statement about hardware and search, not about understanding.
  • What each refuses to count ➔ the behavioural rule refuses “but it’s only search” as evidence; the mechanist rule refuses the scoreboard as evidence.
  • The deflation trap ➔ the moment chess fell, chess stopped counting as intelligence (The AI Effect (Defining AI)) ➔ any purely mechanist verdict must answer why the goalpost moved after the result.
  • Hsu’s framing“The computer doesn’t play chess like a human, any more than an airplane flies like a bird.” ➔ concedes the mechanism differs and denies that this diminishes the achievement.

⚖️ Framework Contrast

PositionTest applied to Deep BlueVerdict it reachesWhere it breaks down
Behavioural (Turing’s substitution)did it produce grandmaster-beating play under fair conditions?intelligent — the output is what was ever testablelicenses a lookup table of every position as “intelligent”
Mechanist (“only search”)what machinery produced the moves?not intelligence — enumeration plus an evaluatorthe same objection dissolves human cognition into neurochemistry
AI-effect deflationis the task still hard now?not intelligence — chess is solved-ish, so it never countedunfalsifiable: no achievement can ever survive the test

🧩 Case Application Drill

⚠️ Common Mistakes

  • 💡 Retelling the match ➔ the dates, the moves and the accusation are evidence, not an argument; every fact cited must be attached to a claim about intelligence.
  • 💡 Asserting a verdict with no rival named ➔ the response is graded on engaging the counter-position; “it was just brute force” without the Turing objection answered is half an argument.
  • 💡 Sliding the goalposts silently ➔ if the verdict is “chess never counted”, say so and defend it against The AI Effect (Defining AI) — otherwise the position is unfalsifiable.