AI History Battle

games

Design the auction

It is 2002 in Mountain View, and a young search company must sell advertising slots beside its results — millions of tiny auctions a day, run by algorithm, against bidders who are themselves algorithms. Design the auction: sell ad slots to strategic bidders so that truth-telling is each bidder's best strategy and revenue is predictable. Naive designs invite shading, sniping, and bid-cycling wars that destroy revenue and trust alike; the theory of mechanism design says incentives can be engineered, if the rules are chosen with mathematical care. The stakes compound absurdly: this auction will run billions of times daily and fund most of the consumer internet. A subtle incentive flaw here is not a bug — it is a permanent tax on the entire web.

mechanism designincentives

Who this problem belongs to

The two figures whose methods fit it best, out of 51 in contention.

1928–2015 · midcentury
93

This problem is analyzed, top to bottom, in concepts Nash created. His 1950 thesis defined equilibrium for exactly this situation — many self-interested players, no cooperation assumed — and proved existence; his bargaining solution and the Nash program tied cooperative outcomes to non-cooperative play. Mechanism design is the engineering inverse of his theory: choose the game so that its Nash equilibria (ideally, dominant strategies) do what you want, and the canonical analysis of the generalized second-price auction proceeds by characterizing its Nash equilibria and comparing revenue with truthful VCG outcomes. The honest caveats: Nash predates Vickrey (1961) and Myerson (1981), never designed a market himself, and his RAND-era experience left him skeptical of naive rationality assumptions. But no carrier's actual mathematics is more load-bearing here. The auction's stability is defined in his terms.

b. 1971 · deep-modern
88

Kleinberg stands at the precise intersection this problem occupies: web-scale algorithms meeting economic incentives. His 1998 HITS algorithm shaped link analysis at the birth of modern search; his small-world results treated emergent structure among self-interested actors; and he became a central figure in algorithmic game theory as it crystallized around 2000, later co-authoring with Tardos the standard textbook treating matching markets, VCG, and sponsored-search auctions explicitly. In 2002 he is an active, working expert on exactly the web infrastructure in question, fluent in both the approximation-algorithm mindset the billions-per-day scale demands and the equilibrium analysis truthfulness requires. He is not a classical auction theorist — Myerson's optimal-auction machinery is economics he applies rather than originated — which is the only daylight between him and the top score.

In the mind map

The same ideas, as concepts rather than history — in John's ML knowledge map.

Game Theory Reinforcement Learning

51 figures are scored on this problem. Draw it in a battle to see where you land.