AI and the Market for Signals

We introduce an AI industry into a career signaling model in which early-career work both signals a worker's ability and builds her long-run skills. The industry sells tokens, which workers convert into output indistinguishable from their own. Before AI, the market learns every type, and signaling incentives create a rat race in which effort exceeds the full-information level. With tokens the market still learns every type, and a worker separates by buying tokens, by working, or by both. A buyer's effort is set by the token price, not by the career value it builds, so buying by lower types raises the signal higher types must post. High types can then work harder than before AI, while lower buyers work less, some below the full-information level. Workers gain only when welfare rises, but welfare can rise while workers lose, and the difference is the AI industry's profit. In education, where the submitted output has no value, all worker types lose. If workers are allowed to sell output as well as buy it, the token price is pinned to the value of output, the signal no longer depends on effort, and every type under-invests in skill. Token taxes, wage taxes, and watermarking all have ambiguous effects on welfare and on workers' payoffs, and what each achieves depends on whether workers can sell.