Large language models, trained on personal data, may soon be able to mimic individual personalities. This would potentially transform search across human candidates, including for marriage and jobs -- indeed, several dating platforms have already begun experimenting with training "AI clones" to represent users. This paper presents a theoretical framework to study the tradeoff between the substantially expanded search capacity of AI clones and their imperfect representation of humans. Individuals are modeled as points in $k$-dimensional Euclidean space, and their AI clones are modeled as noisy approximations. I compare two search regimes: an "in-person regime" -- where each person randomly meets some number of individuals and matches to the most compatible among them -- against an "AI representation regime" -- in which individuals match to the person whose AI clone is most compatible with their AI clone. I show that a finite number of in-person encounters exceeds the expected payoff from search over infinite AI clones. Moreover, when the dimensionality of personality is large, simply meeting two people in person produces a higher expected match quality than entrusting the process to an AI platform, regardless of the size of its candidate pool.
翻译:暂无翻译