Competition Policy for Invisible Agents

The title of today's conference is The Invisible Mind, which is wonderfully evocative.

To me it sounds like either a lost Hitchcock film or the name of a popular philosophy podcast.

Since Adam Smith, economists have talked about invisible hands, and today I want to talk about invisible agents.

An invisible agent is an AI system that acts for someone in a market.

It might search, compare, recommend or buy.

It might set prices, design bids, screen suppliers and negotiate contracts.

It may work for a customer, a firm, a platform or a payment provider.

Often it acts at machine speed across many transactions, with the human setting broad instructions rather than watching each step.

Economics has always cared about agency.

We ask who has information, who has the incentives, who bears risk and who captures the surplus.

But AI agents put a kind of mysterious intermediary into the mix.

When a person walks into a shop, we know what roles they're playing.

The buyer wants a good product at a good price.

The seller wants a sale.

The shop assistant may be helpful, though maybe a bit overenthusiastic about selling you the extended warranty if they also get a commission for doing so.

But in an AI-mediated market, the picture is a little murkier.

A shopping agent may say it's helping the consumer, yet it might be trained by a platform, paid by a merchant, optimised for engagement, or tuned to protect a particular market structure.