Coase, Property Rights, and the AI Debate Everyone Is Having Backwards
Jeffrey Wernick
The AI debate has settled into two camps. One says the technology is too dangerous to distribute, so it must be concentrated in a few licensed hands. The other says it is too dangerous to concentrate, so it must be distributed widely. Serious people occupy both camps, and both are answering the wrong question. The question is not who should hold the power. The question is what rights everyone else holds against whoever ends up holding it. Neither camp asks, because both camps are populated by people who build, fund, or deploy the technology, and the missing rights belong to everyone else.
Take the concentration camp first, because its path is the most traveled and its destination is known. Stigler, 1971: “A central thesis of this paper is that, as a rule, regulation is acquired by the industry and is designed and operated primarily for its benefit.” There is no version of AI safety regulation that escapes this. The incumbents write the thresholds. The thresholds price out entrants. The safety agency becomes the industry’s moat, and the moat gets defended as protection. The end state is already visible: a public private partnership for warrantless surveillance, in which the laboratories receive liability protection and procurement, and the state receives the data and the chokepoint. The third party doctrine already blesses it. Records handed to an intermediary lose their Fourth Amendment protection, and the most intimate record ever assembled, your conversations with the machine, sits with intermediaries.
The distribution camp wins the argument against concentration. Zuckerberg is right that the notion of safety through extreme concentration of power is inherently problematic, and history is on his side. But distribution alone answers the harm question with a shrug, and the shrug is not necessary. The harms are unpriced because the rights are undefined, and that is a fixable condition, not a property of the technology.
Coase, 1960: “It is necessary to know whether the damaging business is liable or not for damage caused since without the establishment of this initial delimitation of rights there can be no market transactions to transfer and recombine them.” The entire modern treatment of externalities rests on that sentence. Harms do not require a regulator. They require defined rights and workable transaction costs, and then bargaining and liability allocate the costs to whoever values them most. Coase does not fail for data. Coase has never been tried for data, because the precondition has never been met. There is no general property title in personal data. The delimitation Coase requires does not exist, so there is nothing to bargain over, no one to sue, and no price on the taking. The largest industry in history was built on an input whose price is zero because its owners were never given title.
Berners-Lee, who built the web: “Solid changes the current model where users have to hand over personal data to digital giants in exchange for perceived value. As we’ve all discovered, this hasn’t been in our best interests.” And: “You should have complete control of your data. It’s not oil. It’s not a commodity.” Personal data, he added, you should not be able to sell for money, “because it’s a right.”
Three reforms meet Coase’s precondition.
Define property rights in personal data. Title, not consent theater. A checkbox on a forty-page policy is not a conveyance, and behavioral data taken at a price of zero is taken, not traded.
Repeal the third party doctrine. Handing data to an intermediary should not forfeit it. The doctrine was built for bank records and dialed phone numbers, and it now covers everything a person thinks out loud to a machine. Property that dissolves on deposit is not property.
Abolish Section 230. The common law had the answer in Cubby v. CompuServe, 1991: “CompuServe has no more editorial control over such a publication than does a public library, book store, or newsstand,” liable when it knew or had reason to know, not otherwise. The bookstore standard, being refined case by case, the way liability rules have always been discovered. Congress interrupted that developing common law and substituted a sweeping statutory immunity, and the immunity built the platforms we have. Let the common law resume where it was interrupted.
Define the rights and the rest follows. Courts price harms case by case, the way every liability regime developed. The firm that wants your data buys it from you at a price you accept. The firm whose model damages you answers to you, not to an agency it staffed. No thresholds for incumbents to write. No central licensing authority for Stigler’s rule to capture.
Transaction costs are the strongest objection. Millions of small data holders cannot each negotiate with every counterparty. Markets solve this with intermediaries, standard terms, and class actions, once there is a right to aggregate. Copyright clears billions of micro-uses through collecting societies. Transaction costs are an engineering problem inside a rights regime. They are not a reason to hand the field to an agency that Stigler’s rule will convert into the industry’s instrument.
Property rights answer the question of what machines may do to people. A separate problem remains: how machines deal with one another when neither can safely trust the other. The design for that already exists. Satoshi, 2008: “If a greedy attacker is able to assemble more CPU power than all the honest nodes, he would have to choose between using it to defraud people by stealing back his payments, or using it to generate new coins. He ought to find it more profitable to play by the rules, such rules that favour him with more new coins than everyone else combined, than to undermine the system and the validity of his own wealth.” And the conclusion: “We have proposed a system for electronic transactions without relying on trust.” No certified participants. No trusted overseer. Defection priced, verification open to anyone, seventeen years in production. That is what safety looks like when it is built instead of promised.
Until the rights exist, caveat emptor is the only honest doctrine for the user. A certificate issued by the seller about the safety of the seller’s own machine is an advertisement, not a protection.
Hal Finney, who received the first Bitcoin transaction, described another tradition: “The computer can be used as a tool to liberate and protect people, rather than to control them.”
That is the choice the AI debate keeps avoiding.
The concentration camp would decide who may build the machines. The distribution camp would let everyone build them. Both begin one step too late. Before deciding who may possess the technology, decide what the technology may do to you. Define what is yours. Define what may be taken. Define who is liable when you are harmed. Preserve those rights when your property passes through an intermediary. Then let courts discover the boundaries, markets price the rights, and machines compete inside them.
The problem is not that we have failed to decide who should control AI.
The problem is that we are deciding who should control AI before deciding what they have no right to control.
Editor’s Note: At a moment when the once vaunted model of responsible journalism is overwhelmingly the play thing of self-serving billionaires and their corporate scribes, alternatives of integrity are desperately needed, and ScheerPost is one of them. Please support our independent journalism by contributing to our online donation platform, Network for Good, or send a check to our new PO Box. We can’t thank you enough, and promise to keep bringing you this kind of vital news.
You can also make a donation to our PayPal or subscribe to our Patreon.