AI applications routinely perform a myriad of operations on and off-line.  Some are fairly mundane, ranging from running e-commerce database queries, preventing financial fraud and deep-space data analysis.  More sophisticated AI operate autonomous vehicles; play chess (Deep Blue), defeat the best human contenders in Jeopardy (Watson), take action in response to verbal commands (Siri) and some come very close to passing off as humans, solving CAPTCHAs (though not, as of this writing, the more current variety known as PlayThru).  Then there are (as yet theoretical) applications that aim at legal informatics.  These include AI that provide advice on website terms of use and on used-car warranties.

The range of operational possibilities is essentially limited only by the imagination and the current state of the art.  And with hardware and software sophistication growing exponentially, and predicted to “vastly” exceed that of humans, AI implementation and widespread adoption is not only here to stay but is destined to rapidly increase and take on operations that are currently difficult to fathom.

This post considers a discrete question as a starting point for a broader analysis:  Can an AI agent effect a valid transfer of an exclusive copyright? Let’s get the answer right up front: No.

For a valid transfer to occur, 17 U.S.C.A § 204(a) requires that it be (i) in writing and (ii) signed by copyright owner or their “duly authorized agent”. Under the current legal environment in the US an AI agent could satisfy (i.e., produce) the first prong, but would fall short of satisfying the second.

This is because AI agents are not considered “agents” in the context of the law of agency. (For more on this, see the 4-part Germinating Seeds of Agency posts.) UETA and (to a drastically lesser degree) UCITA determine the “software agent pedigree”, labeling these agents as merely transactional software programs. For purposes of satisfying §204 under current law, the AI agent would thus need to be (i) duly authorized and (ii) be accepted as an agent under agency law principles.

It is tempting to immediately engage in an analytical framework that seeks to satisfy these two prongs.  But amending the law is unnecessary and undesirable (from a time perspective) and I think there is a more practical alternative that can deliver the desired result of allowing AI agents to effect a legally-valid copyright transfer.

The solution comes from understanding and satisfying the core legal concerns.  Section 204 was designed to ensure that (exclusive) copyrights are effectively transferred when there is a clear, deliberate act on the part of the copyright holder. The writing does not have to be in paper (ESIGN), nor a lengthy document.  However, §204 surpasses the statue of frauds’ mere evidentiary-centric purpose, requiring the written agreement to go a little further and sufficiently evince a properly deliberated grant took place. (This is a pretty rigorous requirement that even appears to override customary industry practices to the contrary, see e.g., Effects Associates, Inc. v. Cohen 908 F.2d 555 C.A. 9 (Cal.),1990.)  Bottom line, §204 should be satisfied so long as there is sufficient evidence that a deliberated grant took place.

Given AI’s sophisticated transactional and decision-making capabilities, and that a party using it is conspicuously signaling their intent to be bound to the AI’s decisions, a grant by the AI can be properly understood as being the result of a deliberated grant effected by a “duly appointed” tool selected by the copyright owner.  The fact that this tool is not an “agent” under agency law should be rendered irrelevant.  Such a grant satisfies the core concern of §204.  It should also satisfy the courts.

***Postscript***

December 14, 2020: In the eight years that have passed since the above was posted, we have seen significant attention to the question of whether AI entities should be granted patent rights. I’m of the opinion, which I’ve written about here and have offered in a number of interviews and CLEs is that that the answer is, flat out: “no.” Public preoccupation with this issue may be (possibly, but even then unlikely) relevant when we not only reach into, but mature into, a state of Artificial General Intelligence (AGI). While an exact timeline is a matter of some disagreement, the common denominator is that AGI is not going to happen for a long time. Now, with this AI patent rights discussion backdrop, the question raised in the above 2012 post gets a bit more color. A very small, incremental step towards endowing AI with meaningful rights could begin with satisfying its status vis-a-vis §204.