Consider the following hypothetical: Researchers at a US-parented drug company develop an artificial intelligence (or “AI”) system that can identify new therapeutic targets with minimal human intervention. The drug company sells the system to its foreign affiliate in a lower-tax jurisdiction. What is the appropriate valuation of the system on this outbound transfer (e.g., based on the cost to create it or based on the value of the IP it is likely to generate)? And, when the AI system later successfully creates a new therapeutic, which entity will be entitled to the non-routine returns from sales of the therapeutic: the US parent that developed the system, the foreign subsidiary that owns the system that developed the therapeutic, or some combination of both?

As is generally the case in transfer pricing, the answer to these questions will of course depend on the facts and circumstances of the particular case.

These questions are further complicated because they hinge in large part on intellectual-property law, which is struggling to keep up with the advances in AI technology. Recently, the US Patent Office rejected an argument that the sole inventor of two inventions was an AI tool. Although the Patent Office determined that the inventor listed on the patent applications must be a human, controversy in this area will likely continue, and Congress might be forced to intervene in the future. If it were ever determined that the inventor could be the AI program, would this support an argument that the AI program should be treated for transfer pricing purposes as having performed the IP development (i.e., the “D” in the company’s DEMPE functions)? If so, the affiliate owning the AI system should presumably be entitled to a significant portion of the income from the invention.

In recent years, practical applications for AI have multiplied, and the hypothetical above no longer seems like something that comes from a science-fiction movie. AI systems will invariably blur the line between the inventor, the inventor’s toolkit, and the invention, making questions about IP ownership—and by extension transfer pricing—especially nettlesome. The existence and use of AI systems may also present significant tax planning opportunities.