Tech Transfer eNews Blog

Don’t undervalue or under-protect your data for use in AI systems

By Jesse Schwartz
Published: September 27th, 2023

A detailed article on IP protection and licensing issues surrounding data used in AI systems appears in the September issue of Technology Transfer Tactics. To subscribe and access the complete article, or for further subscription details, click here.

Artificial intelligence is making its way into virtually every industry, and tech transfer programs are finding opportunities to work with AI both as developers and as owners of data and other IP that might be joined with AI systems to make something new. But this exciting new frontier brings with it questions and challenges about how to manage co-developed innovations and address all the facets of licensing and commercialization that are more familiar in other areas.

The first issue to consider is ownership, says Nicholas J. Zepnick, JD, partner with the Foley & Lardner law firm in Milwaukee, WI. He says he has had sophisticated clients who failed to paper out the relationship in a deal because it involved AI and not a commodity with which they were more familiar.

Zepnick offers the hypothetical of a university that has data of interest to an AI developer that wants to use it to train its AI model. “The institution says, alright, we’re going to get paid X dollars for use of our data. They let the other party use it, and then they just take their check and go home,” he says. “But often, that fails to value the data.”

Classifying the data can be a big part of the process and adds value to the data for licensing purposes, Zepnick says. If the data is a large set of photos, the AI may learn from classifying those photos, for example. Once that is done, the AI model has been improved and is more commercially viable.

“Who owns that resulting AI model? Who owns the future use of that? What I see a lot of clients doing is taking the immediate win, which is in my experience a nominal payment for the use of the data, and failing to see the potential value of the AI model [once it] has been trained, and the revenue that will come out of that,” Zepnick says. “A lot of times, these kind of companies that are developing the AI tool are getting a pretty big windfall out of the whole deal, to be honest.”

To the university or other owner of the data, the information contained in an AI system might seem mundane or even trivial, Zepnick says, but if the AI developer wants to train its system using that data, it most certainly has real value.

“If you’re trying to teach an AI system to identify taillights and have the car brake, and I’ve got a million pictures of brakes lights, that has value to you because otherwise you would need to take pictures of all these cars,” he says. And in many cases the data could never be reproduced, making it even more valuable.

“If I have health history data from a gazillion patients who’ve taken combinations of drugs x, y, and z, it may not be possible for you to get that data in another way. You might not be able to train your model in the same way simply because I have the collection of all of this stuff, and you don’t.”

Click here to continue reading this article with a subscription to Technology Transfer Tactics. Already a subscriber? Click here to log in.

Posted under: Tech Transfer e-News