Q&A: The legalities of intellectual property with AI-based solutions

Artificial intelligence and large language model use have grown within healthcare, and questions arise about who can claim ownership of the intellectual property associated with solutions developed utilizing these emerging technologies. 

Dr. Terri Shieh-Newton, intellectual property attorney, immunologist and member of the law firm Mintz, established the Life Sciences Artificial Intelligence group at the firm. The group comprises a team of AI-focused IP practitioners, including microbiologists, physicists, immunologists, chemists, electrical engineers and computer scientists, who meet monthly to discuss current patents or concepts related to new AI models.

Shieh-Newton sat down with MobiHealthNews to discuss the legal viewpoint of establishing intellectual property when utilizing AI and what companies should be aware of when determining ownership. 

MobiHealthNews: What advice do you give clients when discussing AI use in healthcare and ensuring the use of unbiased data?

Dr. Terri Shieh-Newton: As an IP attorney, frankly, I don’t really get into the whole design of the database. That’s more on the data scientists. But what I will do in working with them, and as they’re telling me the results, is I might ask probing questions about where certain things came from and how did you assemble this data set? And how did you train this? And what were some of your exclusion parameters? Sometimes when they’re trying to group different things, they’ll use classifiers. And so, what kind of classifiers did you use? Because, you have all these data points, and how you draw the line will kind of delineate that. And so I think it’s really kind of up to me to ask those probing questions. 

Although I may not have designed it, I may shed light on certain things because they may be thinking one way. And from my angle, what I’m trying to do is to get a reasonable, strong patent, and in order to do that, and especially in light of the recent Amgen v. Sanofi decision, it is really incumbent upon all of us to think about what is it that you’re claiming? You’re claiming this huge scope, but yet you just have a few data points, and I think that’s the issue, right? I mean, the Supreme Court, the Federal Circuit, they really pointed that out. However, with machine learning and some of the datasets, are you able to actually sample more, so you’re actually more enabled and have that written description there, support there that wasn’t there before? And I think that’s where we’re coming in with some of the questions about where are you getting your dataset? Is it skewed toward a certain way? Are you actually eliminating a population or some criteria that actually would help you strengthen the breadth of your patent?

MHN: Can you discuss how companies may identify ownership of intellectual property as AI begins developing solutions?

Shieh-Newton: That’s an uncharted area in terms of there’s no subtle law on that yet. But I think what it comes down to is, who is the one who put the algorithm together? Who’s the one who’s doing the data training? What kind of model is it? You know, if it’s a supervised learning model, then there’s some thought process. If you start out with junk data, then you’re probably going to get junk results. So there is some thought process there as to how one is curating it.  

And then there might be different modules that get separated out. So there is something there where there’s a deliberate attempt to maybe divert the workflow or the calculations one way or the other. And ultimately, that might be the person that ends up being the inventor because we know right now the AI can’t be an inventor. So then that begs the question of how well are all the protections in terms of the appropriate employment agreements? Or [are] the inventorship assignment agreements already in place so that if there’s a dispute because we know case law is changing all the time. So, what is happening right now may not be reflective of what really is going to happen a year from now. I mean we don’t have the same system like in Europe or some of the other jurisdictions where the company automatically owns all the data and everything.

MHN: It could get murky for companies that don’t have those contracts in place.

Shieh-Newton: I think that’s pretty standard with employment contracts. What’s the messier issue is who owns the data. Because I think these days, there’s a lot of collaboration, and there’s a lot of data being exchanged.

I think it’s a general principle, right? The more data you have, the better the training set you have. I mean, if you have 10 data points versus 10,000 data points, you’re able to get much better training. But then, where did that come from?  

In academics, depending on who it is, I guess some of them are very sophisticated, but others are just more free. And they just want to exchange information. 

The trickier scenario is that different institutions or different companies are collaborating. And then it’s super hard to track where did that data come from, and then if that came from a hospital, was there some sort of release? Is there some sort of HIPAA concern? So I think those are the things that don’t fall squarely under patents but are part of the overall workflow that we do have to take into account because I think one of the worst-case scenarios is you have all this data, and you actually do come up with a great discovery and then somebody comes knocking along saying, well, but for that data that I gave, you wouldn’t have discovered this great thing and so, therefore, I deserve a piece of that.

But I think there are ways of progressing, and everybody has good intentions, and everybody wants to help advance medicine, advance cures and things like that. But it just takes advanced planning and the right agreements in place. And then it’s like, OK, everything’s settled. Now, everybody go share and make good progress.