• Strategy
  • Payers
  • AI
  • Cybersecurity
  • Device Development
  • Studies/Published Research
  • Ethics
  • Regulatory
  • Digital Biomarkers
  • Data Science
  • Companion Diagnostics
  • Clinical Care
  • Digital Therapeutics
  • Academic Research
  • Virtual Reality
  • RWE
  • Digital Endpoints
  • Medical Devices (Class II and Class III)
  • Commercialization
  • Clinical Trials
  • Drug-Device Combinations
  • SaMD
  • Remote Patient Monitoring
  • Patient Engagement
  • Profiles/Q&A

Protecting Data, AI, and Devices Appropriately


The link between artificial intelligence and human capital warrant legal protection for the latter

Inputted data, queries that educate that data, and the ensuing unique output—they are the sum and substance of artificial intelligence (AI). And while it may be a machine that is doing the computing, it was likely human capital that helped decide, create, and ultimately produce that data. If so, that human capital requires legal protection.

This legal protection, says Gunjan Agarwal, a partner at Fox Rothschild, during the 2022 Medtech MVP Conference held this past June, has to go beyond the standard intellectual property (IP) protections. The human capital invested in each step of the AI product are together responsible for the end result, so each must be separately protected.

Argawal explains that data drives AI; data are constantly changing; and AI- and IP- protection and ownership laws are always in flux. As such, a comprehensive strategy to identify the valuable assets that require legal protection should be developed—whether those assets are data models, output, or the services these assets create.

“If it's just a machine doing something, it's better kept as a trade secret and protected using contracts,” he says. However, if it is possible to include and clarify the human involvement that went into the system’s design, start to finish, then the property could be eligible for patent or copyright protection. Based on those clarifications, “identify the available or the best IP protections, whether it be copyrights, patents, or trade secrets, and wherever you see the gaps, fill it in with contracts,” suggests Argawal.

Typically, IP issues concern business objectives, he said. With AI protection, it's slightly different.

When AI-created data are designed as a product or a service, protection is tantamount, says Agarwal. If another group has created something similar and already has IP protection, he says that third party “can stop you from selling your own products or services that you might have invested millions of dollars in developing.” IP protection builds market value, and if protected properly gives the creator exclusive rights to that AI, he continues. Additionally, this applies to investors, lenders or potential acquirers of a company, whom all want to that IP is protected.

Or, looked at from another angle: Argawal notes that “If you have your own IP built around your tools or data, it can be a cross-licensing opportunity.”

Procrastinating to secure protection for any AI-based IP property is probably not a good idea; 40% of all applications made to the US Patent and Trade Office last year contained AI components.

Securing trade secrets

Trade secret protection is important to acquire, Agarwal said, because there are no government-approval or registration requirements. And, considering the “iterative nature” of AI and data, the AI model can change continuously because “your trade data is continuously expanding.”

Trade secret protection is like an umbrella. “Things that do not qualify for other types of IP protection, can be protected as a trade secret.”

Applications of AI

Agarwal gave specific examples of the use of AI, and one involved radiology images. Before the advent of AI, radiologists ran screens for various biomarkers; with AI, the computer system assumes that role. But the system first has to have the data. In the case of radiology, the radiologists have spent “many hours labeling different parts of the [images] and that is what makes it important input.”

As for the queries devised to create the algorithm, yes, they are very important from an IP protection perspective, he says. And finally, the output that’s been inferred with a human-trained machine learning model—that inference alone, he notes, is one of the most critical aspects of data creation.

When devising IP protection strategies, Agarwal repeats to his audience, companies should stress all human intervention, because IP protection is “not available for something that only a machine does.”

Six points to consider for legal protection

Here are some important points to consider when legal protection for big data and AI is required:

  1. Always draft contracts to protect yourself. While doing so, think about who in your firm or practice is providing the component you're trying to protect.
  2. If your company is hiring an AI specialist to create a model, make sure the agreement contains specific language about the person’s current assignment, and not future assignments. Case law exists, Agarwal said, which found agreements that try to protect assignments for future work do not “count as an actual assignment.”
  3. Those with valuable data who are not engaged in creating AI models should avoid giving exclusive use of it to one company.
  4. AI is more than just machine learning. Different algorithms apply in different situations. And the IP protection may depend on what type of AI you're using.
  5. If there is only one way to find what you are mining the data for, include that restriction in your contract. Agarwal used the example of analyzing data for cancer biomarkers. If another group is also looking for the same markers, chances are good that its members are using a similar method.
  6. As for output with respect to No. 5, the patent office will not allow too broad a claim to award a patent.
© 2024 MJH Life Sciences

All rights reserved.