Artificial intelligence is rapidly transforming industries by enabling new and previously unimaginable capabilities and efficiencies. As companies race to integrate AI into products and services, strategic approaches to protecting and licensing AI innovations are pivotal.
Building a Portfolio
The general approach to protecting AI technology remains largely the same as for other software and computer-implemented inventions. An effective and efficient intellectual property program for AI requires effort and attention.
Identifying valuable IP demands attentiveness, a company culture that values innovation, and legal engagement to protect valuable discoveries. This is why building sophisticated and mature IP programs is so hard.
AI technology hasn't changed those needs but has heightened them. Understanding the value and importance of new developments is the first step in protecting innovations, no matter their form.
A vigorous approach to identifying important AI developments, however, requires companies to investigate the development cycle. A new harvesting framework isn't required. Instead, existing processes can be expanded to cover AI development workflows.
New information or innovations can be found throughout the AI lifecycle, beginning with data collection and curating, then in model training and testing, and finally in deployment and integration. Protecting AI innovations requires identifying key developments in each of these phases.
Although the general IP approach may be the same, AI innovations pose some unique challenges.
For example, some valuable aspects of AI don't lend themselves easily to traditional utility patent protection. Important information such as training data, negative know-how, and labeling and annotations don't, on their face, appear patentable. But companies can leverage other forms of IP protection, including trade secrets and copyrights, to safeguard valuable non-public information. Building a robust AI IP portfolio requires a holistic approach to IP protection, not a patent-only strategy.
To determine the best form of IP protection, consider both the nature of the asset and the form that the commercial product or service will take. A patent might be the best option when the AI innovation improves conventional computer technology or a specific, practical application of AI software—or when it's integrated into a consumer product that can be easily reverse-engineered.
In the AI game, data is the queen of the board. Data often has significant value apart from any product. Curation, labeling, or annotation transforms ordinary data into a valuable commodity.
Collections of data and the know-how to store, manage, and deploy them aren't likely patentable themselves, but can be protected by trade secret. Opportunities to implement trade secret protections are especially important for software-as-a-service companies that provide AI-based SaaS platforms in which the key AI innovations remain hidden from outsiders. However, if a dataset is made publicly available, the best approach uses robust contractual restrictions in license agreements.
Licensing AI Technologies
Whether commercializing AI technology or licensing a tool, licensors and licensees must find common ground on contract terms that promote their business objectives while managing risks.
A license covering AI technology should specify restrictions on the licensee's use of the technology or its output, as well as any limitations on how the licensee's existing data can be used in the AI system.
A deal will only make commercial sense for both parties if it allows the licensee to use the technology for the licensee's intended purpose while maintaining restrictions the licensor deems necessary to support its business model and allow for compliance with its third-party contractual and regulatory obligations.
In AI licenses, terms regarding data ownership, usage rights, and restrictions are important. The licensor might seek to use the licensee's data to train and improve the AI system. Valuable information for this purpose includes the data generated by or for a licensee from an AI application or data the licensee feeds into an AI program. However, the licensee's goals may pose a conflict, as a licensee will typically want to ensure that its trade secrets or other confidential information fed into an AI platform don't become public knowledge.
One solution is using a tiered cost model, allowing a licensee to pay more to prevent the licensor from using data to train and improve its AI platform. To provide more security for licensee data, parties can agree to install the AI software on the licensee's own server rather than provision via a SaaS model. This would allow the licensee to use its data to train the AI system but prevent that improved system from being licensed to other companies (and potential competitors) down the road.
From the licensor perspective, agreements should ensure that any rights the licensor receives to licensee data survive after the agreement ends. This may require deviating from standard license terms requiring each party to return or destroy the other party's confidential information upon termination.
Mitigating Risk
A primary goal of IP portfolios is to mitigate the risks inherent with technology development. This threat could come in the form of competitor risk (both indirect or direct players in the market), collaboration risk (third-party vendors or manufacturers), or internal risks (employees intentionally or inadvertently disclosing non-public information).
An IP strategy must include pragmatic pillars and strong IP rights. Contract provisions that indemnify the business from IP risks and clearly define ownership of background and foreground IP are a must.
But contracts are often a last line of defense. Thoughtful IP programs conduct IP diligence up front to identify potential issues and guide the contract-drafting process. For AI, this is especially important. Understanding the terms of service of AI tools is paramount to securing a company's data and not accidentally giving it away.
Companies with AI technology should build IP portfolios strategically, keeping in mind principles governing IP in software—along with commercial realities—and implementing a plan to mitigate risk. When in- or out-licensing AI innovations, companies should carefully negotiate terms governing use of the AI applications and data.
Co-authored by David Soucy
Originally published by Bloomberg Law
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.