Early-stage companies are innovating new artificial intelligence-based solutions, but they often face questions as to whether such technology can be protected and the best strategy for doing so.
Without an understanding of how to protect their R&D investment and claim technology as proprietary, startup companies are leaving a tool behind, possibly forfeiting market share and investments as a result.
The considerations below will be useful for companies trying to understand the opportunities to protect their innovation.
Artificial intelligence innovations are patentable
AI software is patentable, and applicants are seeking protection at a remarkable rate. In 2000, the U.S. Patent and Trademark Office (USPTO) had received about 10,000 applications directed to artificial intelligence, and by 2020, that number reached about 80,000 applications, of which 77% were approved.
Despite some challenges involving types of software and business methods that are ineligible for patent protection, we have been obtaining patent protection for AI innovations for many years. In fact, the USPTO even issued guidance for eligibility that gave an example of training a neural network. Patentable innovations may relate to an improvement in a particular model, an implementation of a model, improved training or other aspects.
Disclosure of technology, whether planned at a conference or a partner meeting, or unplanned and incidental, may cause a forfeiture of patent rights.
The USPTO characterizes AI innovation as including the following component technologies: planning/control, knowledge processing, speech, AI hardware, evolutionary computation, natural language processing, machine learning and vision.
If a company has an innovative feature that distinguishes itself from competitors, then patent protection may be a worthwhile tool to gain a competitive advantage, perhaps even in conjunction with copyright and trade secret protection.
Direct patent coverage to detectable features
Patents can be useful for excluding others from making, using, selling, or selling an infringing product or method. When asserting that another company infringes a patent, we look to the claims, which define the property right. If another company practices each and every element of a claim, then that company is infringing.
In an effort to more easily identify those infringing companies, perhaps through the use of their marketing materials or inherency in the functionality of their software, claims may be directed to the aspects of the innovation that are more easily detectable. For example, aspects of the algorithm regarding mathematics for an optimization algorithm may be difficult to detect.
While there are mechanisms for obtaining information from the accused party, a patent holder may be able to establish a stronger position if the claims read on the apparent features of the party’s product. Likewise, that infringing party will be more readily aware that their system is likely infringing upon reviewing those claims. When those competitors conduct clearance searches to identify potential risks of infringement, it would be beneficial to have patented claims that cause significant concern that a court may stop the company from using that product or make it pay significant damages.
Conventional contractual agreements may be obsolete
AI developers routinely enter agreements with third parties to access their data for training or deployment purposes. A third party may own certain IP rights protecting a training dataset, such as a trade secret or copyright. During training, the AI model updates its weights or hyperparameters using the training data, resulting in a trained variant of the original. The third party might be able to assert some ownership rights over the trained AI or its outputs, as it may arguably be the product of the third party’s training data.
The agreement should clearly delineate ownership and license boundaries between the training data, the AI model, the trained AI model and the output data. There are many variations on this theme — undefined rules governing the ownership of the data and the AI software can inject hidden vulnerabilities. The data itself may also be subject to certain privacy laws and each entity using, receiving or providing data should be aware of those regulations.
Conventional agreements directed to software or data are unlikely to clearly delineate these boundaries and should be scrutinized by AI developers or data owners.
Trade secret protection is also available
An innovation may be protected by a patent that will eventually become publicly available and/or a trade secret that should remain as a secret. In AI software, functionality of the system may be protected by a patent, whereas certain details of the algorithm or even the code may be protected by a trade secret.
It is critical to note that trade secret protection is not given to information or documents that are merely not the subject of patent applications. Instead, a trade secret program should aim to restrict access, label the materials accordingly and store those materials in a safe place. In some instances, a document describing the trade secret may even be useful.
With such a program, an innovative solution may take advantage of both patent and trade secret protection.
Be aware of deadlines that can preclude protection
Disclosure of technology, whether planned at a conference or a partner meeting, or unplanned and incidental, may cause a forfeiture of patent rights. For U.S. patents, an inventor has only a year from the date of any public disclosure to file for patent protection. Most international jurisdictions require inventors to file for patent protection before any public disclosures.
What entails a “disclosure” is fact specific, but it is generally considered a sale or offer of the AI, a presentation or publication about the AI or underlying algorithms, or using the AI in business, even if the end user is not aware of how it works. Startups often run into trouble by disclosing the technology when pitching investors or negotiating with third-party business partners.
A company can sometimes protect itself from disclosures through confidentiality or non-disclosure agreements (NDAs), allowing it to engage with partners or investors privately. It is critical for companies to consider and define who can disclose the details of the AI and under what circumstances, particularly when patent protection is directed for those details.
The way forward: Integrate IP considerations into strategy
As AI becomes more prevalent in our economy and integral to innovation, startups cannot afford to ignore IP rights. To avoid missing out on opportunities to fundraise, commercialize products and illustrate innovation, AI-based startups should ensure IP considerations make up an integral component of their business strategy.
Here are six considerations to keep in mind:
- Consider various IP protections for innovations of current products, future products and enhancements likely to be adopted by competitors.
- Remember to pursue protection before disclosure. It may be worth filing an initial application on the fundamental technology (e.g., as a provisional patent application) for an opportunity to obtain broad protection on core ideas.
- Evaluate innovations and IP strategies at various milestones, including the release of new products and decision points during patent examination.
- Outline a deliberate IP strategy aligned with the company’s corporate strategies.
- The IP strategy is not a one-time decision, and it should be reevaluated regularly. Continue to build upon the existing applications (e.g., with continuation application filings) to provide more robust protection in critical areas of innovation or inventions.
- Focus international filings on strategically critical markets, as it is not needed to pursue protection everywhere for every invention.