Imagine That IP Law Blog

Artificial Intelligence (AI) is everywhere these days, from personal assistants to self-driving cars. With the continuous introduction of new technology, the use of artificial intelligence in business has been growing rapidly, and in ways that people have never imagined. AI is used in many software products and services, and is also being integrated into manufacturing processes. AI frequently automates routine tasks that were previously performed by people, eliminating tedious work, making business processes more efficient, and creating new capabilities and opportunities. Every business in the future will be using AI to some extent.

Although there are different ways to implement an AI system, the area with the most activity today is using machine learning, where the software learns and adapts. One type of machine learning is using a deep learning neural network. Deep learning neural networks mathematically simulate how neurons work in a human brain using linear algebra, and thus, these neural networks operate as a very simplistic human brain. They learn, and are not simply programmed as with most software. In order to learn, the neural network receives training information. The output of the neural network is compared to a desired result and then the complicated math within the neural network is adjusted to more closely achieve the desired result. After this training process is repeated a large number of times, the network is considered to be “trained” and it can predictably do the desired task. For example, many cell phones use facial recognition, to classify the features of a person’s face. They learn the features of the cell phone owner’s face and can use those features to recognize the owner with a high degree of accuracy. Because of this training phase of the AI system, the system is really only as good as the data set that is used to train the algorithm. If the training data is skewed, for example by being racially biased, the output of the AI system will be likewise biased. This is only one example of the way in which an AI algorithm can ultimately provide a biased output.

From an Intellectual Property (IP) perspective, AI is essentially software. It is a software program performing one or more tasks. However, there is a significant difference with an AI program in that the software learns and adapts as it is used, and continues to receive new information. Therefore, the software in use a year or two later is different than the software that is new; i.e. “out of the box.” This adaptation feature is very different from software or devices that are not able to learn and adapt.

The areas of IP that cover AI are copyright, trade secret, patent and, although not directly IP, privacy. Copyright protects original works of authorship that are produced in a tangible form of expression. Copyright protection obviously applies to the software code used for the AI functions. A copyright is inherent in the created software code and that code can be filed at the US Copyright Office to obtain a Copyright registration. Copyright protection also potentially applies to products produced by the AI. For example, someone uses an AI to analyze data, the report produced from that AI could be copyrightable. A growing trend now is to use AI to create stories and images. For example, the Dream app by Wombo can turn a person’s words into a drawing. One recent issue is whether the AI can be considered the author of such a work. So far, the Copyright Office says no.

With respect to Copyright protection for AI, much of this will be handled with standard software licenses and service agreements. The vendor of the AI software will license the software product to the user or provide a service with the AI. However, one area to be on the look out for in such agreements is to make sure that any copyrightable materials that produced using the AI are owned by you. This would typically be the case, but due to the learning and customization capabilities of some AI software and services, it would be desirable for you to own the rights to those aspects as well, to the extent possible. For example, AI software for document review may become very adept at identifying relevant documents for your particular clients, the more that you use it. Once the AI software is fully optimized, it may be a huge hindrance to lose that service. That could place you in an unfavorable negotiation position with the service provider, if that provider realizes the value that the AI service provides to you. Additionally, you may also be inadvertently providing the service provider with a free service, as training their AI software makes it better. Although this has always been an issue with early adopters of software products, who have to suffer through the bugs and glitches. It is a bit different with AI, where the AI is learning on confidential information.

Trade secret is a practice or process that is not known outside of the company. The most referred to is the formula for Coke. Due to the intertwining of AI into business processes, there is much potential for trade secret protection. For example, Tesla performs all of its non-destructive testing on automobiles using AI without humans. This is obviously a big advantage that Tesla has over competitors and it would not want it leaked out how it does key aspects of this non-destructive testing. Since trade secrets just exist, the key way to maintain a trade secret is by taking reasonable measures to protect it, such as having employees sign non-disclosure agreements (NDAs) and having non-compete clauses in their employment agreements. Also, having rules and policies protecting the trade secret information is helpful to show that reasonable measures were taken. Since AI software is always learning and adapting, the changes to the software as well as the output of the software are both subject to protection of a trade secret.

Patent protection allows the patent owner to be able to prevent another from making, using, distributing, importing or selling the patented invention. Patent protection can be obtained for a new non-obvious product or process. With respect to AI, patent protection can be obtained for the AI software or product if novel and unobvious. Additionally, the use of an AI function in a service or product may also be patentable. For example, an AI based facial recognition system may be integrated into a financial software program in a novel way. That overall financial software program may be patentable. However, if the invention is to use the AI to simply automate a human function, it may not be patentable under 35 U.S.C. 101, as not being patentable subject matter under the Alice line of cases. Generally, simply automating prior systems using AI or other software is not patentable subject matter.

Similar to Copyright, the issue as to whether an AI can be an inventor for a patent has been raised. In the recently decided Thaler v. Vidal, No. 2021-2347 (Fed. Cir. Aug. 5, 2022), the Federal Circuit confirmed the district court’s decision that “inventors” must be natural persons, so an AI cannot be an inventor. For the patent application at issue, the only listed inventor was the AI, referred to as DABUS. Since the patent application lacked a valid inventor, it was considered incomplete and not accepted by the USPTO.

Privacy is a growing concern relating to AI. Customers and employees should have a reasonable expectation that certain information is to be kept private. This is even a legal requirement in some industries, such as the health care industry. However, there are examples of this when this expectation is clearly not met. Recently, a father learned of his daughter’s pregnancy by consistently receiving coupons for baby clothes and cribs from Target. The daughter had not registered with Target as an expectant mother. However, Target was running a predictive marketing (AI) program that identifies expectant mothers. The predictive marketing program analyzed the daughter’s purchases, such as scent-free soap and cotton balls, that the daughter was likely pregnant. This likely was not something that Target’s marketing department foresaw.

AI is frequently used to analyze large amounts of data. This data will sometimes include information that a person wants to keep private. Additionally, it may be information that is believe to be benign (as with the Target example above), but in hindsight, may have been better kept private. Implementation of AI systems within the corporate context presents an incredible opportunity for increased efficiency, and can supplement already overworked staff. However, if not implemented correctly, it can bring unintended downsides and risks to an organization. Relying on an AI or any computerized tool comes with the side effect of losing immediate reviewability by a human. A poorly managed AI system could easily cause greater liabilities than introduce efficiencies, and properly evaluating, monitoring and auditing an AI system could make the difference between these outcomes. Employers should be aware of all of the benefits and risks before utilizing such a system.

Reprinted with permission from the August 31, 2022 issue of The Legal Intelligencer ©2022 ALM Media Properties, LLC. Further duplication without permission is prohibited. All rights reserved.