Artificial intelligence (AI) is proving central to many recent commercial transactions, such as asset or business acquisitions or licenses, and this will only increase as interest in AI explodes. Whether such commercial transactions concern an investment into AI-driven technology, competitive industries, or delivery of services that may be revolutionized with AI, here are four crucial considerations when making a deal.

What is it? Identifying the Assets at Play

In an AI-focused deal, it is crucial to correctly define the AI-related assets at play, including those that may be impacted by or utilize AI. By way of example, vague words like “software” or “application” may not capture all the novel elements of the AI asset and, in that case, a purchaser may not receive everything it needs to operationalize the business.

On the other side, using a term like “the artificial intelligence solution” could mean that the seller might inadvertently make promises (representations) about the asset, the scope of which are much broader than the seller intended opening them up to liability for issues with, for example, hardware on which the solution runs. In a licensing transaction, a licensor may end up with conflicting obligations to different licensees or issues with its own licensors who have rights over, for example, certain algorithms, or may appear to license certain intellectual property (or even physical property) that it may not have ownership over.

Given the unintended consequences that may arise in an AI-focused deal, a solid definition of AI is grounded in technicalities. Both parties should work closely with counsel to ensure that the definitions being used in a definitive document accurately capture the asset in question.

At a high level, any definition of AI should include concepts related to machine learning, proprietary algorithms, and any software, hardware, equipment or systems that make use of or employ neural networks. Although such language may not always fit all circumstances, these concepts are worth noting when drafting the scope of the AI definition. In addition, both parties should carefully consider whether data sets (as discussed below) should be included in this definition of applicable AI.

How Did it Learn?

AI needs to learn in order to fully deliver its expected value. As a result, the quantity and quality of the data that the AI is learning from usually is a key driver in the value and complexity of commercial transactions. This particular issue has already hit mainstream discussions with, for example, musicians and authors contesting the right of AI-driven content creators to train AI using these musicians’ and authors’ work (without paying for the privilege to do so). Is it fair, for example, for an AI creator to upload copies of a popular musician’s works to train the AI to make its “own” music? 

Avoid missing the mark or misunderstanding how the AI can learn by investigating which datasets are relevant to the AI, who owns the datasets, how much data is there, and how good it is. These, and similar probing questions, will allow you to understand the specific data used to train an AI model. 

At an absolute minimum, it is crucial that parties agree as to whether or not the training data sets are included in the transaction; in other words, if a party is licensing or purchasing the AI, are they also purchasing or licensing the underlying datasets? Express legal terms related to transferability and ownership will likely become the norm, and as a result, we expect that sellers (or the AI owner) will be required to make robust representations about their title and ownership interest over the data sets or, at minimum, their right to use the data sets.

Can it Keep Learning?

AI is valuable not just because of what it knows but because it keeps learning. Without this skill, and the necessary data sets to facilitate this learning, the technology may quickly become irrelevant. Similar to a proprietary formula for a drug or food, a buyer may want to confirm that the AI is capable of training and retraining, reproduction, further expansion and development, and debugging.

So, as a seller or licensor, it is in your best interest to consider that a buyer or licensee may want to continue to develop and enhance your AI well-beyond simply feeding it more data, and to determine whether or not your technology can live up to that expectation. On the other side, a buyer or licensee should never assume that future or more complex capabilities are built into the current AI being purchased or licensed.

The Good, the Bad, and the Ugly

Over the past few years, certain AI systems have been accused of racial, gender and other types of unacceptable biases. Further, government agencies are figuring out how to apply their regulatory tool boxes to this new technology. With these issues in mind, we may see certain related market expectations of sellers; for example, investors may request a representation or promise that the AI was developed, or may require a covenant that the creator cause the AI to continue to operate, ethically, responsibly and in a way that is compliant with industry standards, community standards, and anticipated regulation, much in the same way some venture investors currently require their portfolio companies to adopt policies related to diversity, equity and inclusion. This may give rise to heightened scrutiny, and certainly, carefully crafted representations or covenants regarding the capabilities of the AI in its current or future state.

The takeaways within this article are intended to cover only a portion of what may be useful when involved in an AI-based transaction. Legal counsel and corporate leadership should work closely to ensure that all aspects of AI are being carefully analyzed and thoroughly discussed to ensure the protection of both parties in this ever-developing space of AI.