Stay up to date with the latest Gevers news by signing up for our newsletter.
Artificial Intelligence (AI): Dissecting the AI Value Chain
Dissecting the AI Value Chain
Digital technologies continue to evolve rapidly and the use of Artificial Intelligence (“AI”) has become integrated into all kinds of applications and industries. AI is the backbone of search engines, chatbots, customized marketing, self-driving cars, smart devices, cybersecurity and many more applications. AI and data are driving companies to innovate and sometimes transform their business models.
Artificial Intelligence can be defined as the discipline of computer sciences aimed at developing machines and systems, which are able to carry out tasks considered to require human intelligence, with limited or no human intervention. There are many potential applications for AI many of which are linked to human skills such as reasoning, learning, planning and creativity.
From a legal perspective, AI is a challenging topic as it is a new phenomenon where different stakeholders are involved. Many of the characteristics and implications of AI had not been anticipated and are therefore not covered by legal frameworks. While the EU has made European Data Strategy and artificial intelligence a priority, including a proposal for an AI Regulation laying down harmonized rules for the EU of 21 April 2021, the application of intellectual property rights to AI remains a point of debate. In view of this, companies need to rely on due contractual frameworks for the use and protection of data and AI.
In this article, we will zoom in on the different value points of the AI chain and how they interact. For illustration purposes, we will use the example of an AI system that assists city councils in making decisions on certain safety measures, based on the input of traffic flows data and city characteristics.
1. Traditional vs. AI setting
In a traditional setting, the subject matter of a service and the process of performing it are identifiable and traceable. The parties will agree on a certain division of rights, taking into consideration pre-existing rights (background) and newly developed rights (foreground). While these discussions may be challenging, the parties generally have a good view on the subject-matter of their cooperation. In a traditional software development setting for instance, a client or the principal, will ask for a specific functionality or outcome and the software developer will develop a code to perform such functionality with the data of the principal. The parties will make arrangements on the ownership of the software rights and may provide each other with certain licenses to ensure the smooth continuity of their operations.
An AI setting however, is often more complex because there are more stakeholders in the process who each contribute a certain value to the final product. This value chain is roughly made up of the following components; the AI Tool, Training Data, Production Data and AI Output.
2. Components of the AI value chain
In this section, we will discuss the different components of the AI value chain, the IP rights that may affect them and their impact on the relationship between the various stakeholders. Ideally, the ownership and use of each of these components has been clarified between the concerned parties.
- AI Tool: This is the software that has been developed to perform certain – intelligent – functionalities. In our example, this would be the AI software that is capable of analyzing traffic data and making predictions on hazardous situations.
In most cases, the developer of the AI Tool will own the rights to it, which may include patents, copyright and software rights. If another entity wishes to use this AI tool, it can either (i) obtain a license from the AI Tool provider, if it gets access to the AI Tool, or (ii) pay for the services of the AI Tool provider, if the AI Tool provider operates the AI Tool by itself.
- Training Data: In a supervised learning environment, the AI Tool must be trained in order to achieve a certain level of functionality through the use of training data. In our example, this data will consist of labeled images of the different motor vehicles, bicycles, pedestrians, dangerous intersections, normal streams of traffic and the corresponding collisions between traffic participants, and traffic infrastructure (traffic lights, pedestrian crossings etc). It is critical in the training of an AI Tool that the Training Data used are of a good quality and, to that end, it often requires a great deal of effort to produce data of this quality.
The Training Data may be protected under the sui generis database law, copyright or trade secrets, but this is not always certain and will depend on the situation. In general, the Training Data provider will own the rights to it and will grant a license to use the data for specific AI training purposes. However, the Training Data provider may also be interested to obtain certain rights in the AI output (see below), in order to further grow its database. In view hereof, parties may consider granting cross-licenses to each other.
- Production Data: Once the AI Tool has reached a certain level of functionality, it is ready for use. This requires certain source material or “Production Data” which will be fed into the AI Tool. In our example, the Production Data will consist of city specific traffic information such as images of traffic cams and satellites, images of the city infrastructure, information on the number of vehicles and their driving speed.
The Production Data may be protected under the sui generis database law, copyright or trade secrets, but, like the Training Data, this is not always certain and will depend on the situation. Production Data is often produced and owned by yet another party, and the use of it for the purpose of the AI Output will be governed by a separate agreement between the Production Data provider, the AI Tool Provider and the principal.
- AI Output: This data is the compilation of the result from the work performed by the AI Tool based on the Production Data. In our example, this would be a prediction of places where accidents or traffic jams are likely to occur and where additional safety measures are to be considered. In some cases, the AI Output can be further subject to a validation, which further enhances the value of the AI Output.
The AI Output may be protected under the sui generis database law, copyright or trade secrets, but again, this is not always certain and will depend on the situation. With regard to the AI Output, there are several parties that may assert rights to it; (i) the AI Tool owner since the AI Output has been generated by the AI Tool, (ii) the Production Data provider as the AI Output was based on the Production Data, and (iii) the principal because the principal has ordered and paid for the AI Output. Moreover, each of these parties may have an interest in using the AI Output for secondary purposes; (i) the AI Tool owner may be interested to use the AI Output to improve its AI Tool, (ii) the Production Data provider to grow its database, and (iii) the principal to license the AI Output to third parties – in our example for instance car manufacturers or parking facility operators.
For each of these cases, the parties will need to agree on ownership, access and exploitation rights, and address potential points of conflict. For instance, if the AI Output is used to improve the AI Tool, the principal may wish to ensure that the improved AI Tool will not be used for the benefit of a competitor.
When it comes to the monetization of these different components, the possibilities may be different depending on the AI method that is being used. In practice, many use cases of unsupervised AI remain confined to internal purposes, such as the performance of machines and customer satisfaction, without further monetization of the AI components. To the contrary, use cases of supervised AI are often better suited for further monetization due to their scalability such as cancer diagnostics in the field of life sciences
When dealing with AI, it is necessary to make clear arrangements and formalize them in a written agreement. By doing so, stakeholders will create clarity and avoid discussions. Special attention should be paid to the different Value Points in an AI setting and decisions have to be made on ownership and access rights issues.
The GEVERS AI team will be happy to assist with any questions you may have or to set up a short consultation, please contact us at mailbox.contracts@gevers.eu or reach out directly to one of our specialized attorneys.
To learn more about Artificial Intelligence (AI), please watch for our next quarterly newsletter, to be released January 2022, that will expand on this interesting subject matter on an IP level.