No one can deny it anymore. AI is coming and will take over many aspects of our daily life. However, what kind of aspects? Will they take over our role as a consumer? Nowadays, Artificial intelligence takes wider forms when it intervenes in the retail industry. It can take the form of simple Amazon recommendations based on pre-established parameters invisible to consumers, but can also fit the role of a real shopping assistant by helping consumers throughout the process of purchasing like Ebay chatbots. In some scenarios, AIs take charge of the whole purchasing process by ordering themselves goods or services leaving no room for the average consumer. To sum up, the Fourth Industrial Revolution is currently allowing AI to intervene in any step of the purchasing process.

Reactive v. predictive retail industry
The uniqueness of such intervention rests in the ability to predict what consumers want without needing a complete picture of branding information. At the antipodes of such conception, our world has mainly developed on a reactive retail industry model in which consumers were at the heart of the purchasing process reacting to branding information. This model has served many societal foundations but most importantly, it has served as a basis for trademark law. Trade marks grant information to consumers in order to help them make purchasing decisions and thereby reduce search costs. The rationale of trademark protection is essentially based on a reactive modeled industry.

What about trademark law rationales?
What happens when AI takes part in the purchasing process? Is there still a good reason to grant exclusive rights? Is it necessary to present consumers with branding information? For instance, taking a product suggestion type of AI which suggests a few brands to consumers but leave some aside, what determines its choice and does a brand still play a role for its suggestions?

Typically, a recommender system will suggest products to consumers based on a content-based filtering method or a collaborative-filtering method. This means that based on previous likes, views or purchase by the end user or similar users, the AI will suggest products to consumers by predicting that if it was viewed by him or by similar users in the past, he might like to purchase it. Strictly speaking, this leaves trademark considerations aside and begs the question of whether trademark law is still needed as AIs are perfectly capable of predicting what consumers want based on other parameters. However consumers are unpredictable and one cannot deny that their previous activity on a platform was certainly influenced by trademarks. This could be because the trademark was used as an indicator of origin or quality but also as a symbolic image which highly influence a consumer in his purchase. If AIs were to completely take over purchasing decisions independently of trademarks, would consumers really be satisfied with the products selected regardless of the trademark attached, or no-name products? This means that if machine-learning algorithms want to aim a high level of accurate suggestions, they may need to be trained to include trademark considerations.

On top of that, AIs that do not consider or even depict trademarks in purchase suggestions may trigger other discussion. A recent case has ruled in favour of an extension of these trademark rationales. In Mitsubishi v. Duma, the CJEU has recently ruled that a non-use of a trademark in the context of de-branding some goods and than rebranding them with a different sign, would harm the trademark functions as it deprives the trademark owner of displaying his goods bearing the sign on the market. Even though the outcome of such a case is debatable, and it does not involve AIs at all, maybe it would serve as a basis for assessing the apparent conflict between trademark law rationales and AIs in the context of goods not being displayed in online platforms. In other words, this interpretation may suggest for AI that not displaying a brand to the end consumer limits the communicative function of the mark. Since trademarks are not only protected to indicate the trade origin of a product but also to communicate certain qualities of the product, one could argue that a trade mark when perceived by a human consumer, still fulfills a function in a world of AI, namely being a carrier of information and feelings. 

This means that AIs should take into consideration all trademark law functions, including the non-essential ones, so as to be compliant with the rationales of trademark law, as well as to offer customers desirable experiences. Caution of course is needed to not run into anticompetitive claims of preferring certain undertakings over others. As a general rule, one shall always take into account relevant consumers, which is one of the most important notions in trademark law. Working on better information to consumers given by AIs may be the solution to our problem. This might be achieved by equipping AIs with the ability to autonomously react to branding information.