How AI Chatbots Could Change Trademark Law

Trademark law is essentially a law built on human imperfections. Phonetic, conceptual, and visual confusion of trademarks, imperfect recollection, and slurring of words all presume that so-called average consumers are inherently imperfect, and indeed they are. However, will the rise of artificial intelligence (AI) chatbots such as OpenAI’s ChatGPT and Google’s Bard change that? Will they import perfection into the brand search and product purchasing processes or are we just adding another imperfect player to the mix?

AI’s Emergence

For years, intellectual property (IP) professionals have been predicting how the rise of AI will impact the purchasing of branded products. Indeed, the authors have been writing, speaking, and commenting on this potential phenomenon since 2017.

Although the recommendation and buying processes of branded products and services already incorporate aspects of AI, the most readily and widely used are product recommendations via online retail platforms and marketplaces, such as Amazon. Many consumers do not realize that they interact regularly with AI technologies, most notably via interactions with banks and the like. So far, the retail market has not felt the full force of AI in product and service recommendation and purchasing.

Some market observers believed the rise of retail assistants, such as Amazon Alexa, might herald a new revolution and once again fundamentally change the way a consumer purchases a product.

Observers also thought that applications such as Amazon Alexa would herald the rise of voice search (or maybe its reemergence, remembering that many products were bought on the basis of recommendations from human shop assistants), potentially changing the dynamic between the phonetic, conceptual, and visual comparison of trademarks. In this way, consumers would become accustomed to relying on recommendations from an AI application or even, going further, allow such applications to buy branded products for them. The so-called change of the retail model from a “shopping then shipping” to a “shipping then shopping” model, if only in a limited form, was thought to be close.

Although, Amazon is reportedly still committed to its Alexa platform, which is very popular with many consumers who use it to enjoy music and obtain basic information, its impact on product and service recommendations, and subsequently the ordering of products, has been more limited. Where could the revolution come from in the shorter term?

AI Chatbots and Online Search Applications

The answer to that question may well be the impact of AI chatbots in the world of online search applications. OpenAI released its ChatGPT chatbot application for public testing on November 22, 2022. This technology is based on GPT 3.5 software, which is a so-called autoregressive language model that uses supervised, deep learning to generate human-like text conversations. AI chatbots are based on so-called generative AI systems, with these systems being reinforced via learning by interactions via human feedback loops. “GPT” in ChatGPT stands for “generative pre-trained transformer.”

Google’s BERT technology was one of the first so-called transformer models. Such an application must be “trained” on learning data. This data is often taken from the Internet in other languages besides English. ChatGPT can reportedly “speak” 51 languages. Every human interaction reinforces or changes its “learned norms.” Many readers will no doubt have played with ChatGPT. It was reported to have had 57 million users in its first month of operation. Poems have been penned, blogs written, and philosophical questions have been answered, but, potentially, the biggest commercial impact of AI chatbots could be in the online search market and not on trademarks.

Microsoft paid $10 billion for a 46 percent stake in OpenAI and announced on February 7, 2023, that it would incorporate AI chatbot technology, of a similar nature to ChatGPT, into its Bing search engine.

A text conversation now appears alongside the “natural results” of a Bing search. In an article in The Economist on February 9, 2023, Eric Schmidt, Google’s Chief Executive Officer from 2001 to 2011 said ChatGPT is the “first broadly visible example” of what a true AI assistant to a human may look like in the future. Microsoft’s CEO Satya Nadella called the announcement of Chat GPT’s incorporation with Bing as “a new day in search” and that the race to catch up with Google in the online search market had begun.

Search Accuracy

Google became the market leader in search engines because the majority of consumers around the world perceived its search engine algorithm as most accurately reflecting the most useful websites in any information search; these are Google’s “natural search results.”

From there, Google used keyword advertising sales to build the world’s fourth largest company by revenue.

However, the natural search results that search engines such as Google generate are still essentially lists of websites. Sure, Google’s algorithm ranks them according to relevance and utility, but they are still essentially lists of information and website links that are many pages long.

Conversational

AI search chatbots could fundamentally change that picture and how consumers search for products and services. For example, ChatGPT which is based on a large language model, considers billions of parameters, often many, many more than conventional search engines. It is trained to go through billions of pages of material and then produce conversational texts prompted by the consumer’s request. The computational power required to make searches of this nature is much more expensive than what is needed for traditional searches. But as The Economist reported in its February 9 article, if 20 percent of the most lucrative searches were conducted on this basis, this could significantly disrupt the online search market.

In theory, ads could be embedded in such conversational texts and, indeed, ad bidding could develop in the same way keyword advertising has in traditional online searches. The technology also raises the risk of chatbots developing biases or being open to manipulation by either their owners or outside forces, in the same way that search optimization techniques in the traditional online search market has raised the same risk.

On February 8, 2023, a day after Microsoft’s announcement about the incorporation of a ChatGPT-like application with Bing, Alphabet, Google’s parent company, announced that it would incorporate its own AI chatbot technology, Bard, into its search function (and China’s Baidu is following suit with its own AI chatbot technology). However, a mistake in its first public demonstration overshadowed Bard’s launch. When asked “What new discoveries from the James Webb Space Telescope can I tell my nine-year-old about?”, Bard offered three answers, including that the famous telescope “took the very first pictures of a planet outside of our own solar system.” But that is not correct; that happened 14 years before the James Webb Telescope was launched.

Making Mistakes

What does all of this tell us? Firstly, how online searches occur could be fundamentally changing for the first time since Google launched its search engine technology in 1998 and quickly became the leader in this market in most of the world.

AI chatbots applied to online searches could be the first significant embodiment of AI technology to impact how people search for and purchase branded products and services in that very important sector of the market. Much has been made of the creativity of AI chatbots, and no doubt many of us have seen some of our contacts on LinkedIn—which Microsoft also owns, as it happens— “testing” ChatGPT. However, the true impact of AI chatbots—beyond their use as an amusing toy—is in online searching.

What the Bard launch also illustrated is that AI chatbots are not perfect. OpenAI’s GPT 3.5 software was trained on data only up to 2021 and knew nothing of the world after that date. Although the software being used with Bing is slightly different, and more up-to-date, and the loop mechanism helps with accuracy, some AI chatbots have exhibited a fundamental flaw by using imperfect data sets to make up an answer to a question or prompt if they do not have the correct information. These are so-called hallucinations. Could AI hallucinations be added to well-known concepts of trademark law such as imperfect recollection and confusion?

Questions for Trademark Owners

AI applications were often perceived as importing a type of perfection into the purchasing process. These perceptions were perhaps influenced by classic films such as Metropolis, Blade Runner, and The Terminator, in which AI applications were portrayed by the filmmakers as perfect and all powerful. There would be no imperfect recollection, there would be no slurring of trademarks, and there would be no confusion. Some commentators, notably Michael Grynberg of DePaul University, in an article in the Kentucky Law Journal in 2019, have even raised the prospect that there would be no need for trademarks as a shorthand for brand values, as AI applications do not operate in the same way as the human mind. Who needs shorthand when you have an infinite source of product information from the Web? However, as we explain above, AI chatbots have their flaws.

Lord Justice Arnold of the England and Wales Court of Appeal raised a number of interesting points in an online discussion about AI and the impact on trademark law, organized by Anke Moerland, Associate Professor of Intellectual Property Law at the University of Maastricht in the Netherlands in 2021.

The judge used the example of NIKE and NIKF to discuss questions such as:

• Could AI applications be confused via “fuzzy matching” programs?
• Though AI applications could learn not to confuse NIKE with NIKF, could indirect confusion still occur?
• Could an AI chatbot associate and link a NIKF-branded product with a NIKE-branded product? Would there be free riding via the AI chatbot?
• Could an AI application mishear or misunderstand a consumer’s commands?
• Amazon Alexa can adapt to a consumer’s accent, but could errors happen with AI chatbots and other AI applications?
• How will AI chatbots tell the difference between genuine and counterfeit products?

Justice Arnold raised the question of reputation. Amazon Alexa generally makes only three product recommendations when suggesting branded products to consumers. Indeed, there is a fear that AI chatbots may entrench established brands without the long lists of websites available to consumers in conventional search results.

Justice Arnold also raised the question of whether repute (a generally held opinion) could influence an AI chatbot’s product recommendations. Can AI chatbots be manipulated? The obvious example is counterfeit products. It is often hard for humans to compare genuine and counterfeit products instantly. Would AI applications be any better?

Further, there are reports that some AI chatbots are being manipulated to import bias into their recommendations.

These answers and recommendations at present may be simple answers to informational questions. However, what happens when commerce gets involved and product and service recommendations are being manipulated? Bias is, after all, inherent in many human interactions, and it can be learned, so why not with AI chatbots?

Competing AI applications could also counter and manipulate each other. You just have to think of the importance of online searching for branded products, which led to the emergence of the search engine optimization (SEO) industry, a practice that is based on optimizing the search algorithms of online search platforms for the benefit of brand owners. Could similar industries develop in relation to AI chatbots?

Trademark Law Can Adapt

The possible impact of AI on the world has been the subject of much speculation, particularly since the release of various AI chatbots for public use.

We and others have speculated in particular on the impact of AI on how products are purchased and thus by definition its impact on trademark law, which essentially regulates the interaction between brands and the purchasing process. It seems likely that, despite some false dawns, the rise of AI chatbots, such as ChatGPT, could fundamentally change online searching.

At the same time, chatbots have proved to be fallible and imperfect. They learn, they make mistakes, just like humans do. Indeed, many AI applications import imperfection to make them more palatable to humans (since humans seem to relate better to AI companions that make mistakes).

Though the growth of AI applications raises new IP infringement issues, trademarks basically still do what they are meant to do; they are still “fit for purpose.” Trademark law has adapted over the decades from the human shopkeeper to the development of supermarkets, to the rise of the Internet and social media and will continue to adapt with the rise of AI, including chatbots.

Register and learn more about INTA’s upcoming Business of AI Conference.

Although every effort has been made to verify the accuracy of this article, readers are urged to check independently on matters of specific concern or interest.

© 2023 International Trademark Association