|1 minute read|
|The impact of artificial intelligence (AI) on trademark law is as yet, untested, but here the authors provide a practical insight into some of the possibilities of the effect in practice that this technology could create. With the rise of voice searching and purchasing, there will be an increased importance on the phonetic aspects of trademarks, and the responsibility of accurate searching could be pushed back to the consumer. The authors suggest that we will see a shift in the law regarding the liability of AI and a new legal test for infringement as we see a need to identify who (or what) is making the purchasing decision.
When you ask Alexa, Amazon's artificial intelligence (AI) assistant, how 'she' works, she says: "Lots of people have worked hard to teach me, and I am still learning more." She does not deviate from that response, and although Alexa does not give much away, the way in which Alexa – and indeed all AI assistants – works will have a practical impact on how trademark law will work in the future.
In our previous Managing IP article on the topic of AI and trademarks (in December 2017), we raised – probably for the first time in print – the possible impact AI may have on some of the core concepts of trademark law. Notably, how will AI impact the fundamental and accepted concepts of trademark law such as 'average consumer', 'imperfect recollection', 'likelihood of confusion', 'concepts of aural, conceptual and visual comparison' and the 'slurring' of trademarks? The ultimate question was, 'can AI be confused?'
As we explored previously, trademark law today is still fundamentally based on nineteenth century concepts of how products are purchased, but when AI is involved it means that the purchasing of products and services moves from a reactive to predictive mode. As such, many of the current concepts of trademark law may simply not be relevant anymore and the law will have to adapt, as to a degree it has always done, to be flexible to the new ways products are purchased.
Enough about the theory, what about the practical implications?
However, what will be the practical implications on the average trademark attorney or trademark lawyer of this 'AI revolution'? After all, trademark law fundamentally concerns itself with the way products and services are purchased and the relationship between the consumer and brands. Trademark law is a very 'practical' aspect of the law.
Sébastien Szczepaniak, head of sales and e-business at Nestlé, in an article in The Wall Street Journal (written by Saabira Chaudhuri and Sharon Terlep, February 2018) predicted that within five years, as a direct result of the rising use of AI assistants in retail, 50% of searches for products will be done by voice. This will no doubt only increase over time as the popularity of AI assistants rises. Of course, in the UK and the EU, as in the majority of countries of the world, the oral use of trademarks is considered trademark use. Therefore, by definition, use of voice searching for brands is trademark use.
Practical implications of 'voice' searching rise
When comparing trademarks, we are all familiar with the analytical framework for comparing trademarks, i.e. their phonetic, visual and conceptual impact. Historically the importance of one form of comparison over others has waxed and waned, and with the shift to voice ordering, the phonetic comparison is expected to take centre stage.
By definition when the majority of product or service searching is conducted by the voice, the visual impact and visual cues of branding will be reduced, if not eliminated, in the process. Further, at present the majority of online shopping is conducted via typing on a keyboard or mobile phone and browsing through search engines such as Google or via online shopping websites or marketplaces. However, the rise of voice searching means that in the 'trinity' of trademark comparison, at least with regards to the voice purchasing of products, phonetics will rise in importance, followed by the conceptual impact of the brand, with the ultimate relegation of the visual impact of a trademark.
We also contend that such a dominance on the phonetics of a trademark will lead to fewer instances of typo-squatting or domain name squatting, as fewer people will be relying on having to actually type the brand name into a search engine. Gradually, the parties behind such squatting may cease to bother registering similar domains, as it will not be worthwhile pursuing, or at least the attractiveness of this form of infringing activity will fall. Typo-squatting will reduce as consumers move away from typing all together – it is not just shopping where people are using their voice, as technology in general is moving towards voice command (think Siri, Google Assistant and Cortana). As such, attorneys and lawyers are likely to see a reduction in the number of domain name actions crossing their desks.
The development of AI applications and issue of liability is a major point of debate as this new form of technology rises in importance
Coupled with this, the nature of search engine optimisation as we currently know it will change with the rise of voice search. These changes are not unprecedented and neither is their impact on the practicalities of trademark law. In the early days of the rise of online shopping, there was a spate of cases concerning the use of metadata and metatags and trademark infringement claims, the most notable in the UK dating from 2004 between Reed Executive and Reed Business Information. This form of trademark infringement is now almost unheard of, as the importance of the use of metadata in search optimisation has fallen, particularly on Google. No doubt with the rise of voice search, new forms of optimisation will develop and with that new forms of trademark infringement. Indeed optimisers are already attempting to determine how Amazon ranks listings in Alexa, with popularity (sales), rating, reviews, price, delivery speed reportedly having an impact. Further, as online searching is commanded by voice, how will consumers see sponsored content?
AI assistants introduce some new issues to the phonetic comparison of trademarks. By definition, AI assistants are based on voice recognition software. How a particular consumer pronounces a brand and how the AI assistant interprets this may introduce another factor into the product purchasing process. It could also mean that such AI assistants may have to reconfirm brand name suggestions by the consumer. It has been recently reported in the press that Amazon employees do 'sample' certain recorded conversations between consumers and Alexa, subject to privacy protection controls, to improve the accuracy of its voice recognition software and leads to Alexa learning how individual consumers pronounce words and what they mean. By definition this will equally apply to the pronunciation of brand names.
It is, however, highly unlikely that the visual impact of trademarks in the comparison framework will be removed completely with regards to all product and service sales simply due to the inherent structure and make-up of the retail market. The rise of online shopping and its impact on the retail market has been widely commented on in the national and international press and news. Market research firm Mintel reported in 2018 that in the UK, online retail sales grew by 15.8% in 2017 to reach £59.7 billion ($77.7 million) which accounted for 16.3% of all retail sales in the UK. However, that still means that 83.7% of retail sales happened 'offline'. Although the percentage of online sales in the retail market as a whole will undoubtedly grow, and AI will also affect offline retail sales, particularly in the form of chatbots and in-store robotic shop assistants such as Pepper, the visual impact of brands will not be completely removed from the retail market, simply because 'bricks and mortar' sales will remain a key ingredient in the average retail marketplace. However, it may mean that as online sales increase, and voice searching within online sales becomes dominant, the occurrence of 'trademark problems' such as lookalike supermarket products falls, as the importance of visual cues in marketing consequently reduce.
Trademark law is inherently adaptable and indeed it is already clear from the case law that the courts take into account how products are purchased when balancing the importance of the phonetic, visual and conceptual impact of trademarks in comparison analysis. However, practically, AI will affect this assessment and this 'balancing act'.
Who is responsible for accurate product searching?
The increased importance of voice over the written word in product searching may push some of the 'responsibility' for the accurate searching for brands back to the consumer. In a conference on AI and its impact on IP law held at the USPTO on January 31 2019, Dana Brown Northcott, (associate general counsel for IP at Amazon) raised this exact point. It may also mean that when making product and brand suggestions to consumers, AI assistants will have to re-emphasise the brand name or ask for clarification.
Consider the difference between purchases made in the pure form of AI, when the human actually takes no part at all in the assessment of the product by brand (a pure AI purchase) and purchases where the human takes an active role in the purchasing decision by telling the AI assistant what brand to purchase (an impure AI purchase). Concepts such as 'imperfect recollection' and the 'slurring of trademarks' will simply no longer be relevant to trademark law when it comes to purchases via AI assistants, where the AI assistant takes complete control. AI assistants do not get confused and have a perfect memory. But, where there is a combination of human and AI interaction in the purchasing process, the importance of concepts such as the 'imperfect recollection' and 'slurring' may actually increase as it is shifted back to the consumer.
Will we see a new element to the legal tests?
Based on the above, it seems as though there may be a need for a shift or diversification of the current legal tests for comparison of trademarks. It would appear that there will be a need to identify who (or what) is making the purchasing decision. If the purchase is a 'pure' AI purchase, there may be no need to assess the likelihood of confusion or whether the trademarks are similar as the trademark did not play a part in the decision. With this route, the AI would only potentially be a secondary infringer. In the instance of a mixed AI and human purchase, the consumer has been involved and states the brand when ordering, so the element of confusion would remain.
Of course, such a shift in the law would happen slowly over time as the cases are decided and precedent laid down, but it is inevitable and something which practitioners will need to keep in mind for future.
Bid ranking and comparative advertising
The rise of AI assistants may introduce another new practical problem for trademark owners and their representatives. The Wall Street Journal article, referred to above, identified another issue 'new' to AI assistants. As the importance of AI assistants rises in retail, consumer goods brands are increasingly concerned over their ranking in the products suggested by AI assistants.
Szczepaniak in the aforementioned article also states that you need to be "first position or you go home". Chaudhuri, one of the authors of that article, identified that in her local grocery store seven brands are on offer for toilet paper, that on Amazon's website there are over 30,000 results for toilet paper, but that Alexa only offered two brands in Chaudhuri's voice search for toilet paper. Although undoubtedly the choice of brands on Alexa and AI assistants will increase, and Amazon is circumspect in identifying the exact basis of its ranking on Alexa, it has identified popularity (sales), rating, reviews, price, delivery speed and of course being available via Amazon Prime as factors in determining whether a product is available on Alexa and how it is ranked.
Although Amazon is clear that it does do not this at present and apparently there are no plans to introduce it, as with bidding on AdWords via Google, could brands not also bid for higher ranking in AI assistant listings? The potential for AI assistant suppliers to generate income by allowing the bidding on ranking in their product suggestion system does raise certain issues, most notably competition issues. However, there is also the potential for AI assistants to engage in comparative advertising. For example if a consumer mentions a certain brand, could an AI assistant for a fee suggest an alternative brand or brands depending on a paid-for ranking? Although there are unconfirmed reports that Google is testing voice triggered adverts, to the best of our knowledge no AI assistant at present employs such a bidding system, or engages in such comparative advertisements. It seems an obvious means to generate income in the same way that Google generates revenue from keyword advertising bidding and introduces exactly the same issues as raised in the well-known Interflora v Marks & Spencer decisions.
Further, certain AI assistants could concentrate on price comparisons, or indeed on the basis of any 'comparison'. The issue is then whether those comparisons are fair and accurate and, if not, whether trademark infringement is happening.
Damages calculation and strict liability
It is often forgotten that the move from personal liability to liability tied to the legal construct of the company as a legal entity developed over time and was not an obvious concept. The development of AI applications and issue of liability is a major point of debate as this new form of technology rises in importance, and trademark law and trademark lawyers cannot ignore this debate. Some commentators have raised, as with companies, that AI applications should have a new form of legal personality as a means of getting around the liability issue. However, such concepts have been widely criticised on the basis that they open a Pandora's box of AI applications then logically being given 'rights' to mirror the 'responsibilities' of liability.
Professor Luciano Floridi, (a professor of philosophy and ethics of information at the University of Oxford) has suggested that one might look to Roman slave law which dealt with the issue of liability via the concept of the 'intelligent slave', with the slave owner being liable for the actions of his or her slave. The 'intelligent slave' in this case being the AI application and the 'slave owner' being the AI application provider. However, in our previous article, we postulated that the liability of the provider of AI applications, with regard to trademark infringement, is likely to be dealt with in the same way as the Google France cases (CJEU, March 23 2010) and indeed in L'Oreal v eBay International (CJEU July 12 2011) in that an AI application provider is only likely to be held liable for infringing activity on their platforms if they are actively aware of such activity and do nothing. Practically, this may well lead to the introduction of the same 'take down procedures' in place in relation to keyword advertising and online marketplace platforms. Further, it is very unlikely that even if an AI application provider might be aware generally of infringing activity on their platform that they would be held liable unless it were tied to a specific instance of infringing activity.
It is often forgotten that the move from personal liability to liability tied to the legal construct of the company as a legal entity developed over time
However, how AI works will also affect the calculation of damages and account of profits. Case law in the UK is clear that with regard to trademark infringement, a claimant is only due damages or an account of profits which strictly relates to the infringing act. It is important in assessing damages that a court has to determine what proportion of the defendant's consumers have been confused and be compensated accordingly. A claimant is not entitled to damages for persons who have not been misled or have purchased a product in a way unconnected to the infringing act.
In a pure form of AI, it is possible for a consumer to completely delegate the purchasing decision to the AI assistant, the so-called 'shipping-then-shopping' model put forward by the Harvard Business Review, and indeed for the consumer not to be aware at all of what products are available on the market by brand. AI assistants affect the purchase of a product in two ways: the information available to the consumer in making a purchasing decision and secondly the purchasing decision itself. In theory a consumer could ask an AI assistant to buy a product based on past purchasing decisions and a criterion completely unrelated to a brand. Would this be trademark infringement? If so, should a plaintiff be compensated at all in such a situation if it can be shown that a consumer or indeed the AI application did not make the purchase based on a brand, but on the basis of other criteria such as price, colour, quality or design? One might argue well how you could possibly know how purchases are made in all instances, that it is not possible in the 'real world'. Well in the world of AI that may actually be possible in future.
The hardware for AI applications is relatively simple. For example, the Alexa speaker in your home is a simple device technologically consisting mainly of a speaker. The key to AI assistants such as Alexa is in the connection to your WiFi and its connection to the cloud and in the case of Alexa its connection to Amazon's services. The key to the Alexa system for example is the Alexa Voice Service (AVS) which is the software system which connects the Alexa device to the cloud and links the consumer comments and commands to Amazon's services such as Prime shopping. AVS is the 'brain' to the device in your home. Indeed Amazon is willing to allow AVS to be incorporated in other devices not under its control to enable those devices to be connected to its services. However, in theory the software powering AI services could provide a record of any purchase, and indeed all purchases relating to certain brands and why certain products are purchased and how. This in theory raises the potential for a court to understand how and why certain purchasing decisions were made, in contrast to the 'real world' where a certain amount of doubt is always at play. Indeed it raises the possibility of a court having a record of all purchasing decisions relating to a particular brand. By definition, this could affect the calculation of damages and an account of profits and will potentially affect issues of strict liability in trademark infringement cases.
AI and counterfeits
AI is already being used in the fight against counterfeits. For example Amazon's Brand Registry is only the start of its efforts on this front, with Amazon developing AI machine learning to identify fraud vectors and identify products which could be counterfeit and remove them from its pages.
However, AI is also being used by counterfeiters. Erling Vestergaard of the EUIPO, speaking at the Fide Conference "Can robots invent and create? A dialogue between Artificial Intelligence and Intellectual Property", hosted by the University of Alicante on March 14 2019, identified specific instances where AI applications have registered domain names which have recently lapsed and have high traffic automatically. Online shops are then automatically set up to sell counterfeit products from those domains. Further, the AI applications adapt their registration strategy based on the enforcement strategies employed against them in certain countries. In short, AI is learning anti-counterfeiting measures and adapting them to a retail environment and trademark attorneys and lawyers will have to take on board these practical realities in their enforcement strategies.
So back to you, Alexa
So where are we and where are we going with AI and trademarks from a practical standpoint? Interestingly, when you ask Alexa 'What is the impact of artificial intelligence on trademark law?' she responds 'Sorry, I do not know that one.' If only we could simply get all the answers from Alexa.
Lee Curtis is a partner and chartered trademark attorney, and Rachel Platts is a trainee trademark attorney, both based at HGF in the Manchester office.