A big retail trade association has formally asked Europe’s technology chief to rethink its forthcoming Artificial Intelligence Act, which will apply to AI technology, particularly as it relates to advertising. The European retail association Eurocommerce, which represents industry giants such as Amazon, H&M, Inditex and Ikea, has written to EU tech commissioner Henna Virkkunen asking for an exemption from the disclosure requirements of the AI regulation for AI-generated marketing content unless it is specifically created to deceive consumers.
The letter, which was seen by Reuters, questions the principle of where to draw the line between fitting the use of artificial intelligence in trade and that of content which needs particular measures of transparency. Eurocommerce director general Christel Delberghe made it clear the industry believes that AI-generated ads which are not meant to deceive should not fall under the definition of ‘deep fake’. The difference, the retail association argues, is important to guarantee that the regulation will not stifle the adoption of AI by businesses transitioning to such technology in responsible ways.
This request follows weeks before the disclosure requirements of the AI Act come into effect in the EU. The new rule will require businesses to plainly identify and label content created or AI-enhanced by AI, especially when it is defined as a “deep fake. But retailers say this sweeping interpretation doesn’t make a distinction between a misleading content and more common commercial images that consumers will still recognize as a tool for marketing.
In practice, the regulation, as it stands, has a significant impact on the retail industry. The major retail giants have already embraced AI in their marketing strategies, and it’s changing the landscape of creating visual content. The possibilities of AI have been so appealing in the eCommerce sector because of the efficiency that German online retailer Zalando has shown in cutting down on the content production rate by a staggering ninety percent with the help of AI technology. Likewise, H&M and Zara, two fast-fashion brands, have started using AI-powered facelifted models, which raises intriguing questions about authenticity in the fashion industry but does not necessarily amount to deceptive practices.
Rather than compliance costs, the concern from retail point of view is its impact on doing business. Retailers could find themselves having to label a huge proportion of AI-supported content if the regulation is extended to all AI-generated or AI-edited advertisements, warns the Eurocommerce letter. Delberghe said as much when he proposed that the labeling be limited to just those products that contain the active ingredients that make them effective. This is a well-developed argument that addresses a real consumer protection policy problem: warnings or labels that become ubiquitous can become ineffective at conveying information to consumers.
The retail association’s request also comes amid a wider discussion of the potential for AI to be regulated in a manner that safeguards consumers and doesn’t inhibit innovation. The question of what constitutes “where” the limits of the AI Act are is not exclusive to retailers. In various industries, businesses attempt to learn how this regulation will affect their daily operations, especially in the areas where AI is utilized for routine commercial uses, rather than possibly deceptive applications.
The difference the retail sector is requesting regulators to draw is not without some logic. That’s a difference between using AI to produce an image of a living room to display a sofa and producing a deep fake to manipulate public opinion or deceiving consumers about what a product can do. In the first case, consumers are aware that they are viewing a commercial representation that is intended to be used for advertising purposes. In the second, the purpose is deceit, and they call for special measures of transparency.
But the difficulty for the regulators is designing regulations that can effectively differentiate between these various uses without allowing the bad actors to find loopholes. An overly general exemption may lead to a failure to meet the objectives of the AI Act in terms of consumer protection. If it is too narrow, it could prevent legitimate businesses from going about their lawful activities without undue compliance costs, stifling innovation.
The retail sector doesn’t want to be completely exempt from transparency rules, but expects a more nuanced approach that would restrict exceptions to what is actually misleading content, not simply all AI-generated images. This stance accepts that there is some AI produced content which should be clearly marked, but claims that the current definition of a “deep fake” is too broad to act as a good regulatory measure.
The questions consumers have in the context of this debate are also important. How would consumers react to seeing a product where all the ads are AI-generated and they were told so? There is evidence that consumer behaviour can become desensitised to warnings that are overstated or too common, and that this can reduce their effectiveness.
The European Commission has yet to comment publicly on the retail association’s request, but the topic is likely to be hot in the days ahead as the deadline for implementation looms on August 2 approaches. Watch will be keen to see if regulators will be willing to implement changes to the disclosure obligations of the AI Act, or if they will stick to the status quo for consumer protection.
