Generative AI is changing the world, and will change retail. Some ways are obvious; but it is so powerful and its impact will go far beyond what we can predict today.

The progress of generative AI in the last year may be as significant as the invention of the printing press. Generative AI creates content; ChatGPT is probably the best-known example. Released in November 2022, this chatbot can compose poems, pass legal exams, invent cocktails and plan your holiday. It is particularly good at writing code, and has become a useful tool for software developers. However, it also “hallucinates” – it confidently states things which are just not true. Chatbots like ChatGPT have been primarily trained to write impressive and plausible text, rather than to tell the truth. (Of course, it can also be a human trait to prioritise providing any answer over the correct one).

Image generators, like Dall-E 2, Midjourney and Stable Diffusion have astonishingly good artistic ability, and in a few months have forever changed stock imagery. Need a photograph of a unicorn in front of the Eiffel Tower at sunset? No problem. Music and video are following.

So, for idea generation for marketing campaigns, writing marketing copy and product descriptions, SEO optimisation, or for generating imagery, generative AI is fantastic. It is a hugely powerful aid to creativity. If one wants to see what a dress with waves crashing over it could look like, or design a logo for a new bakery, or create a poster for a festival, or create a planting scheme for a garden, it can quickly generate images and ideas. Virtual changing rooms and augmented reality are already here, but AI will make it much better. AI is changing the way we search the internet; ecommerce websites are currently engineered to work well with search engines like Google – retailers will need to adapt this as consumers search and understand things differently.

Chatbots more broadly will be used for customer service such as dealing with website queries. AI will manage stock levels, and provide personalised recommendations – Netflix already tells us what film we will like next. Pricing can be adjusted, within parameters, to react to competitors.

There is a lot of money at stake and there will be a lot of disruption; there are also a lot of legal issues to consider. Expect 1) lawsuits, 2) investigations, and 3) new laws and guidance from regulators.

Foundation models, like GPT-3 and GPT-4, on which Dall-E 2 and ChatGPT are based, have been trained on a huge amount of data. OpenAI have not disclosed what training data was used for GPT-4, but GPT-3 was trained on, amongst other things, “Common Crawl” – which is a lot of the internet. This means problems relating to copyright and personal data:

  • This creates copyright issues, both in how the AI is trained, and when it generates content. There are already claims saying that AIs were trained on copyright works: Getty Images has sued Stable Diffusion for copyright infringement, and some artists have also brought claims against Midjourney and DeviantArt. Generative AI can also be used to create infringing content. AI can create stories featuring famous characters and in the style of famous authors; images of Buzz Lightyear and Mickey Mouse; and paintings in the style of Warhol and Hockney, so there are many arguments about copyright infringement yet to come.
  • European data regulators are looking at ChatGPT’s use of personal data. The European Data Protection Board has established a task force; Italy’s regulator temporarily had restricted access, and regulators in Spain and Germany are also investigating. Lots of what retailers will want to do with AI will involve personal data.

Aside from legal issues, very eminent computer scientists are concerned that the pace of AI developments poses serious risks for society and perhaps humanity more broadly. The 2018 Turing Prize was won by three godfathers of AI, one, Yoshua Bengio, recently signed an open letter calling for an immediate pause in the development of AI systems more powerful than GPT-4. Another, Geoff Hinton, has resigned from Google Brain so he can speak freely about his concerns.

Legislators across the world are playing catch up in considering how the law needs to change.

The EU’s AI Act is currently going through the European Parliament, and is evolving significantly as it does so. Its requirements include stringent obligations on “High Risk AI”, including implementing Risk Management Systems; lawful, relevant, representative, error-free and complete training data sets; human oversight and transparency. There is debate at the moment as to whether generative AI will be classed as “High Risk”. Most obligations are on the providers, but retailers should take note of the fundamental principles and consider their own implementation. For example, consider whether you should be telling consumers when you are using AI generated content, especially for marketing campaigns.

The UK government has published its own white paper, proposing less regulation and a “common sense” approach, but without any specific details yet.  With no new specific regulator proposed, the current plan is that UK businesses will need to answer to existing regulators, including the data regulator, the ICO.  The ICO has taken a more business friendly approach than its EU counterparts on some issues, such as international data transfers.  However, retailers should not wait for guidance; there are steps they should take today, including implementing a policy on permitted uses of generative AI.

To find out more about Browne Jacobson and the services they provide to the retail industry, click here.

This article was also published in The Retailer, our quarterly online magazine providing thought-leading insights from BRC experts and Associate Members.