Understanding its limitations will allow marketers to harness the opportunities.

Artificial intelligence is generating a lot of commentary right now, especially with the advent of more accessible versions such as Chat GPT.

In a way it’s reminiscent of many technological changes that have come before, where an immensely useful technology is hyped far beyond its actual usefulness, before settling down into doing what it is actually best at. Anyone remember blockchain hysteria – at its peak about 5 years ago?

AI is an incredible technology, capable of recognising and replicating staggeringly complex patterns in colossal datasets. It develops this skill by being trained using datasets of the same type which are tagged and targeted to allow the AI to learn the rules that apply.

This ability to apply complex pattern recognition to massive datasets means that AI is making a huge difference to diagnostics, drug development, engineering and probably will to many other things that we haven’t yet thought about.

The interesting thing is, that the skills that make AI brilliant at this type of work, also make it look like it’s brilliant at lots of other stuff. As Jason Lanier, interdisciplinary scientist at Microsoft aka the ”Godfather of virtual reality” or “the Dismal Optimist”, points out in a recent Guardian article, we call it artificial intelligence, but it isn’t really intelligent.

Simulated intelligence would perhaps be a better term.

A limitless opportunity? Not quite… or, perhaps, not yet.

A competent creative brief from a human user can certainly produce interesting results from both an imagery and written word perspective. But this is where the limits of AI become apparent. The AI recognises data as data. The AI’s output is correct so long as it holds true to the patterns and rules that it has learned. It has no clue about the representation of that data in its real-world form.

Once we get beyond factual writing and illustration, the meaning of words and images themselves is often an abstraction. Our reaction to them is often more about how they make us feel, what memories they stir up or what they cause us to imagine than the components of their content. Which is why a lot of AI creative output, no matter how unusual, feels flat. A huge amount of our frontal lobe gives us the power to put ourselves in the place of others, to imagine how they might feel. It also gives us the power to imagine how other people would react to our actions, words and things we create. An ability no AI possesses.

If briefed to do so by a human, an AI can create an image of Donald Trump riding a walrus whilst eating a burger in the style of Van Gogh. Amusing for sure, but it has no way of comprehending what the reaction from a human being to the image might be, or even what the image represents. All it has done, in reality, is to faithfully take the patterns of data that it recognises from its training as images of Trump, a burger, a walrus plus images created by Van Gogh, and combine them into one single dataset which we then see as the requested image. This amusing ability is a side effect of what AIs are good at, even though, at the moment, most commentators seem to be focusing solely on this aspect.

Apply human expertise to artificial intelligence and you’ve got a powerful marketing tool.

Now, there’s another side effect which is just as interesting and possibly more useful. Using AI to collect information may free us (at the moment anyway) from the almost unnoticed restrictions that social media and search algorithms use to filter what we see. Ask an AI to summarise information on any particular subject and it is likely to give a different answer each time. This element of randomness mean we have to use our judgement when choosing between the options.

Going back to more core uses, particularly in marketing, AI is already proving very helpful by picking out patterns in customer purchase and behaviour data, looking for trends in market and customer quantitative data etc. Some interesting uses are emerging which involve facial and expression recognition. These are designed to look for emotional responses in video of market research respondents as they are being exposed to various stimuli. There may be applications for taking a master campaign and creating multiple executions in different formats and languages… ready for QA by a human! After all, nobody wants to see a toothpaste ad where the model has two rows of teeth, in the style of the Xenomorph from Alien. Take a look at some AI-generated imagery of human faces and you might be very surprised at how often this actually happens.

Using AI for commercial creative purposes may have more serious consequences than xenomorphs obsessed with dental hygiene. Predictably, several lawsuits alleging breaches of intellectual property rights have already been launched, one notably by Getty. It seems likely that regardless of whoever or whatever does the scraping of the source material, it will likely be the final user who is found liable for any breaches of IP rights legislation.

It is, of course, fascinating to discuss the future AI-related demise of poets, artists, copywriters, and art directors but to spend too much time doing so risks missing out on the real opportunities offered by what AI is good at right now. Whilst, of course, still having fun with the side effects.