Last week I visited Romania for a good friend’s wedding. While wandering the streets of old town Bucharest, admiring the mixture of Hanuri, French and Soviet architecture, I was met with jarring images of AI-generated advertising. Malformed hands around a Heineken bottle, multiple people who look eerily perfect and almost similar, well within the uncanny valley. I’m sure this is not unique to Romania, but it did spark some thought into the knock-on effects and parallels in the other facets of humanity.

By now you’re likely familiar with the common sentence structures, punctuation, and adjectives/adverbs used by AI. Whether it be an over-use of em dash “—”, it’s not X, but Y sentence structure, or describing every idea as something that’s “quietly” occurring. These types of posts plague Linkedin and Reddit. I would argue that this is at the detriment to the platforms and user experience because it’s hard to know if the idea is genuine or a half-baked AI hallucination.

Recently, it was brought to my attention that something interesting is happening in the realm of fictional literature, best described through the introduction to a man named Elias Thorne.

You see, fictional writing has been seeing a rise in protagonists of this name (and a few variations of). Before AI, he wasn’t a known protagonist at all. How has this occurred?

The answer is in the data. It seems that when the AI labs attempt to scrub copyright content within their training data, they replace recognisable character names with synthetic stand-ins e.g. Harry Potter is changed to another name selected from a pool, and Elias Thorne has been over-used as a stand-in. Now when something like ChatGPT tries to write a story about a clockmaker or lighthouse keeper there’s a high statistical probability they’re named Elias Thorne. Thus, our fictional reading experience is now suffering too.

These are excellent examples of how the incorrect use of AI leads to a convergence to mediocrity. Many people do not realise that when you prompt an LLM it will provide you with the most probable result according to its training.How does this translate to businesses?

While these systems remain genuinely useful for compression, extraction, and large-scale grunt work, using them for creativity or final products will never make a company stand out. We see that the most successful AI implementations rely on human wisdom, experience and unique creativity layered on top of AI insight/generation. To use it otherwise is the quickest way to join the middle of the pack and thus degrade the experience of your end users.