Smoking gun of useless AI exposed a fake Maumela bio

I had an extraordinary conversation with Gemini this week. For those who haven’t met Gemini – and if you are part of the Google ecosystem, you will surely have – it is the global search engine giant’s artificial intelligent (AI) bot… or perhaps persona would be a better word.

I stay away from AI as a general rule – and never use it for writing (it still feels as though that would be plagiarism) – but on this occasion, I needed to check if a picture posted on the internet and purporting to be that of the alleged Tembisa tender looting kingpin Hangwani Maumela was correct.

Most news sites and social media platforms were running the image – and it seemed to fit because it was of a youngish dude showing off in a Lamborghini branded T-shirt.

That made sense because he apparently has a fetish for the Italian-German supercars – as shown when the Special Investigating Unit seized a few of them from his Joburg mansion last year.

One of the links provided by Google was to an alleged biography of Maumela and it immediately raised the cynical gooseflesh in my journalist’s brain.

Everything written about him was more than biographical, it was hagiographical. He was, it claimed, a simple lad from Limpopo who worked hard and made it good, getting law qualifications and awards and finally being made a judge of the high court.

Stretching over two or three pages, the bio was fulsome in its praise in the main – and extremely sparse on details. No dates, no names of other people, no photos, nothing to verify the content or enable a cross-referencing of the claims.

It looked like either he – or someone close to him – had deliberately set it up (probably with the help of AI) to make him look as good as possible for the future when search engines or LLMs (the large language models on which AI is based) come looking for information.

So, the logical thing for me to do, given this dreck had surfaced in a Google search, was to ask its AI arm, Gemini, if this was not a scam?

Oh yes, said the bot – it definitely was fishy. But – and this was the part which stunned me – what happened was not as the result of devious human prompting, it was because this is the way AI programmes scraped the Net when used by fake bio sites.

How could that happen without specific prompts? Gemini said it was simple: AI had seen the reports of Maumela linked to a court judgment (a finalised seizure had been issued earlier in the week) and had constructed the rest by itself.

Without any mention of Tembisa? Yes, said Gemini. Happens all the time.

If that is indeed the case, then it is the smoking gun of AI incompetence. No way a human searcher – or even the Google of old – would have put something like this together without the other juicy details about Maumela which are freely available on the Web.

My weird interaction happened as the world was digesting the news that a ChatGPT bot had “gone rogue” – without any human intervention – and hacked into another, open-source, platform used by AI developers… and that mass destruction had only been averted by people on that platform roping in a Chinese AI source to ward off damage.

I only hope that when the anaesthetic has knocked me out and the surgeon is hauling out the scalpels, he or she hasn’t been trained by AI.