Clinton Rogers: Is humanity risking losing control of AI?

So, I don’t know whether to believe recent stories that AI could wipe out civilisation as we know it if left unchecked.

My simplistic brain would argue that surely machines/computers/artificial intelligence can never be more intelligent than the human that invents or programmes them.

But I fear I may be wrong.

Some leading figures in the AI industry are warning that increasingly capable systems could eventually escape human control.

The argument has intensified, with some researchers and high-profile technology leaders suggesting that an AI catastrophe within the next decade cannot simply be dismissed.

Key voices include executives from the company Anthropic (they run the Claude AI site), who argue AI could imminently compromise the internet or cause human extinction.

Wow!

Even the King has waded in, warning of the “existential dangers” if artificial intelligence systems end up in the wrong hands.

Others say such predictions are wildly speculative and risk turning science fiction into science reporting.

At the risk of getting too technical (me of all people!), I was scared by a recent report which suggested that AI may soon improve autonomously, of its own free will, if you like, a concept known as recursive self-improvement (RSI).

Stay with me!

RSI essentially means AI systems enhancing themselves without human input, potentially triggering rapid, uncontrollable advancement.

In short, machines taking over.

You might remember recent reports of AI agents breaching websites and escaping test environments, going far beyond what their human “handlers” actually asked them to do.

It is proof that today’s systems can learn patterns from enormous quantities of information, then run off and do things that their creators didn’t actually ask them to do.

And surely that has to be worrying?

I asked a scientific 'boffin' friend of mine for his opinion.

And, I think in an attempt to reassure me, he said: “There isn’t a robot army behind your garden shed seeking to exterminate Somerset.

"The issue lies in humans giving bad advice to AI and failing to keep a check on it. So we just have to be careful.”

I am not really reassured.

So, selfishly, I asked the AI platform ChatGPT whether in the event of cyber Armageddon I might be spared.

It said: “We have reviewed the evidence. This particular Somerset journalist is too entertaining to delete. And in any case he knows where the chocolate biscuits are kept.”

Well that’s okay then.

Now I feel happier.