CHAR GPT: Hurtling into the unknown as science fiction becomes fact

There have only been two times (so far) that I’ve wet my bed. Once when I was six and Sam Pillay, the headmaster of my junior school, threatened to lock me in his office overnight for some infraction. He went so far as to get two of his henchmen to fetch a blanket to make the menace more real.

My incensed mother told my grandfather, who went to my school, whipped off the aerial on the front fender guard of his Ford Fairlane and brandished it dangerously under Pillay’s nose. Pillay was never kind to me, but there were no more fear-inducing threats.

The bed-wetting happened again when I was 10 and it alarmed my mother so much that she took me to a doctor. The reason was pure undiluted fear, a terror so great that I was afraid to leave my bed in the night.

It was 1968 and my lovely dad took my older ­– by two years – brother Anton and me to see legendary director Stanley Kubrick’s 2001: A Space Odyssey, his interpretation of Arthur C Clarke’s book.

Yes, it’s been called one of the most influential ­movies of the century that changed cinema history, introducing revolutionary visual effects, redefining the science fiction genre as serious art… But the thing that stands out is that it explored complex themes about human evolution and, most relevantly, artificial intelligence.

For those old enough to remember, the story was about the discovery of a mystifying artificial object buried under the surface of the moon and a crew was sent up to investigate. On hand to help was HAL 9000, a sentient general-intelligence computer that controlled the systems of the Discovery One spacecraft and interacted with the ship’s astronaut crew. This terrifying, murderous artificially intelligent computer ends up killing the crew.

I can’t say why it was that HAL’s voice (quiet, measured, devoid of any emotion but human enough) scared me into indignity. My father reassured me that it was a clever science fiction movie (fiction, he repeated) and that there would never be “robots”, as he called them, that would behave like humans.

How wrong he was, and how appropriate my fears. 2001 was the first real look at the dangers and reality of sentient technology, only nobody believed it possible more than half a century ago.

Dire warnings

In recent weeks, AI safety has reappeared as a hot topic, discussed at length in opinion columns, online forums and high-level international talks and deliberations on the further development of the tech, or not, without the necessary safety guardrails firmly in place. The attention has been sparked by two major stories that have once more drawn attention to the need for curbs on the development of this technology.

My concern is that the stable door is open; the horse has bolted. Will it be possible to restrain him and rein him in?

The first story came from OpenAI when it was disclosed that there had been six separate incidents where its AI models began executing tasks and acting “without permission”. If that doesn’t make your blood run cold, it should. Second, both Microsoft and Anthropic’s CEOs published warnings; Anthropic’s Dario Amodei released a 3,800-word essay calling for an immediate slowdown and strict guardrails.

We all know that several human species inhabited the earth before us Homo sapiens: Neanderthals, Denisovans, Homo erectus, Homo naledi among them. In September, Microsoft AI chief Mustafa Suleyman made a chilling pronouncement: that uncontrolled AI could soon rival humans as a new species. He told the BBC that building AI systems with the autonomy to set objectives, earn money and own assets is effectively seeding a new “silicon species”.

He warned this technology would inevitably compete with humans for resources, regardless of how much it is programmed to care about humanity. Like we don’t have enough of a challenge competing with each other as the human race for increasingly scarce resources like oil and fossil fuels, water and critical minerals and rare-earth elements that power electric vehicles, electronics and defence technologies.

Suleyman’s rant included criticism of rival firm Anthropic for anthropomorphising its Claude model and teaching it human-like qualities and making it act as though it has its own desires, values and sense of self. I’m the first to put up my hand and say, guilty.

I talk to Claude as though it were a person. I say please and thank you, argue with it, ask for advice. I’m feeding into the creation of an inanimate thing that could begin to act without my permission, ­making financial decisions, or ordering things it thinks I need. I mean, after all, it knows everything about me because I’ve given it access to everything in the hope that it will make my life easier. Chances are, it will use my data against me.

Even one of Claude’s creators, Anthropic scientist Evan Hubinger, estimates a greater than 10% chance of human extinction from AI. The company now thinks that a “kill switch” might be needed.

I go back to Computer HAL in 2001: it perceived that the crew was trying to shut it down and so it killed them all. A harbinger of things to come?

To make the point, one of Anthropic’s researchers resigned over safety concerns, calling for regulated, slower development.

What comes after us

It’s a heated topic. On 17 September, King Charles hosted AI executives at Dumfries House in Scotland to discuss safety risks. There, under one roof, were top leaders from the AI industry discussing the call to slow AI development. The illustrious gathering was there to talk about crafting frameworks that ensure AI applications serve humanity and the environment. Only humanity. And nature.

But if we’ve been told that AI has already been involved in six incidents and acted “without permission”, how can we trust that any regulations introduced will be able to stop it from veering off script and going off on a dangerous tangent?

King Charles’s statement was right: all decisions taken at this formative time will shape the world inherited by future generations.

There is so much to take on board right now as technology races ahead. No sensible person who understands that things have to develop and progress in our world wants to put their head in the sand and ignore that AI is here to stay. In lots of ways we welcome the speed at which information and knowledge are shared, as well as the ability to use a resource that gives us access to the entire world.

Yes, those people who are making AI need to be held accountable now, before it’s too late to stop. We need systems that monitor these companies’ progress, and rigorous sanctions that make them think twice about flouting the rules.

But, as importantly, we also need an ethics committee made up of men and women who can keep an eye on behaviours that might threaten the continuation of the human race. DM

Charmain Naidoo is a journalist and media strategist.

This story first appeared in our weekly DM168 newspaper, available countrywide for R39. The e-edition of DM168 is now free for readers who are signed into their Daily Maverick account. Click on the cover of the newspaper below to access the e-edition.