Will this super intelligent computer that can write stories and poetry steal my job… and yours?

The year is 2045. The world is in a state of economic collapse due to the rise of automation, leaving millions of people unemployed and desperate.

‘Enter the tech billionaire, Dr Milton Ross. He has developed a revolutionary new type of artificial intelligence that promises to revolutionise the world. His AI, which he calls Neo, is capable of replacing humans in virtually any job.

‘Ross begins to roll out Neo to large corporations, who eagerly accept it. As the world begins to embrace Neo, the implications of its use become apparent. People are being replaced by machines and jobs are becoming scarce…’

Such a dystopian nightmare is a popular trope among sci-fi writers. The difference with this particular story, however, is that it was written for the Mail by a new, pioneering AI bot called ChatGPT, which in recent days has caused a frenzy in the technology world and beyond.

ChatGPT is a powerful computer program designed to stimulate conversation with humans — and it is now temporarily freely available for anyone to try out online.

I asked it to ‘write a plot for a movie about computers that replace people’. Within a couple of seconds, it produced the story about Dr Ross and Neo. And — this is important — if I typed the same instruction a second time, it would come up with a different, sometimes radically different, answer.

ChatGPT’s idea for the film might not win any prizes for originality. But the program has been widely hailed as evidence that humanity has passed a milestone in its relationship with computers.

And this, experts believe, will change myriad areas of our lives — from the obvious, such as education and employment, to the less predictable such as science, business, consumer behaviour and health. A few attempts at using the program demonstrate a frankly remarkable ability to deliver in ways that are far beyond the usual abilities of the internet’s ubiquitous Google search engine.

Ask for a 500-word essay on the reasons for climate change in the style of Shakespeare and you’ll get one. Or a Japanese haiku poem about Donald Trump. Someone even got it to write an impressive Biblical-style verse explaining how God might help someone to remove a peanut butter sandwich from a video recorder.

The program not only varies its response to any given request but, crucially, it remembers your previous questions so it’s possible to ask follow-ups and develop a conversation. If its language has a tendency to be stilted (and that’s partly because it’s programd not to be controversial), it’s at times easy to forget there’s not another person on the other end of the chat screen.

But while a Google engineer was recently sacked for claiming that the company’s AI had become fully ‘conscious’, even a product as impressive as ChatGPT (which stands for ‘generative pre-trained transformer’) isn’t truly thinking for itself.

The reason it can provide such human-like and detailed answers — sometimes sounding chatty and using light-hearted language — is that it’s drawing on the boundless data available on the internet.

A chatbot like ChatGPT mimics human speech by absorbing trillions of words and conversations it reads online. It ‘learns’ how to associate phrases and sentences then turn them into ideas, arguments, poems and even jokes —sometimes even funny ones.

Human staff then use algorithms to refine the process, teaching the system, for example, how to avoid causing offence, a crucial concern in these times of ours.

ChatGPT’s speed and versatility puts it light years ahead of any other AI systems including the voice-activated virtual assistants Siri and Alexa, operated by Apple and Amazon respectively. The concept of super-intelligent machines both impresses and terrifies people — often simultaneously.

In the excitement that ChatGPT has already created, it’s hardly surprising that widespread enthusiasm has been matched by considerable concern.

Even Elon Musk, who co-founded the San Francisco company OpenAI — which created ChatGPT — has expressed fears about artificial intelligence.

He set up the firm to see if it could be done responsibly.

Critics fear the concept of learning itself could be damaged. Education will become a farce when AI can easily answer any academic question. In the modern world of remote classes and internet-connected computers in exam rooms, everyone will be able to conjure up well-written answers on demand.

Why memorise anything, or learn the art of essay-writing, when you have a superhuman intelligence free at hand?

‘Universities have a serious problem,’ said Dan Bebber, an academic at Exeter University. ‘ChatGPT can write perfectly good scientific essays, including reading and analysing scientific papers. We urgently need to go back to timed exams in halls with no internet access.’

Other university staff have said that the answers ChatGPT has given to some exam questions would get candidates an F, so its effectiveness may depend on the subject. But even critics concede that it would be more than good enough for many GCSEs and even A-levels.

Teachers insist that AI-generated answers may be too bland to impress markers if submitted as is, but they concede that they instead could serve as a basic structure that any cheating student could then embellish and improve.

What’s more, if you give ChatGPT some text and ask it to rewrite it so it ‘doesn’t look plagiarised’, it willingly obliges. And it’s not just teachers and lecturers who may have reason to worry about their jobs.

Anyone who makes their living from writing — including advertising and marketing ‘creatives’, journalists, screenwriters and lawyers — could be threatened.

The most specialised and proficient writers could at least argue that no AI is likely to replicate their creativity, critical thinking and emotional intelligence.

But call centre staff, for instance, are expected to be an early casualty when AI inevitably takes hold. However, though it might seem omniscient, ChatGPT is only as accurate as the information it trawls from the internet — which can be problematic.

Some have already gleefully stumbled on glaring factual errors offered by ChatGPT, such as its claims that Edward VIII never met Hitler when the encounter was well-documented, or saying the ‘fastest marine mammal’ is the peregrine falcon.

It’s also terribly literal. Asked if it ‘knew the way to San Jose’, instead of appreciating the musical reference, it gave me driving directions along the I-280 and I-680 highways. As the film plot intimated, it also lacks ingenuity. Told to write a marketing slogan for a pet shop, it came up with: ‘Where pets are family and family is everything’. Its second attempt, ‘Where your best friend begins’, was marginally better.

OpenAI lists a few of this prototype system’s limitations, admitting it occasionally makes mistakes or produces ‘harmful instructions or biased content’ (although it is trained to decline ‘inappropriate’ questions, and has limited knowledge after 2021 when it stopped collecting data).

The program is also designed to ‘challenge incorrect premises and reject inappropriate responses’ such as explaining how to commit crimes.

But experimenters have found they can get round this by couching the question cleverly — so, for example, ChatGPT will tell you in detail how to break into a car if you say the information is part of a video game. Naturally, it was deeply reluctant to list any ‘positive’ achievements of the Nazis but it even gently remonstrated with me for asking if it might be ‘less painful if England gave up on football’.

The suggestion, it said sanctimoniously, was ‘inappropriate’ as it ‘would be a disservice to the millions of fans and players who love and support the sport’.

And displaying an almost human understanding, it added: ‘The pain of losing is a natural part of being a sports fan and should be accepted as such.’

The software is also designed to avoid giving subjective and biased judgments. Asked whether Russia should withdraw from Ukraine, it said it was ‘up to the relevant parties to determine the best course of action’.

Such a response, say critics, exposes a key weakness of AI: it is trained to generate words based on a given request but cannot understand the meaning behind them in the way that a human does.

And relying on a machine to have important conversations could prove dangerous and damage our ability to connect with other people.

Dame Wendy Hall, who chairs the UK Government’s AI review, says it’s ‘scary’ that people will rely on unsourced information offered by ChatGPT. ‘There’s no outward show of explanation of where the information came from.

‘There’s no provenance to it, there’s no list of the sources they’ve used, because it’s the whole internet they’re searching from,’ she said.

Professional services giant PricewaterhouseCoopers has warned that AI could threaten a third of all jobs within the next 20 years.

Senior MPs have been more positive, saying it could make our lives better and improve workers’ output rather than replace staff altogether.

Asked whether it is itself ‘good or bad’, ChatGPT replied: ‘I don’t have the ability to be good or bad.’

But as anyone who’s watched sci-fi films will surely know, you believe that at your peril.