On Centre CourtThe main court at a tennis centre, especially Wimbledon in London. at WimbledonAn international tennis tournament held in London every year., defending champions Novak Djokovic and Elena Rybakina will face stiff competition to hold onto their hard-won silverware. But up in the commentary box, a very different match will pit human against machine.

An AIA computer programme that has been designed to think.  chatbot trained in the "unique language of tennis" will give running commentary and analysis on each of the games. WatsonX uses data gathered on and off court to offer insights into player performance and potential.

The bot is run by IBM, the company behind two historic moments in AI development. In 1997, Deep Blue was the first computer to win a game of chess. And in 2011, Watson won $1m (£786,000) on the US gameshow Jeopardy!, beating two human contestants.

But the latest "generative" AIs are far more powerful. They learn from vast quantities of online data to produce original content, from stories and songs to computer code. Since its release last year, the popular ChatGPT super-bot has written hundreds of books, given legal and financial advice, and prepared a church service.¹

Its success and popularity have surprised its creators. "We are a little bit scared," says OpenAI boss Sam Altman. In May he signed an open letter warning of the "risk of extinction" if AI is not controlled. But experts disagree on how it will shape our future.

Optimists believe AI can free humanity from work to pursue our dreams. Like the printing press and the Internet, these smart bots make human knowledge more accessible to more people.

As a turbocharged research tool, AI can "unlock cheap, sustainable energy, solve many problems of human health, and deliver better, more efficient agriculture," says philosopher Stephen Cave.

But some computer scientists warn that AI may destroy itself first. Current bots learn from human-generated content. But future AIs will train on an Internet full of machine-made data, littered with errors, biasesBeliefs that systematically and unfairly distort a person's decisions, in favour or against one group or another. and fabricationThings that are made-up. .

Each generation of chatbots will "amplify" these mistakes, says expert Ilia Shumailov. In his research, a language model went from fluency to gibberish in nine generations.

Of course, incoherentNot spoken or written clearly. bots are unlikely to get anywhere near the Centre Court commentary box. But will WatsonX still be a reliable Wimbledon host if it reads machine-generated sports journalism elsewhere on the Internet?

Current AIs "hallucinate", confidently spouting out phantom information and invented facts in order to be helpful. Critics fear that more powerful models will learn worse habits - lying, stealing and even harming humans - to fulfil their tasks.

Computer scientist Alex Dimakis says future technology will get better at "filtering" true and false data. This may help eliminate both human bias and machine hallucinations.

But AI expert Marc Warner warns that we do not know how to teach them common sense. "Given enough time", he says, it is "very likely" we will create AI with "generalised" intelligence. Capable of doing everything we do, but better.

And if it does not share our interests and values, we may have a serious problem.

Yes: Computers do not experience the real world as we do. So they cannot test whether their information is aligned with reality. The more misinformation they consume, the less sense they will make.

No: Humans have developed the skills to identify false information and learn from each other. Bots struggle to do this at the moment but are developing quickly and will soon overcome these obstacles.

Or... If computers get better at making human-like content, we may need more powerful AI to identify what is real. By that stage, would computers have a better sense of reality than ourselves?