In 2016, AI pioneer and Nobel laureate Geoffrey Hinton argued that there was no point in training radiologists, since it was “obvious” that AI would outperform human radiologists within five years i.e. by 2021. As pointed out in an article in Radiology Business, the number of radiologists employed at the Mayo Clinic actually increased by 55% since that prediction in 2016. This is not an isolated instance. The US Bureau of Labour Statistics projects that healthcare employment will increase from 2022 to 2032, outpacing the average growth for all occupations. The picture may turn out to be nuanced. A large study published in September 2025 in the European Journal of Radiology Artificial Intelligence concluded that AI will improve the productivity of radiologists. However, it suggested that these gains will mostly accrue to employers, AI vendors and private equity firms that operate many US healthcare organisations. Radiologists themselves will most likely come under salary pressure as AI takes over more of their role, with job roles changing rather than disappearing. The New York Times recently asked Mr Hinton about his ill-fated prediction. An Article in Radiology Business reported in May 2025 that Mr Hinton conceded that he was wrong about timing but not the direction AI would take the speciality.
There is no shortage of wild predictions about AI that have proved unfounded. AI Pioneer Marvin Minsky predicted in 1970, “In from three to eight years, we will have a machine with the general intelligence of an average human being.” We are now well into 2025, and there is no sign of this so far. We may be closer, but there is no AI today that approaches the intelligence of an ant or a squirrel, never mind a dog or a dolphin. Elon Musk said, in 2019, that “I feel very confident predicting autonomous robotaxis for Tesla next year.” In fact, Tesla started a limited trial in Austin, Texas, only in June 2025. To be fair, Mr Musk is noted for his optimism, having also said “Human missions to colonise Mars will start in 2024”. Mr Musk was not alone in his optimism for autonomous cars, such as the prediction in The Guardian newspaper in 2015 that autonomous cars would take over by 2020, allowing us all to be backseat drivers. Five years after that date, and we have some trials of self-driving taxis, from San Francisco to Wuhan, but we are a long way off from a car that can drive itself in a completely autonomous way in unrestricted road conditions known as “SAE level 5”.
Professor of theoretical physics Michio Kaku famously stated, admittedly some years ago, that “our most advanced robots have the collective intelligence and wisdom of a cockroach, a mentally challenged cockroach”. LLMs can recognise patterns, but do not demonstrate adaptive intelligence, self-awareness, abstract reasoning or emotional intelligence. This does not mean that they are not useful: a pocket calculator can multiply much faster than you, which is useful, but no one is giving IQ tests to a pocket calculator.
Even the fast-growing AI industry itself is subject to wildly overoptimistic projects. Analyst firm Forrester predicted that the AI market would reach $1.2 trillion by 2020, when in fact it was $17 billion in size that year, an error in prediction that was off by a factor of over 70. In 2024, the AI market was worth around $279 billion; it is certainly growing quickly, but linear projections of growth in this market are tricky. We are already seeing signs that AI may not be quite the miracle of productivity that was promised. A large MIT study found that 95% of AI projects failed to show any form of return on investment. Researcher Atidya Challapally of MIT Media Lab said: “When we spoke to executives, they would often say that the internal tool was very effective, but when we spoke to employees, we found zero usage”. The MIT study is not alone. The Financial Times reported in September 2025 that US company executives frequently reported on earnings calls that they were using AI, but were unable to explain whether any benefits had occurred from this investment. The clearest benefits that the FT found were from companies building data centres for AI vendors. It also noted that such statements in annual reports were increasingly highlighting risks associated with AI, especially around cybersecurity. The future growth of the AI industry may not look like a straight line heading up and to the right, despite the enormous sums of money currently being invested in it.
As Nobel laureate physicist Niels Bohr said, “predictions are hard, especially about the future”.
