But on the other hand, a lot of AI talk sounds like hype \u2013 for all the big claims about artificial intelligence, it seems to make basic mistakes quite often, hallucinating false facts in search results, giving people the wrong number of fingers when it generates images, and more. Relying on AI at work has already got people into trouble: last month, a barrister was referred to the Bar Standards Board for using AI in material filed to court.<\/p>\n
Channel 4\u2019s Dispatches decided to put AI to the test<\/a> in the workplace. In Will AI Take My Job?, four professionals are pitted against an AI \u2018clone\u2019 of themselves in a head-to-head contest. A photographer is asked to do a fashion photoshoot for a designer, who judges her shots against AI-generated promotional images. A composer goes head-to-head against AI, both making a custom soundtrack for a documentary about a paraplegic diver swimming with sharks. A trainee lawyer competes with AI to make a claim form in court.<\/p>\n New FeatureIn ShortQuick Stories. Same trusted journalism.<\/p>\n GP v AI\u00a0<\/p>\n Perhaps the most striking contest pits Dr Tom Rustom, a GP from East Surrey, against a customised AI system based on the technology underpinning ChatGPT. Dr Rustom was filmed assessing six patients the accuracy of his diagnoses was compared with those of the automated GP.<\/p>\n \u201cHonestly, I haven\u2019t felt that sort of pressure since medical school exams,\u201d Dr Rustom tells The i Paper of the experience and the moment he learned what he\u2019d signed up for on the show. \u201cI was glad I put some antiperspirant on that morning, because I was very, very scared. And all the permutations of, you know, \u2018What if I lose?\u2019, \u2018What if I misdiagnose somebody?\u2019, \u2018What if I miss something, you know, really obvious?\u2019\u201d<\/p>\n Rustom does not embarrass himself on the show. But he admits he was quite impressed with the AI system<\/a>, which diagnosed a majority of the patients correctly, and was often relatively close even when it didn\u2019t get the exact condition right.<\/p>\n A deadly mistake<\/p>\n \u201cI was intrigued by its ability to make a diagnosis,\u201d he says. But in the case of one patient, he thought the AI made a mistake that could be deadly. <\/p>\n One patient presented symptoms of a stomach ulcer that Dr Rustom thought needed immediate hospital attention. The AI gave a diagnosis very similar to his, but merely suggested the patient take iron supplements. <\/p>\n \u201cThat\u2019s a medical emergency,\u201d he explains. \u201cThat person needs to go into hospital, needs urgent blood tests, possibly IV fluids, and blood transfusions \u2013 they need all sorts of investigations\u2026 If you\u2019ve got an actively bleeding stomach ulcer, I mean, the worst outcome is death. So quite frankly, that one scared me. It angered me a little bit.\u201d<\/p>\n The AI<\/a> was much quicker than Dr Rustom \u2013 it took 25 minutes to assess six patients, while he took 67 minutes, but given the serious risk it missed, he doesn\u2019t feel like his job will be replaced by AI any time soon. All the same, he said he was optimistic about AI\u2019s ability to help GPs do their jobs better, even if it shouldn\u2019t be allowed to make diagnoses any time soon.<\/p>\n AI takes on the law<\/p>\n Dr Rustom fairly obviously won his competition against the AI, but other contests were closer. Charlotte Jaques, a trainee solicitor from Derby, was presented with a real case from a builder who had installed a bathroom for a client, who then refused to pay the \u00a34,500 fee they had agreed via WhatsApp. Jaques was required to put together a case for the Small Claims Court \u2013 a task she estimated would take around three to four hours, while a specialist legal AI <\/a>would do the same.<\/p>\n Both documents were then judged blind by Zainab Zaeem, Jaques\u2019 supervisor at her law firm. Zaeem explains that she was genuinely impressed by both documents.<\/p>\n \u201cAt the time of the experiment, I was skeptical, right? Because I hadn\u2019t tested the AI systems at all,\u201d she explains. But when she looked at the drafts, she thought both were good. \u201cOne was meatier than the other one \u2013 it had a bit more information than the other one \u2013 but they were equally good drafts. Both would be acceptable in a court of law.\u201d<\/p>\n In the blind test, Zaeem eventually picked the draft produced by her colleague as the better of the two, because it would require fewer questions from the judge on the day of the hearing. There was, though, a huge difference in the cost to the client between the two documents. <\/p>\n Four hours of Jaques\u2019 time would cost the builder \u00a31,080 \u2013 almost a quarter of the disputed sum he was trying to claim \u2013 whereas the document generated by AI <\/a>cost just \u00a3120. The \u2018client\u2019 acknowledged it was nice to be able to have a conversation with a human solicitor, but said he would choose to go with the AI<\/a>.<\/p>\n Zaeem said solicitors are highly regulated and are paid for their attention to detail \u2013 pointing out that in America, a case worth millions of dollars was decided over the placement of a single comma. AI, she thought, wasn\u2019t capable of tracking details that closely \u2013 or at least, not yet.<\/p>\n \u201cI don\u2019t believe, with or without this exercise, that we\u2019re in a position where our profession needs to be threatened, purely because there\u2019s a lot of things that the AI yet cannot do,\u201d she says. \u201cHowever, it\u2019s one of those situations where you should watch this space, because the more the AI grows and learns, there may come a possibility when some of the junior staff will no longer be relevant, because AI works a lot quicker<\/a>.\u201d<\/p>\n Creativity put to the test<\/p>\n The cost difference between the human professionals and the AI<\/a> they were competing against were often stark, but nowhere more so than for Jim Hustwit, a composer from Leeds, who was \u201crepresenting the human race\u201d against the AI music service Suno. A soundtrack from Hustwit took a day to compose at a cost of around \u00a3500, while subscriptions with Suno start from just \u00a35 a month \u2013 and it can generate soundtracks in seconds. It was only when he arrived at the TV studios that the scale of what he\u2019d taken on actually hit him, says Hustwit.<\/p>\n \u201cIf I epically fail here, it\u2019s going to reflect really badly on me professionally, but also it\u2019s going to be really bad for my composing colleagues, the rest of the composing world,\u201d he says.<\/p>\n Hustwit\u2019s particular task involved matching music to the exact pictures on screen in a documentary, matching not just the action but the mood \u2013 yes, sharks were approaching divers, but the moment was an affecting one, not a horror scene from Jaws. Hustwit felt he had an advantage here, not least because he had been scuba diving, and no AI can claim the same.<\/p>\n \u201cEvery film will be very different, and every moment will be different, and every narrative will be different,\u201d he says. \u201cSo I think, in that sense, generative AI will struggle because it can\u2019t feel \u2013 so it can\u2019t try things and see how it feels. It will still need a human being to determine that.\u201d<\/p>\n![\"Pictured:](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2025\/10\/SEI_270473495.jpg\")
His AI counterpart didn\u2019t take some medical emergencies seriously (Photo: Rob Parfitt \/ Channel 4)<\/p>\n