An opinion piece in today's Sunday Times. And it's the national edition, and not just the Scottish edition, as far as I can tell - it's by Dominic Lawson, who's Nigel's son. And if you know that particular fact, then like me you're probably showing your age a bit
Anyway, it was only in the last couple of days I noticed the word 'hallucination', which I hadn't seen before. At least, not in the context of AI
But in the past few days, I've been seeing it all over the place, and not just in the context of the Sandie Peggie case.
Justice is meant to be blind, not to hallucinatehttps://www.thetimes.com/comment/column ... -cqfsc627hThe judge in the Sandie Peggie tribunal is suspected of using AI. Both police and courts are befuddled by this technology: it is no laughing matterHallucination, they call it: the innocent term used to describe entirely fictitious answers given by generative artificial intelligence tools. This is generally regarded as a source of (instructive) amusement. Recently Jimmy Wales, co-founder of Wikipedia, explained what happened when he asked ChatGPT: “Who is Kate Garvey?” In the real world Garvey, who worked for Tony Blair in Downing Street, is married to Wales. But ChatGPT came up with a string of hallucinations, one of which was that “Kate Garvey is married to Peter Mandelson”. When Wales responded to ChatGPT: “Isn’t Peter Mandelson quite famously gay?”, the tool chided him for inappropriate speculation about the peer’s sexual orientation.
Wales has a dog in this fight. ChatGPT, to some astonishment, overtook Wikipedia in monthly visits this year. Still, I thought I’d try the same sort of query, asking it who Rosa Monckton is. As far as I know, she is my wife, and in the professional world launched Tiffany & Co in the UK. But on one of my visits ChatGPT informed me she was the “co-founder of Pimlico Carpets”. Er, no. Although to be fair to ChatGPT, she may once have bought a carpet in Pimlico. On another visit it asserted that “Rosa Monckton’s full name is Lady Bowes-Lyon (after her marriage)”. So, either my wife is bigamously married into the family of the late Queen Elizabeth the Queen Mother, or I am, without previously being informed of it, a scion of the Anglo-Scottish aristocracy.
But such hallucinations cease to be amusing when they pollute the legal system, as a result of laziness — or worse — on the part of professional practitioners. Over the past few days there were two apparent examples, in cases of high public importance. The Sunday Times had already discovered how West Midlands police misled parliament over the evidence for its decision to ban supporters of Maccabi Tel Aviv from attending last month’s match against Aston Villa. Now we learn that WMP’s dossier, revealed to MPs, included a reference to a match Maccabi Tel Aviv played “against West Ham in the Uefa Europa Conference League group stage on November 9, 2023 … at the London Stadium”.
No such match took place. The Birmingham-born MP Nick Timothy pointed out: “The match was a fiction. It was a hallucination by AI.” Not true, says the WMP chief constable, Craig Guildford. While admitting it was “completely wrong”, he said it was a result not of the use of AI but of “social media scraping”. However, he provided no source. In any case why should the police, of all organisations, require “social media scraping” to discover when Maccabi Tel Aviv last played a match in this country? And we know WMP are assiduous users of AI, from a press release they put out in September boasting how “we updated our technology to unlock the power of AI”.
Last week an employment tribunal in Scotland ruled on claims of harassment under the Equality Act 2010 by an NHS nurse, Sandie Peggie, who had objected to being expected to share a changing room with a trans woman — that is, a biological male — called Dr Beth Upton. Judge Alexander Kemp cited an earlier tribunal case involving Maya Forstater, which, he wrote, had stated that the Equality Act does not create “a hierarchy of protected characteristics”. Forstater, now head of the charity Sex Matters, tweeted, definitively: “This ‘quote’ from my judgment doesn’t come from my judgment. It is completely made up.”
Michael Foran, an expert in this field of the law at Oxford University, declared: “The [Peggie] judgment included supposed quotes from specific judgments that do not appear in those judgments. That in itself is extraordinary.”
The Scottish Courts and Tribunal Service has guidelines stating it is “not using or considering the use of AI in relation to any form of decision-making”, for the good reason that AI tools may, according to the UK Courts and Tribunals Judiciary, make up “fictitious cases, citations or quotes, or refer to legislation, articles or legal texts that do not exist”. Kemp was obliged to reissue his judgment, with the fake citations removed but with no admission that AI use was the cause, referring instead to “clerical mistake[s]”. Bull.
The starkest example of what is beginning to happen in the courts came a couple of months ago, when Judge Mark Blundell exposed in the upper tribunal the conduct of a barrister, Chowdhury Rahman, who had presented an appeal on behalf of clients turned down for asylum by the first-tier tribunal. Rahman claimed the original judge had made an error in law. When Blundell examined these submissions he found this lawyer had cited cases which were “entirely fictitious” and that Rahman “appeared to know nothing” about any of the authorities he had cited, some of which “did not exist”.
Blundell concluded: “It is overwhelmingly likely, in my judgment, that Mr Rahman used generative artificial intelligence to formulate the grounds of appeal in this case, and that he attempted to hide that fact from me during the hearing.” On his chambers’ website Rahman is described as having “a unique judicial mind”. One would hope his approach is unique, but I fear not.
In the US, the home of AI, there have been hundreds of cases of such legal misdemeanour (to put it politely). The US technology publication 404 Media has analysed many of those in which “lawyers have been caught using generative AI to … generate fictitious citations [and] false evidence … that ultimately threaten their careers”. It tried to talk to all the offending lawyers, with some difficulty. One claimed his misuse of ChatGPT came about because “a serious health challenge … causes me to be dizzy and experience bouts of vertigo and confusion”. Now why didn’t the West Midlands chief constable think of that excuse?
More seriously: we know that even the latest iterations of generative AI can have a hallucination rate of about 35 per cent. Rather like a hugely sophisticated version of auto-complete, this technology is designed to predict plausible text patterns, not to know facts. It is very useful if the person employing it is expert in the field under investigation, who would sense if the “answer” was fishy; it is treacherous and possibly career-ending in the hands of the ignorant or uncritical.
The astounding sums now being plunged into it by the Silicon Valley behemoths (not billions but trillions of dollars) has been compared to the biggest capital expenditures of the 19th century — the railways. This was described as the “railway mania”: there was a vast investment bubble, and although many fortunes were lost, it was also transformative.
However, while much of the track was built speculatively, it was never the case that a third of it, or more, took passengers to fictitious destinations.
Login
Register
FAQ
and the popularity of Ola in Scotland 
