The news that two healthcare trusts in London are to experiment with a system to look up symptoms by text message, to triage the kind of non-urgent queries at present handled by the NHS 111 service, raises many questions. They may not seem urgent when people are dying in the corridors of an NHS hospital for want of money, but in the long term they are just as important. Some are purely medical: is this an area that requires the attention of a human being, or is it one where purely factual answers will suffice? When will this project start using artificial intelligence? Some have to do with the way that the NHS is being privatised around the edges in ways that disadvantage the central public parts of it. Widest of all is the general question of the automation of brainwork, which might have effects quite as gigantic as the replacement of manual labour by technology has had.
Two kinds of claims are made for AI in medicine. The weaker and more plausible is that it can automate the processes where no judgment is required, only the clear and consistent following of well-understood rules. This kind of thing is what the 111 service is supposed to do: the question that it answers is not “what’s wrong with you?” but “do you really need to see a doctor?” Some triage is necessary in any healthcare system, and the present system in the NHS is under huge and growing strain.
But the wider claim of healthcare automation is that there will be systems that can augment and eventually replace the judgment of trained human beings. The hope is that deep analysis of unimaginable quantities of data will yield reliable knowledge superior to anything that unaided humans can produce. The placebo effect is important in medicine, and people who believe they are being treated by doctors who have the help of almost omniscient computers will probably do better than those who feel they are getting the harassed attention of an overworked GP even when the diagnosis and the remedies prescribed are exactly the same, as in most cases they will be. But that is not the basis on which we are promised a revolution in the delivery of healthcare. The revolution may come anyway: we are living through an enormous expansion in the reach and variety of machine learning systems, but it will not be for some time. The great majority of diseases do not require heroic diagnosis and exceptional treatment so much as the humane application of well-understood treatments. Much of what’s wrong with the NHS is a lack of money rather than sophistication. Even urgent large-scale threats such as the emergence of antibiotic-resistant pathogens don’t need artificial intelligence to avert, only the consistent use of the intelligence we already have.
It is the apparently small-scale automation of clerical work that we need to think about, because that might happen as quickly as the spread of smartphones did. Vast areas of bureaucracy are about the reduction of complex problems to simple ones for which the correct answers can be written down in a flow chart. This is artificial stupidity rather than artificial intelligence, but the two can merge inside computer systems to produce huge social change. Once the work has been broken down into simple algorithms, these can much more easily and quickly be followed by machines. The 111 service in north London is only one example of a much wider phenomenon. A Japanese insurance company has just replaced 35 claims processors with IBM’s Watson expert system. The Japanese government is preparing to automate the responses to parliamentary questions in a similar way. These are the first signs of a process that may annihilate millions of white-collar jobs in the same way that blue-collar jobs have already disappeared across the developed world. That would be a development to make last year’s political upheavals look like the mere premonitory tremblings of a real earthquake to come.
