The UK government wants to use artificial intelligence to spot signs of disease earlier. In a speech today, prime minister Theresa May said that AI could . The technology could also be used for other conditions, such as heart disease.
The plan is to develop AI systems that can scan data about patients’ habits and genetics, and then cross-reference the country’s health records to spot people at an early stage of disease.
“The development of smart technologies to analyse great quantities of data quickly and with a higher degree of accuracy than is possible by human beings opens up a whole new field of medical research and gives us a new weapon in our armoury in the fight against disease,” said May.
Advertisement
Pattern hunters
It’s hard to know how well the plan will work without more detail, but in principle this is the sort of area where AI can excel. The technology can analyse vast amounts of data easily, spotting patterns that would otherwise go undetected.
In previous tests, an AI pitted against dermatologists was as good as them or better at distinguishing between cancerous and non-cancerous skin growths. Another system outperformed doctors at diagnosing diabetic retinopathy, a complication of . But there has yet to be a test on the scale proposed by May, and that could be a problem.
One potential pitfall is the chance of false positives – finding that someone has a disease when they do not. If rolled out to the whole population, even a test with a very low false positive rate could lead to thousands of people needlessly undergoing stressful tests and examinations.
At the other end of the scale are false negatives, where the computer misses a diagnosis. Despite huge progress in the field, AI still makes mistakes. In 2016 a Tesla car driving autonomously crashed into a truck because its computer systems simply did not spot it. The man in the car driver’s seat died. Recently, an Uber self-driving vehicle killed a pedestrian.
Deadly mistakes
Given that it is often difficult to tell how an AI makes its decisions, it is possible a medical diagnostic system could miss a symptom or signal that would be completely obvious to a doctor.
The risk of using AI in the way May proposes might be justifiable on the grounds that it could catch more cases of disease than it misses, in the same way as driverless cars could prevent more accidents than they cause. But we don’t know whether people will agree with this reasoning if and when the technology goes mainstream.
Finally, any use of UK National Health Service medical data to train privately owned AIs must be done with care, to avoid the public being short-changed. In 2016, 91av revealed an improper data-sharing agreement between Google’s AI firm DeepMind and the Royal Free NHS Trust. If AI is to be rolled out across the entire NHS, it must be done in a way that benefits everyone.
