AI, radiology and the future of work

Economist, AI, radiology and the future of work, Jun 7th 2018

Some AI researchers think that human beings can be dispensed with entirely. “It’s quite obvious that we should stop training radiologists,” said Geoffrey Hinton, an AI luminary, in 2016. In November Andrew Ng, another superstar researcher, when discussing AI’s ability to diagnose pneumonia from chest X-rays, wondered whether “radiologists should be worried about their jobs”. Given how widely applicable machine learning seems to be, such pronouncements are bound to alarm white-collar workers, from engineers to lawyers.... 

One is the nature of AI itself. The field is suffused with hype. Some papers show artificial radiologists outperforming the ones in white coats (see article). Others, though, still put the humans ahead. The machines may eventually take an unambiguous lead. But it is important to remember that AI, for the foreseeable future, will remain “narrow”, not general. No human is as good at mental arithmetic as a $10 pocket calculator, but that is all the calculator can do. Deep learning is broader. It is a pattern-recognition technique, and patterns are everywhere in nature. But in the end it, too, is limited—a sort of electronic idiot-savant which excels at one particular mental task but is baffled by others. Instead of wondering whether AI can replace a job, it is better to ponder whether it could replace humans at a specific task. 

The human touch 

That leads to a second reason for optimism: the nature of work. Most jobs involve many tasks, even if that is not always obvious to outsiders. Spreadsheets have yet to send the accountants to the dole queue, because there is more to accountancy than making columns of figures add up. Radiologists analyse a lot of images. But they also decide which images should be taken, confer on tricky diagnoses, discuss treatment plans with their patients, translate the conclusions of the research literature into the messy business of real-life practice, and so on. Handing one of those tasks to a computerised helper leaves radiologists not with a redundancy cheque, but with more time to focus on other parts of their jobs—often the rewarding ones. 

A third reason for optimism is that automation should also encourage demand. Even in the rich world, radiology is expensive. If machines can make it more efficient, then the price should come down, allowing its benefits to be spread more widely and opening up entire new applications for medical imaging. In the Industrial Revolution the number of weavers rose as the work became more automated. Improved efficiency led to higher production, lower prices and thus more demand for the tasks that the machines could not perform. Medicine itself provides a more recent example. “Expert systems” were the exciting new AI technology of the 1970s and 1980s. They eventually made their way into hospitals as, for instance, automated diagnostic aids. That has been a boon, letting nurses—or even patients—undertake procedures that might previously have required a doctor.