Opinion The learned machine

Today, it is statistically probable only because a machine is yet to turn its processors to editorialising. But tomorrow is another day.

Earth-shaking developments were expected when artificial intelligence (AI) specialist Andrew Ng quit the Chinese search giant Baidu in favour of academics. The earth still stands still, but the medical community has reason to feel shaken. An AI algorithm written by his Machine Learning Group at Stanford University has surpassed radiologists for reliability in the diagnosis of pneumonia. It takes seven to nine years to train a radiologist. The algo got there in one month.

The general perception is that robotics and AI will superannuate only lower-level, repetitive jobs. Inventory managers, production line workers and call centre employees will go the way of switchboard operators and pool typists, who were rendered jobless by automated exchanges and the word processor. But AI is ramping up much faster than expected and now, roles which involve decision-making based on experience, rather than protocols and rote learning, have reason to feel threatened. Today, surgeons programme robots to perform procedures more accurately than their own hands could. In tomorrow’s world, the machine could diagnose and operate solo.

Advertisement

Tomorrow is actually today. The Stanford learning algorithm analysed chest X-rays from public health data, and showed both more accurate positive diagnoses of pneumonia than doctors, and fewer false positives. In addition, it produced heat maps from the X-rays. If doctors can be so easily surpassed, why not CEOs? What do you bet that a machine would assure better profitability? A lot of market punters would be willing to bet on such a contest. What about journalists, then? Are you certain that this editorial was written by a human? Today, it is statistically probable only because a machine is yet to turn its processors to editorialising. But tomorrow is another day.