Back to blog

Artificial Intelligence and learning from your mistakes

by Richard Moss



On the 4th– 10th of July the Royal Society hosted the Summer Science exhibition, where cutting-edge innovators and researchers demonstrated their creations with the hope of inspiring a new generation. Among these were a couple of stand-out exhibits using artificial intelligence (AI) technology. Throughout the course of the day I was to learn what AI actually is and how it is being used to speed up research and treat patients in a more effective manner.

A simple ‘learning’ machine was on display and available to play games with consisting only of paper, sweets and cups. This machine learnt by not repeating moves that resulted in losses. Thus, the more you play the game, the better choices it makes, the more difficult it is to beat. This ultimately results in an unbeatable competitor. If this simple game can teach a computer to beat a human every time in less than an hour, it is natural to wonder how much a more sophisticated computer can learn with a greater data input.

The most sophisticated technology I encountered was that which applied AI to surgeries, both pre and during operations. 2D CT scan images are converted by Machine Learning into a 3D, moveable and interactive impression of the body. This provides the surgeon with a comprehensive view of the patient without the tedious sifting through of individual slices. What’s more, the 3D image can be used in surgery hygienically by using hover-hand movements to control the image thus eliminating any contact-caused contamination. I was also given an idea of what the combination of robotic surgery and machine learning are working towards; surgery with the only human in the room being the patient. Body-mapping and precision tools could lead the safest surgery being carried out by the hands of a robot. Surgical robots can in the near future be able to perform simple procedures such as liver, breast or kidney biopsies independently without human input but with the help of artificial intelligence.

When one sees the capabilities and applications of AI becomes the question inevitably arises; how much can I trust a machine? Having controlled the robots used in this type of surgery to repair a computer simulated kidney, I can safely say I would prefer a robot on the other end of the controls than my unsteady, human hands. Furthermore, it is humans that develop the technology which can do this. Machine learning is limited by both data and choices. In reality these are all limits placed before it by man, what the machine is doing is simply equations. I consult a calculator for the most simple of equations, why would I trust myself more with the larger ones? The computing ability of machine learning allows the tedious, time-consuming calculations to be easily dealt with, without the added risk of human error.

The only way in research is forward, it seems AI is getting there at a greatly accelerated pace. I left the exhibition feeling thoroughly reassured that the collaboration of STEM researchers from each discipline is shaping healthcare and other industries for the better.