Is AI game-changing technology? If so why?
AI is changing the game but it has not yet led to the massive breakthrough many cancer researchers are hoping for. Cancer is a very complex disease involving changes to a patient’s genome, the way the tissue is built around the tumour, and even widespread changes across the whole body through the cancer’s impact on the metabolism and the immune system.
Compared to this complexity, humans can only handle relatively small pieces of information and so it is no surprise that cancer research has traditionally focused on a small number of very well-studied genes and biological mechanisms. The hope is that AI can re-evaluate what we know about cancer and come up with new strategies that human researchers have not yet thought about. For a comparison, think about how AI was applied the board game Go. The technology surprised experts with a move that defied human intuition and experience—the famous ‘Move 37’.
What role does AI play in diagnosis?
AI is pervading all areas of medicine. The disciplines that picked it up most easily were the ones working with images, benefitting radiologists and pathologists most notably. Many systems now exist to support their work with automated image analysis.
One example, from my own work, where these new AI methods had a substantial impact, is the early detection of cancer in the food pipe. People with reflux can experience changes in the tissue where the food pipe meets the stomach. This was generally investigated by endoscopy, a very uncomfortable procedure. My colleagues in Cambridge developed a gentler alternative consisting of a sponge on a string to collect cells along the food pipe. The major bottleneck was the need for specialised pathologists to analyse these cells—there were not enough pathologists available to use the technology. This is where the AI helped. Together with my lab, we developed an AI approach to cut down the analysis time by 60 per cent. Now the capsule sponge is used across the UK and US.
What role does AI play in treatment?
Finding the best treatment for a cancer patient requires teamwork. Oncologists, radiologists, surgeons, nurses and other experts typically come together for multi-disciplinary team meetings. AI is not yet at a stage where it can replace this human intelligence, but it can support it by making relevant data easily available, highlighting important pieces of information and following set decision-making protocols.
In the future, AI could deliver an automated system to pair patients with ongoing trials, regardless of where they live in the country. Currently, you might only get the newest treatments if you live close to a centre of research excellence. Being able to automatically connect patients to the right trials would be a major step in democratising access to cutting edge medical research.
What are the limits and challenges of AI usage?
The major issue is accountability. We trust medical professionals because they are trained in a well-established and well-tested educational system. The same will soon be true for AI—it will also be trained in well-tested systems and humans will learn to trust it the same we learned to trust doctors. But for many of us it still feels like a major difference whether the decision is made by a human or a machine.
Human health is too important for a “computer says no” approach. In the short and medium term, this means that AI tools will be designed as support for human doctors, rather than as a replacement. The human will be accountable; the AI will just be a tool.
