InnovateEOE
InnovateEOE
Artificial Intelligence is a field of study where computers and machines are taught to mimic the problem solving and decision-making capabilities of the human mind (IBM, 2020).
In it’s simplest form, AI uses a large amounts of good quality data (the more the better!) to build a model or algorithm. This model is then shown new data where it can be used to predict the probability that the new data you are showing it, is the same as the data it was trained on.
The main types of AI learning are:
- Reinforced learning – where a machine uses a large number of attempts to learn a simple task e.g. to win a chess match.
- Supervised learning – where researchers tell a machine what the correct answer is for a particular input.
There are 4 main models of AI:
- Classification
- Clustering
- Regression
- Dimensionality reduction
The basis of all AI algorithms is the ‘neural network’ which is similar to the network of neurons in the brain, although more organised. The reason AI is able to complete less complicated decision making than humans may be because of this somewhat limited (though multiple layers can be added) web of neurons in an algorithm, compared to the human brain.
Part of the reason AI advances have been so exponential over the last decade is because of the advances in super-computing.
There are countless applications of AI and many more still to be discovered. AI can be used to generate random computer images of faces, to replenish stores in factories and to transport goods around factory floors. AI can be use to simulate protein folding to study pharmacological drug binding, and AI chat bots can automate processes that would otherwise be carried out in a more costly (monetary and time) manner, by a human. AI has been successfully applied to healthcare in radiology, dermatology, and ophthalmology. Funding opportunities for AI related projects can be found with the NIHR and NHSx
The ethics of using AI to make complex decisions in the hospital are complicated. We must scrutinise algorithms decision making ability for not only accuracy, but also fairness. Whilst application to treatment decisions for patients is complicated, there are numerous other simple potential applications of AI in hospital. These range from automating staffing rotas, to staffing departments to surges in demand (like that seen on a Friday evening in secondary care, for example).
Once an algorithm has been verified in terms of accuracy, it must be implemented into a trust. At this point it becomes a leadership and management challenge as we ensure that clinicians are supported by the power of AI, rather than displaced or challenged by it. This approach must be sought (collaboration) if we are to ensure algorithms actually help clinicians. The algorithms must also be trained by clinicians to be the best possible decision makers. This can relieve clinicians from more mundane, routine tasks.
If you would like more information on AI, please feel free to make contact. You can also visit the AI in healthcare website at the Cambridge Centre for AI in Healthcare (https://ccaim.cam.ac.uk). Here you can find information on future teaching sessions on AI, and also past recorded sessions.