Language evolves. When I was a kid, a bull was a bull. Venturing to the local bull station was a mixture of trepidation and testosterone. The massive beasts, confined to stalls that seemed to burst with muscle and intent. They oozed raw power, A glance from one was enough to send a serious shiver up your spine. They had a job to do. Bulling was their business and there was no way to square that off. Yet, change was in the air. The arrival of a massive tank of liquid nitrogen signalled the end of an era, the bulls became the straw men of the breeding business. The age of the comfort visits, from the Kerry cows, evaporated in the mists of the nitrogen vapours. AI had truly arrived.
Fast forward a few decades and most people couldn’t even tell the original meaning of AI. Yet, if bulls could tell the future, they would have felt like we all do about AI. It’s big, it’s here. Life is never going to be the same again.
When ChatGpT burst on the scene, a little over a year ago, jaws dropped. It was awesome. It wasn’t long before the gloss wore off. The training data was date limited and incomplete. Just like a regular human, it started making stuff up. The hallucinations were funny, reminding us that it wasn’t quite ready for primetime. In a bare year, much of the AI software has jumped forward in leaps and bounds. Unlike a traditional computer program, a genuine AI system creates knowledge from data, telling us something that it wasn’t specifically told. The latest systems are quite capable of delivering extraordinary results. However, it is within these technological miracles that we see one of the fundamental weaknesses of current iterations of AI. They need the right data. Garbage in, garbage out still applies. So, when we’re doing things like creating pictures, videos or text processing, AI is indeed doing all sorts of wonders. These machines already out class us at some levels. One big fear is that machines and AI will replace our jobs. This may well be true, in some cases. But it is much more likely that AI will be the industrial revolution of the 21st century. It will provide us with new ways of approaching problems and provide us with great tools to deliver exceptional services. Everything computerised is getting the AI label, as the bandwagon rolls along. One interesting area is mimicking a pancreas. Connecting insulin pumps to continuous glucose monitoring devices, with insulin dose automatically delivered dependent on blood glucose levels. Strictly speaking, this is not really AI, more of an If This, Then That (IFTTT) system. I am not sure that the typical person with diabetes is ready to turn their life over to a computer program, just yet. These systems can’t fill the insulin cartridges, change the batteries or trouble shoot that an infusion set is not working optimally, so simply put AI helps but it still needs human intervention.
In pharmacy, we have already been using versions of expert systems in our pharmacies. When I open my browser access drug interaction checker, I am accessing the distilled wisdom of many decades of drug use and research. However, it isn’t AI, just a program querying data and regurgitating a pre-programmed answer. This is such a poor use of potential resources. Imagine if we had a system that could take a patient profile, not just weight, height, age and gender. If it had access to current blood profiles, disease states, gene expression and more. Where it could tell you that this patient is likely to have a 100% increase in QT interval, or 85% increase in statin blood levels, or whatever. If the system could, at point of care, be a genuine expert at our shoulder. It is unfortunate that pharmacy, an early adopter at the technology table, has stood almost still in IT terms for over three decades. While dispensing robots look good, they do not add to the evolution of pharmacy as a patient centred profession. The real change will occur when the computers can actually give meaningful advice. Our job will be to interpret and apply this vast new knowledge. As the world ages, this will be evermore relevant. This is not without potential pitfalls. A hugely important issue will be liability. What happens if an AI system makes a mistake and you act in good faith, following the advice and the patient comes to harm? What happens if the computer makes a mistake and you, as a professional, discount or dismiss its advice, and the patient is harmed? Liability and legal peril are going to abound. The dream of personalised medicines is not too far away, but is tantalisingly out of reach. Imagine taking all the patient data and designing a unique treatment just for an individual. A CRISPR machine in the corner, churning out unique proteins. Perhaps a little futuristic, but well withing the realm of possibility. That’s no bull!
Jack Shanahan
MPSI
Highlighted Articles
