As AI seeps into medicine and healthcare, it stands to change how doctors diagnose, monitor and treat patients.
Clinicians still need to be part of the picture, but that picture is changing. So, what can humans provide in healthcare that AI cannot? What do patients and clinicians need to know? And who is accountable if AI gets it wrong?
These are some of the questions to be tackled in an upcoming public discussion titled AI in Our Healthcare: Promises, Pitfalls and the Path Ahead, organised by the Fellows of Trinity College Dublin.
“AI is entering the workflow of patients and their doctors,” says nephrologist Mark Little, who directs the Trinity Translational Medicine Institute.
RM Block
“And because it is so powerful at pattern recognition, we are seeing it have tangible impacts in assessing radiology scans and in histopathology, where it can analyse thin sections of patient tissues. AI works well here because it gets the data in a structured fashion.”
More broadly, Little sees the potential for AI as a tool to help personalise treatments.
“AI can take multiple data sources into account, although it currently struggles with the messy nature of that data in the healthcare system,” he says. “It helps to stratify patients according to risk and point to specific therapies that are most likely to deliver a good outcome.”
Some may ask whether one day machines could replace doctors, but surgeon Paul Ridgway has a more immediate concern.
“The foundation of our profession is that we have to make good decisions, and today AI is supercharging clinical decision making, but we still will need that doctor in the loop,” he says.
“So rather than asking will machines replace doctors, which is where a lot of the public debate is, the more pressing question is if you have an AI system shaping a clinical decision and the decision goes wrong, who is accountable?”
[ Overreliance on AI risks ‘deskilling’ health sector staff, report warnsOpens in new window ]
Ridgway, a professor at Trinity and surgeon at Tallaght Hospital, is seeing rapid advances of AI in clinical decision making and robotics in surgery, and he believes that clinicians should retain a healthy scepticism when using the technology.
“During my medical training in the 1990s and early 2000s, we had to read heart measures from ECG machines, which was very difficult. There was some machine learning to help us do that, but it wasn’t very good, so you had to use it as an indicator rather than the end result,” he says. “And in the same way, today the information from AI should not be the final answer.”
He encourages patients and clinicians alike to ask questions of the technology and those who use it.
“We need to empower doctors to ask what is behind this shiny new technology, and to use any outputs of it with caveats,” he says. “And patients should be asking doctors and nurses whether and how they are using AI tools, and how often they have used them before.”
It takes confidence to ask such questions, and AI’s confident tone is part of the underlying issue, says Brendan Kelly, professor of Psychiatry at Trinity.
“The chief problem with the large language models and other AI tools is their ability to apparently answer questions with a confident tone of voice that is not merited,” he says.
“AI is not a reliable source if your knowledge of the topic is poor. And if an AI tells you something that is new information to you, rather than reminding you of something you knew but had forgotten, then you need to check that piece of information elsewhere.”
People using generative AI to work through mental health issues may also suffer from the lack of questioning and containment that a human therapist can offer, he notes.
“AI tools amplify, magnify and reflect what the person is thinking, rather than pushing back gently in the way a friend or therapist could do,” says Kelly.
“And we have seen cases where people with suicidal thoughts appear to have been provided by AI with advice and information that was unhelpful and possibly even fatal.”
Kelly believes that AI will force us to think more about the human element in medicine and focus on what can help patients.
“We need to better understand the humanity of medicine, and value it, and absorb the technology without changing the core purpose of medicine,” he says.
“We should also not let the techno-panic distract us from the small, quiet gains that AI can bring by organising information well, and as a country we need to focus on being able to use health data in a way that can benefit patients. Imagine the gains we could have with a national electronic health record here.”
Scientist Patricia Maguire sees the potential for AI to deliver enormous gains by capturing signals from data to help us understand human biology and develop more efficient ways to diagnose conditions and develop treatments.
“I see AI as a way to augment human intelligence to understand more about what is going on in the body and how diseases arise and can be treated,” says Maguire, professor of Biochemistry at University College Dublin (UCD).
“Without AI we just can’t see as many signals in the data.”
Maguire, who directs the UCD Institute for Discovery, has experience of moving from eyeballing data to using the power of technology.
“For years I used to spend my evenings looking at Excel spreadsheets line by line to try and understand how levels of proteins were changing,” she recalls. “Then machine learning came along and I could now see the patterns in the data and work with other scientists and doctors and pull more information into the picture.”
The technology could identify patterns in anonymised data, and that could benefit so many patients in Ireland. I have goosebumps just thinking about the potential
— Prof Patricia Maguire
Maguire’s group in UCD is now developing several predictive tools to support doctors and patients, including one that can identify from a blood test early in pregnancy whether a woman is at risk of a potentially fatal condition called pre-eclampsia.
“In research we are really only starting to scrape the surface of what is possible,” she says.
Earlier this month at the All-Ireland Clinical AI Congress, which UCD Institute for Discovery supported, Maguire was struck by the recurring theme of getting the foundations right.
“We need trust, good governance and data, so AI can genuinely help scientific research to understand the complexities of illness and to make Irish healthcare better for everyone,” she says.
“The possibilities are huge because the technology could identify patterns in anonymised data, and that could benefit so many patients in Ireland. I have goosebumps just thinking about the potential.”
[ Roadmap for AI in health is welcome but five years is an eternityOpens in new window ]
At the Trinity event, Cliona O’Farrelly, professor of Comparative Immunology and Chair of Fellows at TCD, wants to encourage more focus in education on the possibilities in research and healthcare, and the pitfalls too.
“Our approaches in university and healthcare education are not changing fast enough to accommodate and exploit the advantages that AI brings, nor to deal with the associated problems and disadvantages,” she says.
“Discussion of the ethical issues is particularly scanty, and we need to keep asking how do we keep the human at the centre?”
- A free, public, two-hour event, AI in Our Healthcare: Promises, Pitfalls and the Path Ahead, will take place on September 30th from 6pm at the Edmund Burke Theatre, Trinity College Dublin. Register through Eventbrite or at adaptcentre.ie/news-and-events/ai-in-our-healthcare-promises-pitfalls-and-the-path-ahead/
