Feature
Does artificial intelligence present a threat or promise for GPs?
Whether legal clerk or journalist, programmer or even doctor – the story goes that machine-learning algorithms can do your job faster, better and cheaper. But is that story accurate?
Five years ago, Oxford University researchers put out an influential working paper claiming almost half of all American jobs could be done by machine. Robots would take blue collar jobs and artificial intelligence (AI) would take white collar work.
That paper opened the floodgates. Dozens of studies and reports emerged, suggesting wildly differing impacts from AI and robotics. A 2015 report by the Committee for Economic Development of Australia suggested 40% of all Australian jobs could be replaced by technology by 2025.
But is it really likely that AI will render doctors jobless?
Early signs suggest that people skills and breadth of knowledge will give GPs the edge.
In early 2017, a London hospital trust began giving patients who were on the phone the option of waiting to speak to a human or immediately talking to an AI ‘chatbot’ – a smart software tool you can speak with – made by Babylon Health that would gauge their symptoms and triage them.
But the trial was abandoned by the end of the year, after it was found that many people were using the AI app to try to speed up access to a GP.
When the consumer-facing Babylon app and competitors like Ada first launched, Chair of the RACGP Expert Committee – eHealth and Practice Systems (REC–eHPS) Dr Nathan Pinskier told the Sydney Morning Herald that there was an issue with standardisation of advice.
‘The question is, how standardised are those support tools and, if you enter the same information into different products, will you end up with similar outcomes and guidance? The evidence says no, at the moment,’ he said.
Dr Pinskier told newsGP that AI presents both threat and promise for GPs.
‘It may give us opportunities to intelligently mine data and have high-quality decision guidance available. But it may change the way GPs practise and cause disruption to the normal workflow, or mean that GPs have to think differently about how they provide services going forward,’ he said.
Dr Pinskier said that the range of ailments a GP treats, coupled with their people skills, would give them some immunity to job losses. He sketches a possible future where AI could actually benefit GPs directly.
‘At the moment, we have fairly primitive forms of decision guidance [from our software],’ he said. ‘We get advice on drug interactions, but the software is not intelligent or consistent enough to give good advice. It gives us constant prompts, and many GPs just turn them off.
‘You get prompt fatigue, information overload.’
That’s where a smart AI bot could come in. Rather than having to break the flow of a consultation by inputting information manually, a GP could simply speak to their AI helper software. Once the GP completes gathering information, the AI could offer possible diagnoses.
‘If it can give us high-level guidance – ‘Under no circumstances should this patient be prescribed this medication’, for example – it would make a big difference,’ Dr Pinskier said.
He suggests that AI could also make time spent in the waiting room useful. Patients could input their own symptoms into AI software, which would then send a summary to the GP.
This kind of generalist AI is still some way off, however.
What AI does do particularly well right now, according to Adelaide radiologist and AI researcher Dr Luke Oakden-Rayner, is achieve narrow, clearly definable tasks.
In an article for The Conversation, Dr Oakden-Rayner writes that ‘[t]he enormous volume of carefully and expensively curated data required to train a system that can do everything a doctor can is currently far beyond our reach … [i]nstead, we will see narrow systems performing individual tasks for the foreseeable future’.
For instance, Google has tested a specific AI tool’s ability to diagnose diabetes-related eye disease in several hospitals in India and found the quality of diagnosis is level with ophthalmologists – a potential boon, given there is a lack of that specialty in India. Stanford University researchers have programmed an AI tool able to diagnose skin cancers as well as dermatologists.
Dr Oakden-Rayner’s specialty – radiology – is one many believe is likely to become automated. AI trained on large, high-quality data sets can now interpret some medical images as well as – or better than – human radiologists.
But even in these specialist areas, it’s far from certain that AI will kill jobs.
Australian startup Maxwell MRI has programmed an AI capable of detecting prostate cancers in medical imaging and plans to roll out detection of breast cancer, lung disease and neurodegenerative diseases by the end of this year.
Maxwell MRI Chief Executive Elliot Smith told newsGP that his goal is to use AI to improve diagnoses. Longer term, he wants his AI to be able to predict which patients would be most likely to develop cancer. However, AI would not be a replacement for radiologists or other doctors, he said.
‘It gives them more time to focus on what they do best – working out what the next step should be – rather than scrolling through imagery,’ Mr Smith explained.
‘We’re not after [medical jobs]. Much like any automation, there are things these technologies will do well. Clinicians provide so much more than drawing circles around suspicious areas.
‘By the end of this year, we’ll see the first round of AI-powered products getting into clinical workflows. Like any machinery, they’ll start doing a small proportion of the work, then more and more as clinicians become comfortable letting them do more of the work.
‘There will be a growing relationship of trust.’
Mr Smith’s view is backed by a prominent medical AI researcher and radiologist, Dr Hugh Harvey.
‘[R]adiologists do not just look at pictures,’ he writes on Towards Data Science.
‘Where radiology [AI] is heading towards is digital augmentation of radiologists, to the point at which their job becomes to monitor and assess machine outputs, rather than manually go through every possible mundane finding as they do now.’
For his part, Dr Pinskier believes AI will inevitably become a vital part of the medicine machine.
‘If we project 50 years forwards, there is no question AI will play a huge role in medicine,’ he said. ‘I’m a sci-fi fan, and a lot of the things you see in sci-fi end up eventuating. It gives you a possible future.
‘We need to convert that into a real future.’
AI artificial-intelligence ehealth
newsGP weekly poll
On average, how many patients do not show up for their appointment at your general practice each week?