The (robot) doctor will see you now – Expert reaction

Artificial Intelligence (AI) will spell the end of doctors as we know them, warn two Kiwi medical professionals.

Medical-robot“We believe that over the coming years AI will perform a significant amount of the diagnostic and treatment decision-making traditionally performed by the doctor,” write Drs William Diprose and Nicholas Buist in an new New Zealand Medical Journal viewpoint article (subscription required).

The two Whangarei doctors highlight the rapid progress of machine learning and artificial intelligence in the health sector thus far, noting that for a safe, sustainable healthcare system, “we need to look beyond human potential towards innovative solutions such as AI.”

“Humans would continue to be an important part of healthcare delivery, but in many situations, less expensive fit-for-purpose healthcare workers could be trained to ‘fill the gaps’ where AI are less capable,” write the authors.

“As a result, the role of the doctor as an expensive problem-solver would become redundant.”

The SMC collected the following expert commentary.

Dr Robert Hickson is a future-scanning and foresight researcher and author of the Ariadne blog. He comments:

“Diprose and Buist are right to highlight the prospects of artificial intelligence in healthcare. Though I don’t see the end of doctors, just a change in some of the tasks that they do.

“The consultancy firm Frost & Sullivan recently forecast revenue from AI and related applications in healthcare to go from US$800 million last year to $6 billion by 2021. That’s a fraction of healthcare, but indicates the pace of development. AI is being developed or applied across healthcare –  diagnostics and imaging, virtual assitants, patient monitoring, drug discovery, and risk analytics – and reflects the broader trends of automation and analytical approaches in medicine. IBM’s Watson cognitive system is already being used to recommend leukaemia treatments, and to help insures evaluate treatment plans. DeepMind – the company who developed the Go-winning AI recently, has also just got access to patients NHS records in the UK to help monitor patients with kidney disease.

“So, over the next decade increasing use of AI as decision support tools for health practitioners seems inevitable as the sector treats more elderly and more complex medical conditions.

“At the moment there is a lot of hype about what AI can do based on relatively simple or well defined problems, like games. What we need to see are more trials that evaluate diagnoses or treatment outcomes from AI systems versus the traditional approaches to see whether AI can provide more general medical support. And as the surgeon and author Atul Gawande has noted, further collaboration is required between designers of AI systems and clinicians, so that the former provides the latter with information that they can use in their daily practice. These will come, in many but perhaps not all situations.

“The main risk is that not every aspect of healthcare can be captured within the bounds of an algorithm. Healthcare practitioners shouldn’t suspend their judgement. As Diprose & Buist point out, doctor-patient relationships aren’t just about efficient diagnoses, and there is a growing interest in holistic approaches to health that AI will need to support.

“Effective use of AI in healthcare will also require good information about the populations that they help serve. We already see limitations in the availability of genetic and physiological information that is relevant to, for example, Maori and Pacific people.”

Prof Ajit Narayanan, School of Engineering, Computer and Mathematical Sciences, Auckland University of Technology, comments:

“The interesting aspect of the viewpoint article is that it pays no attention to failed claims made over 30 years ago that AI would soon replace human medical decision making (see ‘The expert medical computer,’ J.R.W. Hunter, in  Yazdani and Narayanan (Eds), Artificial Intelligence: Human Effects, Ellis Horwood, 1984).

“Many of the so-called AI cases presented by Diprose and Buist are still based on non-intelligent exhaustive search techniques and statistical methods that look for correlations and variances, rather than causal knowledge,  to identify maximally different attributes to distinguish cases.  Because CT and other brain images are now used to identify important features does not make the result ‘pattern recognition’.  Searching millions of possibly non-relevant medical reports to make a judgement concerning an individual patient does not make the judgement more accurate for that individual.

“The AI diagnosis and treatment methods identified in the viewpoint  article do not lead to new knowledge of what causes any of the symptoms under investigation but only identifies their possible presence according to prior human medical decision making which is built into these programmes. As a consequence, the viewpoint article contains a lot of speculation rather than generalization.

“A doctor does more than reach a conclusion. Typically, a doctor decides whether a medical condition is present as well as attempting to determine which medical condition or whether the condition is treatable. These are complex, inter-related processes involving hypothesis generation and refutation with inductive causal and non-precise reasoning.

“Doctors also provide preventative advice on the basis of what they hear and see in front of them.   It is currently not known how a doctor does this except to use something profoundly human that has so far escaped AI: some combination of intuition, experience and compassion.”

Assoc Prof Jeremiah Deng, Information Science, University of Otago, comments:

“I work in the machine learning field and have been involved in medical image analysis and medical data-mining for diagnosis purposes. Artificial Intelligence and data mining are becoming omni-present and all permeating, as we are seeing in the new developments of self-driving cars, smarthomes, robots, e-commerce and beyond.

“Some people say the profession of teachers will disappear, after doctors, which I don’t really buy into.

“Apart from the challenge of domain knowledge (important, but often tacit, hard to encode into computer/robots), there are at least two main obstacles for AI to overcome and be able to function on the critical profession level: creativity, and compassion. These elements are crucial in problem-solving perhaps in any domain, and in particular, medical practice.

“On the other hand, with AlphaGo [a computer program designed to play the game Go] neural networks have found their way back to limelight, which is no doubt impressive achievement, but it remains unclear how “intelligent” the algorithm is. If we are given a “blackbox” which checks you up and produces a 99.99% diagnosis accuracy, can we really trust it?

“Marvin Minsky, Father of AI, once said “we are still in the dark ages” in terms of scientific understanding about human minds. The fast advances of AI and intelligent machinery will definitely impact the society, but whether they will threaten humans or not in a significant way remains a question.”