Patient commentary: Stop hyping artificial intelligence—patients will always need human doctorsBMJ 2018; 363 doi: https://doi.org/10.1136/bmj.k4669 (Published 07 November 2018) Cite this as: BMJ 2018;363:k4669
All rapid responses
I read with great interest the article by Mittleman M and colleagues (1).
I disagree to some extent with the comments “hyping the artificial intelligence (AI)“(1) and agree––at the moment––that almost no one wishes to get a full medical assessment by a robotic doctor, chatbots or AI driven web pages. There should always be a human touch at some point.
Creating a clever machine, vehicle or device has ever been a great passion for human beings, extending back even to the stone age.
AI has been there for a long time and is continuously evolving. Its first leap was by Alan Turing in the 1950s, when he introduced the Turing Test. In short, he proposed a few big questions “Can Machines think? “and “Are there imaginable digital computers which would do well in the imitation game”? (2) He believed simply they could.
The second innovative leap was when the AlphaGo computer won the Go game (which is a rather more complex game than chess) in the Master world champion in 2010 (3). This historic exiting victory over humans triggered lots of research world-wide and spread to medicine in a short time.
There is unfortunately no absolute answer whether AI enhanced robots should take over all doctors’ responsibilities including delivery of “bad news” to patients with great passionate, understanding and consideration; perhaps these sensitive tasks should have the human touch. When it comes to knowledge and skills, there is one Chinese robot doctor out there that has already passed the Chinese medical curriculum with excellent marks (4).
The future will definitely bring more sophisticated robots in the near future.
I believe that AI is a friend rather than a foe (5) and think there is no black and white statement that human doctors will be replaced completely by the doctor robots right now; but having seen the dramatic advancement in the last century and the new millennium this will eventually happen in the future--not in years, but decades.
1. Mittleman M, Markham S, Taylor M. Patient Commentary: Stop hyping artificial intelligence ––patient will always need human doctors. BMJ 2018;363: k4669
2. Turing AM. Mind 59 (October) 1950:433-460
3. Google’s AlphaGo AI wins three – match series against the world’s best Go player.
https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-pl... (accessed 02.02.2019)
4. Dom Galeon. For the first time, a robot passed a medical licensing exam.
5. Pakdemirli E. Artificial intelligence in radiology: friend or foe? Where are we now and where are we heading? Acta Radiologica Open 2019 (Accepted opinion /perspective article on 15.01.2019)
Competing interests: No competing interests
I would like to disagree respectfully but strongly with Dr Cohen at the notion that the notion of the generalist doctor will disappear into subspecialisation. I am all for nurses and allied health professionals playing greater roles and having more clinical autonomy and advanced skills. However, the need for skilled medical generalism is stronger than ever. And being a good "expert generalist" arguably needs more training and expertise than any single organ or disease based specialism. It is no coincidence that the most advertised jobs in secondary care are in areas like acute internal medicine, geriatric medicine and emergency medicine, nor that we have a concerted push to preserve and hopefully expand primary care.
With more people surviving formerly fatal conditions of childhood and mid life, the core business of modern medicine is increasingly the care of individuals with multiple long term conditions, often complicated by frailty, disability or functional impairment and often using multiple services.
These individuals value continuity of care, care co-ordination, and more person-centred approacahes than traditional single disease or single speciality models can provide.
And even for those doctors who do practice more specialised medicine, their patients will still have numerous co-morbidities and also often some psychosocial vulnerability
Algorithms and protocols may work well for diagnostic tests, procedures and fairly predictable single disease trajectories
However, the kind of relationship-based, person-based care where patient and carer/family priorities and the competing risks and benefits of different treatment options, therapeutic goals and limits of treatment in people with many things wrong with them at once, require broad and lengthy training and experience and skills in communication, care co-ordination, ethics etc which go well beyond an evidence base designed around single disease entities and often excluding people with complex needs
Maybe, if anything the use of AI/robotics will lend itself more to well defined single disease pathways in patietns wtih few comorbidities and doctors who have the broad skills and knowledge to support patients with multiple problems will be (as they are increasingly now) those in most demand, most valued by most patietns and hardest to replace with digital solutions
Competing interests: No competing interests
The arguments put forward assume that the current abilities of technology in healthcare, be it in the use of electronic health records or a machine providing emotional support to a patient, are as optimal as they will ever be.
Whilst the ability for an artificial intelligence to be able to pass the Turing test, thus demonstrating empathy and kindness indistinguishable from a human being, is perhaps a long way off, the abilities of technology and AI have been increasing at an exponential rate (and continue to do so). As shown by the progress from rudimentary attempts at social robots such as Kismet in the late 90s, to Erica the ‘android with a soul’, a robot with a specialised purpose of ‘breaking bad news’ could be very much feasible in the next few decades.
Not only feasible, it could be preferable to the current idea of a doctor who is expected to be ever knowledgeable, capable and omnipresent when it comes to patient support. Even in current practice, cancer specialist nurses play an enormous role in the emotional support of patients with terminal illness. With the increasing prevalence of nurse practitioners and physician’s associates, the traditional idea of a generalist ‘doctor’ disappears into sub-specialisation. Artificial intelligence using deep learning has already been shown to be more effective in diagnosis and management decisions in certain clinical tasks, especially those involving image and data processing. As the need for clinical ability in diagnosis decreases with increased reliance on quantitative results and imaging, what role will ultimately be left for a doctor (as we know it) to fill?
1. Turing, A.M., Computing machinery and intelligence, in Parsing the Turing Test. 2009, Springer. p. 23-65.
2. Hilbert, M. and P. López, The world's technological capacity to store, communicate, and compute information. science, 2011: p. 1200970.
3. Breazeal, C. and B. Scassellati, A context-dependent attention system for a social robot. rn, 1999. 255: p. 3.
4. Glas, D.F., et al. Erica: The erato intelligent conversational android. in Robot and Human Interactive Communication (RO-MAN), 2016 25th IEEE International Symposium on. 2016. IEEE.
5. De Fauw, J., et al., Clinically applicable deep learning for diagnosis and referral in retinal disease. Nature medicine, 2018. 24(9): p. 1342.
Competing interests: No competing interests
I am a junior doctor working within a fast moving and high pressured environment to treat patients with both accuracy and compassion, and an academic researching human error and the impact of stress upon performance.
The ability of a human to act as an effective doctor is limited by a multitude of factors highlighted in ‘Could artificial intelligence make doctors obsolete?’ including: time limitations in a busy workplace, by limited training in relation to the immeasurable amount of knowledge required, and by cognitive strain and long hours in a stressful environment. Humans are therefore resource limited and utilise functions that a robot would do in order to treat patients:
As doctors we see a great number of similar presentations of the same condition. We gather an internal database of symptoms and signs which indicate the condition's correct diagnosis. Therefore when we see a patient presenting with X, Y and Z symptoms, we can match them to the hundreds of previous patients with X, Y and Z, and diagnose them successfully. In humans this technique is limited by our own personal experience and we can be biased by particularly memorable cases or indeed, patients that we have misdiagnosed who have then gone on to suffer because of it. This technique is utilised in Artificial Intelligence, however with the input of a multitude of patient presentations robots could be able to apply this rule with much more precision and success than a human doctor, improving healthcare provision.
Over the course of our experiences we are met with the same presenting symptoms over and over. One technique of approaching a specific symptom that humans use is frequency gambling.  We realise that when a certain set of X, Y, and Z symptoms occur it most likely leads to diagnosis A, and rarely leads to diagnosis B. Therefore we are more likely to immediately treat the patient for diagnosis A. Artificial Intelligence utilises this system, however again with imputation of wide reaching data a robot would be able to more accurately frequency gamble whilst potentially being able to find sets of symptoms that were ‘hidden’ to the human doctor.
Medical training requires repetition which results in automated behaviour. The process of undertaking a task again and again allows the human doctor to perform a task efficiently, effectively and safely without having to engage cognitively. This is essential with the strain of being a physician however, can result in attentional slips and mistakes resulting in harm to the patient and harm to the doctor. Artificial Intelligence would function through an automated pattern relying on programmed ‘checks’ which could abort tasks before harm to the patient occurs, a further advancements which could improve patient care.
Applying artificial intelligence to empathy
Humans apply the above techniques to empathy. Mittelman, Markham and Taylor compassionately wrote about their own needs through their experiences of multiple sclerosis, end stage renal disease, mental illness, epilepsy, and Crohn’s disease, however I have never personally experienced any of these conditions and must still show empathy for patients who have these diagnoses. Doctors are often taught frameworks in order to do this, such as ‘SPIKES’ (Setting, Perception, Invitation, Knowledge, Empathy, Summary), and I must apply my training in, and experiences of, breaking bad news and difficult diagnoses to compassionately care. To do this I must draw on similar experiences and reflections of patient care to effectively communicate with patients and discuss subjects which are life changing. Artificial intelligence may be able to comfort patients based on vast imputed experience with a much more informed, experienced, and kind approach than I am able to.
Competing interests: I have read and understood BMJ policy on declaration of interests and have no relevant interests to declare.
Exclusive License: The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, a worldwide licence (http://www.bmj.com/sites/default/files/BMJ%20Author%20Licence%20March%20...) to the Publishers and its licensees in perpetuity, in all forms, formats and media (whether known now or created in the future), to i) publish, reproduce, distribute, display and store the Contribution, ii) translate the Contribution into other languages, create adaptations, reprints, include within collections and create summaries, extracts and/or, abstracts of the Contribution and convert or allow conversion into any format including without limitation audio, iii) create any other derivative work(s) based in whole or part on the on the Contribution, iv) to exploit all subsidiary rights to exploit all subsidiary rights that currently exist or as may exist in the future in the Contribution, v) the inclusion of electronic links from the Contribution to third party material where-ever it may be located; and, vi) licence any third party to do any or all of the above. All research articles will be made available on an open access basis (with authors being asked to pay an open access fee—see http://www.bmj.com/ about-bmj/resources-authors/forms-policies-and-checklists/copyright-open-access-and- permission-reuse). The terms of such open access shall be governed by a Creative Commons licence—details as to which Creative Commons licence will apply to the research article are set out in our worldwide licence referred to above.
1. Goldhahn Jörg, Rampton Vanessa, Spinas Giatgen A. Could artificial intelligence make doctors obsolete? BMJ 2018; 363:k4563
2. Reason J. Human error. Cambridge university press; 1990 Oct 26.
3. Mittelman Michael, Markham Sarah, Taylor Mark. Patient commentary: Stop hyping artificial intelligence—patients will always need human doctors BMJ 2018; 363 :k4669
4. Baile WF, Buckman R, Lenzi R, Glober G, Beale EA, Kudelka AP. SPIKES—a six-step protocol for delivering bad news: application to the patient with cancer. The oncologist. 2000 Aug 1;5(4):302-11.
Competing interests: No competing interests