My first completed novel is set in a world where paid work has become the exception rather than the rule. With ongoing advances in automation and digital technologies, I tried to imagine a world where humans actually carrying out work has become uncommon, while at the same time the government has done little to help those finding themselves out of work and on the breadline. Of course the already privileged we’re better protected, but to find out more about that you will need to read the book.
What I wanted to think about in this post, is the idea of automation and digitalisation of work, specifically the automation of my job. When you think of robots taking people’s jobs, it is often the factory worker you think of, with the repetitive actions they once performed replaced by a machine which can do the exact same thing only hundreds of times faster and much more precisely.
If I asked you to think of a job which you think might be automated in the next decade or so, psychiatrist is likely to be low down on your list, if it makes the cut at all. I think there is a perception, both inside and outside the medical profession, that doctors are somehow immune from being replaced by robots. I know when I first considered mechanisation as a real problem facing us in the near future, I was personally comforted by the naïve notion I would be protected, that I in essence had a job for life.
The more I have thought about this issue, and the more I have understood what my role as a psychiatrist actually is, the less confident I have become in my assertions that my job is secure and I am not facing a future fighting a robot for work. I am not necessarily suggesting my role will be gone within the next decade, though I think I have learned enough from seeing other people try and fail to predict the future to avoid being so bold myself. But, I do think my role is at risk from mechanisation, as when you really think about it most jobs are. It might come in small steps, a task here replaced with a computer, a role there given to someone else, but I think there will be an inevitable trend towards an automated medical world, and I do not feel my job is any safer than any other.
In this post, I want to consider exactly what my role entails, and how I could imagine my job becoming automated. For some areas of my role, it will take technologies which might not entirely exist yet, but they could very well be developed in the not too distant future, and I need to be prepared for this.
Before I begin describing my job and how to automate it, I should be clear that I do not intend to consider the solutions to the real human problems which can arise when work disappears and the robots arrive. There are many people more qualified than I am considering ideas from the Four-Day Week to the Universal Basic Incomeas potentially solutions to the problems automation could cause, and if this is a topic of interest to you I would definitely recommend reading more about them. I have also shied away from the moral, ethical and legal arguments. I am not hear to advocate for or against automation, rather to point out that it is possible.
In this post I want instead to focus simply on what my job is and how I think it could be automated. In doing so, I hope I can convince you that even a job which seems protected from a robotic future might not be as secure as it seems.
If you have read my previous blog post, you might have some idea already what my job entails. In that post, I described briefly some of the tasks and roles I complete as a Learning Disability Psychiatrist. In this post, I want to expand in a little more detail on some of the individual tasks I have to complete, so that we can think in the next section about how we could automate those jobs.
I initially considered running through my week, describe what I do during the day, but I quickly realised this would be both tedious and repetitive. I love my job, but that does not mean you will enjoy reading about it.
Instead, what I have done here is broken my job up into a series of tasks which needs to be completed so that I fulfil my duties and my patients are well cared for. When broken down in this way, it might not seem like I do much in the week, but let me assure you with a busy caseload my time is quickly filled up. Once we have an idea of my job in terms of the tasks I must complete, we can then use this later to consider ways to replace them with automation.
When you first go to see a doctor, they will inevitable begin by asking you a series of questions about what the problem(s) you have come to see them about. This is the medical history, a series of questions designed to whittle down a list of possible causes for your symptoms to one or a few possibilities. If you present with chest pain, the doctor will then ask you a series of questions to determine whether the pain is caused by your heart, lungs or maybe the muscles in your chest wall.
Similarly, when you come to see me as a psychiatrist, I will ask questions to try and determine whether you are experiencing anxiety, depression, psychosis or any other possible causes for your problems. While in theory a psychiatrist will start with all possible diagnoses and then rule them in or out, in reality we can be a little more efficient. It is unlikely teenager with have dementia, even less likely a man with have post-natal depression. By ruling conditions out and asking more questions to clarify what has been said already, with time we can whittle down the diagnosis.
Mental State Examination
The information you tell your doctor is often a history of what has been happening to you. You will say when your symptoms began, how they might have changed with time, what time of day they are worse etc. It is quite possible, that when you are actually in the clinic your symptoms will not be present. I am sure many of you will have had experiences of going to a doctor for a problem only to find it has cleared up by the time you get there.
To get an idea of how things are at the time you are being seen, and to further guide diagnosis, you will often be examined. If the problem is with your chest, then the doctor will likely get out their stethoscope, if it is a rash, then they might get out a magnifying glass to see it in more detail. When it comes to psychiatric problems, we perform an examination of your mental state. In this, we are looking for outward signs of what could be happening in your mind, clues to how you are thinking and feeling. Imagine for example someone is depressed. It is common for people who are depressed to stop eating, so if someone is sat in front of me and has clearly lost weight recently, this would support a diagnosis of depression. Similarly is someone appears distracted and is responding to something I cannot see, it might suggest they are hearing voices.
While each psychiatrist will have a slightly different approach to the mental state, they are broadly similar. My approach is to comment on the following features;
- A person’s appearance, such as the clothes they are wearing, and their behaviours, such as whether they make eye contact or whether they can remain still for the assessment
- A person’s speech, including whether it is abnormally loud or quite, fast or very slow, or in extreme case whether they are speaking at all
- A person’s mood, both as they have described it, but also how they appear. Someone might say they are depressed, but have smiled and laughed and joked throughout. It does not mean they cannot be depressed, but it might make a diagnosis less certain
- A person’s thoughts, which of course relies on what they say and do during the appointment. It might seem difficult to assess someone’s thoughts, but actually it can be fairly straightforward. Can you follow what they are saying? Is it coherent? Are they describing any particular worries or fears or beliefs? All of these would give you an idea of the thoughts they are experiencing
- A person’s perceptions, namely what are they seeing and hearing and smelling etc. Crucially, it is important to note if anything is odd about their perceptions, for example are they responding to a voice which others cannot hear
- A person’s cognition, which in effect means their memory and other cognitive functions like their concentration and attention. I might not comment on this with every person I see, but for some people it can be important
- And finally a person’s insight into their mental health. Do they believe they are unwell? And if they do what do they believe the cause to be? This can be particularly useful when considering treatment options. If someone does not think they are unwell, it is going to be more of a challenge to convince them to take medication.
Once I have gathered all the information I think I need, both from the history and examination, but maybe also from medical records and from family and friends as well, I need to make a decision about diagnosis. In some cases, the cause if fairly clear cut, I make the diagnosis and we begin to consider treatment. At other times it can be less clear. I might be left with a set of options, I might even have to try a particular treatment first and see what the response is to confirm the diagnosis. If someone is given medication for depression, and after a month or two feel happier and have more energy, it is likely they were depressed to begin with.
To aid diagnosis, we have a selection of diagnosticguides available to us. In Europe we predominantly use the ICD 10 while colleagues in the US use the DSM 5. Although there are differences between the two guides, effectively they are a collection of possible mental health problems, with a list of symptoms and signs which should be present for each condition. While we might not always follow the criteria to the letter, especially so in Learning Disability Psychiatry, most of the time a diagnosis is given is all of the criteria are met.
With a diagnosis, we then need to think about what treatment options we might want to consider to deal with the problem. As a doctor and psychiatrist, medication are a mainstay of the treatments I suggest, but it might also include recommending the likes of psychotherapy or ECT depending on the condition and severity.
Arriving at a treatment decision should be a collaborative process, and my role is to suggest the best available options, while the patient considers which treatments they prefer, often guided as much by the side effects as any of the benefits. If things work well, we decide on a treatment choice we are all happy with, before I get out my prescription pad issues the prescription.
Whenever you start or recommend a treatment, it is important to make sure an appropriate monitoring plan is in place, both to check the treatment is working, and also to make sure any side effects are being well managed. For me this would usually involve arrange to see people again in clinic, but it could also mean I ask a nurse colleague to see them at home, or their GP to review them in a given time.
Monitoring treatment also gives me a good opportunity to catch up in general with how the patient is doing, whether there have been any important changes in symptoms or circumstances, and to make decisions about how to move forward. It can also be a good opportunity to provide what can often be seen as one of the most important parts of my job; time to listen to the patient, empathise with them, provide what comfort I can in their time of difficulty. It is not an exaggeration to say that the human contact can be a very important thing for people with any illness, but especially mental health problems when isolation can be a significant component. This whole post is considering how I could automate my job, and to do this effectively, this human contact needs to be taken into account.
The other stuff
The core of my job is assessing, diagnosing and treating patients. Everything which goes around this, from meetings to tribunals to reviews, is to aid and better the patient care we can give. Given that in an automated world many, if not all, of these meetings would be obsolete, it seems pointless describing them in any more detail here.
My Automated Job
Hopefully, the last section has given you a reasonable idea of what I do for a job. I have tried to keep it brief, it has taken me a total of 13 years medical and psychiatric training to get to this point, so any more depth to my role could end up leaving you comatose.
Now that we all have a better understanding of the day to day workings of the job, it is time to consider how the human, or put another way me, can be removed from the process. As I have said above I do not think we are at a point now to begin entirely automating my role, so some of the suggestions and technologies might not quite be up to standard just yet, but I do not envisage the technology is too far away. If you had suggested even twenty years ago that we would have internet enabled devices which we can carry around in our pockets within the decade, you would likely have been laughed at, and yet now I sit here writing much of this post on my phone.
Automated History Taking
When I was describing the process of history taking above, it may well have seemed familiar to you. For anyone familiar with algorithms, you will almost certainly have spotted that taking a history is effectively a complex algorithm.
For those not familiar with algorithms, they a series of questions or instructions which allow people, and crucially machines, to make decisions. As a simple example, if you asked someone whether they wanted a cup of tea, the answer would be yes or no. If they answer no, the cup of tea algorithm would end, but if they said yes, the next question might be with milk, followed by with sugar. Each time, by answering the question you will be led along a slightly different path, until you arrive at the exact cup of tea someone wants.
Computer programmes are effectively complex algorithms making decisions. A programme might say, if the enter key is pressed, move down one line, or if backspace is pressed, remove the last letter. Because computers are, well designed to run on algorithms, and taking a history is a complex algorithm, it follows that a computer should be good at taking a history. You could imagine writing a computer programme which asked you a set of questions, and by answering those questions the computer could arrive at a diagnosis.
If you have ever used something like the NHS Symptom Checker, this might sound familiar to you. On this website, you answer questions, allowing the website to recommend the best course of action for your problem. The issue, which can often come up, is algorithms struggle with nuance, and not every medical problem presents in a clear and precise way. If you have inputted a certain list of symptoms which are characteristic of depression, anyone who says they do not have those symptoms will be told they do not have depression.
The problem which any doctor will tell you, depression can present in a number of different ways. The way to get round this seems to be twofold. Firstly, adding more data points to your algorithm can improve the accuracy and account for some of the nuance which exists in life. The second solution is to offer the diagnosis as a range of probabilities rather than a definitive diagnosis. It might add a little uncertainty, but as a human I cannot often certainty every time anyway.
As a brief aside, it seems prudent to consider one of the most common objections people have to automation, particularly in fields such as health care but also in things like driving, namely failure rates. If the machine you are using to make a diagnosis fails, either entirely or arrives at the wrong diagnosis, it can obviously be catastrophic, and people cite this as a reason to always have a human available to step in. This of course ignores the fact that humans can fail too, and while having a human present as well might reduce the chances of failure, it cannot remove it entirely. Imagine a scenario where humans fail about 1% of the time. It would only take a machine to have a failure rate of 0.1% to see a tenfold reduction in failure. Given the precision which can be achieved with technology, it seems reasonable that failure rates will be much lower than any human could achieve, the challenge is going to be convincing the public this is the case.
Developing history taking algorithms is not outside the possibilities of modern technology, and it could even be dressed up as a human interface asking the questions rather than just ticking boxes on a webpage. Reaching the point where all the questions have been asked of a patient before they arrive in the clinic seems entirelyplausible, and I suspect we are not too far away from this part at least becoming a reality.
There is definitely a skill and art to examining someone. Using the stethoscope correctly, moving joints in such a way as to not cause undue pain but to also examine what movement the patient has, knowing that the repeated glances to the corner and muttering might be more than just nervousness can take years of practice and experience, and replacing this entirely with a machine could be challenging.
Physical examinations, such as examination of the heart, seem more at risk than a mental state. Many of you reading this will be wearing some form of device which records your heart rate and rhythm, and it is not inconceivable to imagine devices which can go further and listen to your heart sounds as well. Technologies which give an idea of a person’s health without ever been touched by a doctor have been in development for years, and I see no reason this will not continue.
But what of the mental state exam. How can we get an idea of someone’s mental state using technology alone? Assessing someone’s appearance and behaviour seems fairly straightforward. Increasingly police forces and governments are using facial recognition software to spot and track people through crowds, adding in a description of their behaviour (agitated, slowed, etc.) does not see a stretch. Given the correct technology, a computer could well give a more detailed picture of a patient with just a moment of footage than I could give after an hour with the patient, and crucially tracking how this changes over time and with recurrent visits would be much easier. Assessing someone’s speech can also be done fairly easily with technology. Recording speech and then assessing the rate and the rhythm is well within the realms of even commercially available dictation software today.
Assessing someone’s mood and thoughts and perceptions can be more challenging, but in reality all I am doing is looking for clues in what they say and how they act during the appointment. So if you have software to record and analyse someone’s appearance and behaviours and speech, an algorithm could be developed to use this information to reach a diagnosis. If someone sits with their head down throughout, this might suggest they are depressed, and this can be fed into the algorithm with all of the other available data.
Automated Diagnosis and Treatment
Reaching a diagnosis and treatment plan in this automated world will be one of the end points of the algorithms described above. Once enough questions have been asked, and enough observations recording, a computer would then arrive at a conclusion based on the path it had followed through the algorithm. In an ideal world it would give just a single answer and treatment, but it might well be that it has to offer a range of possibilities, in particular when it comes to treatment choices, and this seems OK. It is at this point the patient would then have to consider the treatment choice best for them.
Just as another brief aside, we are rapidly moving towards a world with personalised treatment choices. Developments in genetic testing have seen advances in our understanding of not just a drugs effect on the body, but how our bodies respond to a particular drug. Today if you have high blood pressure for example, you will be offered a medication choice from a small selection. In the future, it is reasonable to expect the exact drug you should take based on your genetic profile will be available to you. I have not discussed blood tests and the like in this post, partly because they are used to rule physical health problems out more than anything in psychiatry, but an automated testing machine, covering blood but maybe even genetics as well, does not seem outside the realms of possibility.
Monitoring the response to treatment would require nothing more than the technology used to make the diagnosis in the first place. At regular intervals, maybe weekly or monthly, a patient could be asked to answer a series of questions about symptoms and side effects, and from this the computer would be able to determine how they had responded, and whether any adjustments needed to be made.
What is left to discuss, is probably the biggest hurdle to removing the psychiatrist at present, and this is the human interaction. It cannot be overstated how important this can be for patients, so in any automated world it would need to be replicated as well.
While AI technologies are getting better at fooling people into thinking they are human, they still lack the warmth and responsiveness a person has when they are meeting someone face to face. Replicating this seems a long way off to me, but that does not mean alternatives cannot be sourced.
One obvious solution is to replace the psychiatrist with a listener. It can take years to qualify as a doctor, and as a result doctors are expensive commodities. If you have replaced their technical functions with machines, then it would not be a problem to replace the expensive doctor with a less well qualified person who could offer the patient the human interaction at the same time the machine is doing all the work making the diagnosis. There has been a shift to a degree in healthcare of moving roles from expensive doctors to less well paid alternatives, so it should not be a stretch to do this when the decision making power has been given to a machine.
Of course if you want to remove the person entirely from the situation, then the development of better, more human like interfaces for computers will be needed. This might be a real person’s face made to act and respond to the patient, or else something entirely computer generated, maybe even using virtual reality software. I am not sure any of these options will entirely replace real human interaction, but if diagnosis is being made quicker and more accurately by the machine, and treatments are targeted to the individual, the human interaction part may be less important anyway.
To the future
I do not think anything I have described above is outside the realms of possibility within the next fifty years or so. As machines become more accurate and efficient, I suspect people will begin to demand more automation from healthcare, and the current reluctance to remove humans entirely will begin to crumble.
I have written this post to show how it is possible that a job which seems very much focussed on having a human involved could well become automated, at least in part in the short term. If you think your job is immune from automation, hopefully this post will give you pause for thought and begin to consider what the consequences could be when the machines take over.