Brain Computer Interfaces

My father was a civil engineer, and my mother was a nurse. So I had a mixture of the hard physical sciences and the more clinical and compassionate perspectives on what it was to be a good citizen, a good child, and eventually a good grown-up.

In my teenage years I became fixated on what would nowadays be known as mathematical psychology — an effort to understand the brain, or more generally our sentient cells, but on a formal basis that would keep my father happy. So that's where I started from, and I took all the careers advice I could to equip myself with the right training to ultimately navigate towards where I am now. I could tell you a funny story about that, if we have time.

In the UK, a lot depends on your university course and your Bachelor of Science or Arts, and that in turn depends on the A-levels or Highers you go for to secure a university place. At that point there was a lot of earnest careers advice. I was an experimental subject in one of the very first attempts to use computers, machine learning and artificial intelligence to guide aspiring young students. So I spent the entire day — filmed by the local TV station as this entrepreneurial application of AI to guiding young lives — answering questions about my aspirations and what I'd need to do what I wanted in life, probably very much like the way I'm going to be answering your questions.

I entered it all into the computer, off it went on cards to the University of Liverpool, and then there was the big reveal as to what the perfect job would be for me. My perfect job was a television aerial erector — which had everything in it: electricity, engineering, being outdoors.

So I didn't particularly follow that advice. I was then sent off to a human artificial intelligence for some better guidance. I explained exactly what I've explained to you, and she said, "Ah, Karl, what you need to be is a psychiatrist" — not realising, as I didn't, that there's a fundamental difference between a psychiatrist and a psychologist. So I thought, fair enough. Off we went. Of course, that meant I'd have to apply to medical school. So I spent — I was going to say wasted, but I think that would be disingenuous, though certainly wasted to a certain extent, in terms of my physics and maths education — ten years becoming a doctor. After that I got into academic medicine as soon as I could, in particular academic psychiatry via brain imaging. And then, as soon as that had earned its keep in terms of contributions, I moved to where I really wanted to be: theoretical neurology, and what would now be known as computational neuroscience, artificial intelligence and machine learning.

Yeah, absolutely. That sort of bilateral skilling — on at least two sides of the things you can know — is going to be a central theme in my responses to all of your questions. And it was a very explicit part of my choice of university course. I remember saying in the interviews that I wanted somewhere that would let me study natural sciences, so I could actually study physics and psychology at the same time. The only medical course that allowed that was the place I was eventually successful in securing a studentship on. That theme has been implicit and explicit throughout my career — and, I'd add, through the careers of most of the young, bright things it's been my pleasure to supervise and watch flourish in academic medicine, particularly academic neurology.

Having both — and I have to say it, a lot of your audience are going to tune out at this point — but basically maths is the ultimate language. So I thought maths and Latin would get me through life. The Latin wasn't quite so useful, apart from doing dissection. But it's very useful just to have that formal background. You don't have to be terribly skilled at it. A lot of my contributions have been denoted as mathematical contributions, yet, to be quite honest, the kind of maths I use I get from Wikipedia, not from the physics and quantum physics and probability theory I did at university.

That's a great question. In a sense, I've never left the clinical side — that's why I stumbled when I said it was a waste of my time. As an aspiring and failed physicist, I certainly did waste time clinically, but in terms of having a good and proper life it was an essential move. The motivation for much of the subsequent technical work was to understand how you'd explain psychopathology in terms of the pathophysiology of neurology and psychiatry. My special interest was schizophrenia, and that was honed by my early training experiences.

After the preclinical at Cambridge, the clinical at King's College London, and house jobs in the London area, I went up to the Oxford rotational scheme in psychiatry — simply because it was the closest thing to brain research that wasn't awfully competitive. Everybody wanted to be a heart surgeon in those days, and then a GP; the fantasy was that GPs were not only the right way to show your clinical care but also became very rich. I'd also got a bit distracted at university and medical school by amateur dramatics, so I wasn't in the top tier in terms of results and had to go to something less competitive. Psychiatry in those days was quite a Cinderella subject.

So off I went, and found myself for two years in a therapeutic community of chronic schizophrenics. You're literally living on site in a little Victorian cottage, group therapy twice a day, lots of smoking, lots of psychosis. It was an immersion into a different kind of world, and that's never left me. All of my work since then has really been in the service of trying to understand those false beliefs — the delusions, the hallucinations — even at an elemental motor level of the kind you see in motor disorders. Generally, neurology and psychiatry treat broken inference in the brain, as an organ of inference. All the techniques my team and my younger protégés have developed are ultimately just tools to understand what would nowadays be regarded as a synaptopathy — that is, schizophrenia.

Well, yes — it's very kind of you to cast me in that light, and you're absolutely right from a personal point of view. But there's also a truism there beyond my own career. In any academic field of this sort, in the life sciences, where you're responsible for making breakthroughs that matter in the sense of improving well-being and medical care, you're always going to be more empowered if you're clinically qualified.

Even if you wanted to be a thoroughbred MRI physicist when you grow up, or the ultimate in computational neuroradiology, you'll still have that edge if you can get through your finals — and if you can get a membership of a college, even better. Simply because you've got an appreciation of the things that matter to people and the ways people can go wrong in their bodies or their brains. It also gives you access to certain ring-fenced funding for research that you wouldn't otherwise have. It's a really hard thing to do — we can talk about the difficulties of combining an academic with a clinical career — it's such a niche thing. But if you can squeeze your way through, you end up in a much more empowered position.

The usual advice I give is: do as you're told. That usually works very well, especially in a family context when you get older. But less trivially, doing as you're told in a sensitive way means you're alert to the questions and problems people are currently contending with. To put it another way: you're only going to be successful if you make a useful contribution, and your contribution is only successful if it solves a problem or answers somebody else's question. To make a difference, you've got to answer someone else's — or another field's — questions in a way they couldn't answer themselves. So it's very much a reactive, or response, mode of science. Everything I'm known for, and indeed our group is known for, is largely because the stuff we do lets other people do what they do slightly better.

There are other fields where that doesn't really work, and I'm thinking rather cruelly here of philosophy. The whole game there is to get people to listen to you and not to themselves — but you're not really helping anybody else unless there's a collective aspiration to become, as a community, all great philosophers. It's much easier in the physical and life sciences, and particularly in medicine, because there are constant and ongoing advances you can contribute to. So the bottom line is positioning and equipping yourself with the right skill set to respond to what's needed in the moment, at this point in your career. And that means your career is going to be very much shaped by the context you find yourself in.

An example that's marked my career is human brain mapping. At the point I started to earnestly seek opportunities to help, brain mapping had come online — first with Positron Emission Tomography and subsequently with functional MRI. That was a real paradigm shift, and I just happened to be there at that time. Had it been ten years later, it might have been optogenetics. Had it been ten years before, it might have been neural networks. I can't remember, because I was too busy studying at the time — so probably a bit like you.

The reason I'm saying that is it's important to allow for a certain context sensitivity in your career trajectory. It's difficult to plan the perfect career, because that perfect career is self-building in a context over which you personally have no control and no particular knowledge of what's going to happen. What that means is you'll be better placed to make the most of any given opportunity if you've got the broadest, most eclectic things to bring to the table — which brings us back to that bilateral, multilateral training. You see it implicitly in the way students' career structures are engineered, particularly in medicine: you're encouraged to do this kind of course and that kind of course, to wait a few years before specialising, and even once you specialise you have to do a few general years. That's all to broaden as much as possible a pyramidal base of experience on which you can build and then successively narrow down, given the opportunities and problems that present themselves.

Well, you should clearly only follow your own interests. But within that constraint, follow the direction your interests point in from as many perspectives as possible. I mentioned maths before; nowadays it may be more computer science — becoming fluent in information technology and, in a few years, artificial intelligence and general AI. A technical, hard skill set that is foundational in nature. I use that word very deliberately.

What we're talking about is acquiring, through those early years of wonderfully enhanced neuroplasticity that you're currently enjoying and I have now lost, a set of foundational trainings. About five years ago, with the emergence of big data, there was a backlash against the "data scientist" meme — this idea that we needed a whole herd of data scientists, none of whom had any foundational training in anything. They didn't know molecular biology, they didn't know mathematics, they didn't know clinical medicine. The foundational part — the good old-fashioned understanding of the way the world works, the anatomy of the thing you're interested in, its dynamics — is absolutely essential. For me, maths really underpins it, but you can lift maths to the behaviour of mathematical schemes in silico by becoming fluent with computers, information technology, neural networks.

Of course, that in itself is a great foundation, but unless it's actually aimed at something, it's not well defined. So we come back to it: it has to be something you're interested in. And in the biological and clinical sciences there'll be a lot of foundational issues you have to get on top of — the biochemistry, the anatomy — which is presumably exactly what you're doing at the moment. All of these are different ways of modelling, understanding and formalising the way things work, that you can then put together with a computational or mathematical skill set to really make a difference — diagnosing, predicting, simulating, whether in pharma with computational chemistry or in epidemiological modelling.

My mind goes immediately to the more contentious aspects of the free energy principle, which people find difficult to swallow and therefore implicitly think is daft. But one thing people find difficult to accept — in the part of my world that talks to philosophers — is the notion of consciousness as a process, as opposed to a state.

I'm not sure it would be seen as crazy, but consider simply equating consciousness with a process, in the same sense as, say, evolution. If I'm confronted with some description of consciousness — one of the highest aspects of a sentient brain, and the focus of much of my current research — I want to understand what somebody is trying to say. So I replace it with the word "evolution" and see if it still makes sense. Evolution is a process. It has an objective, which in a deflationary sense isn't really there because of the circular causality in evolution — but you can cast it as having a functional objective and the appropriate dynamics. When I read people talk about conscious states, I try to think: what is an evolutionary state? What is a state of evolution? Does that make any sense?

For me, that would be the best example of something regarded as crazy. There is no conscious state in the sense that there is no current state of evolution. Evolution is something that unfolds in time; it's a process you have to understand. That may or may not be a crazy idea, but if you commit to it, you can get quite a long way in understanding some aspects of higher cognitive function, self-awareness and sentience.

The practical implication — and here we come back to, did you ever stop being a clinician? — is that when you're dealing with psychiatric conditions, nearly all the symptoms and signs, more generally the phenomenology, can be construed as abnormal belief-updating. If you were a statistician, you'd call it false inference. And I mean that in a very simple way. Perhaps this is another crazy idea, but it seems obvious and compelling in its simplicity.

Perhaps you haven't done statistics yet, but when you do, you learn about type 1 and type 2 errors — inferring something is there when it is not, or inferring something is not there when it is. That's exactly what characterises nearly all of the psychopathology you see across a whole range of neurological and psychiatric conditions. Take psychosis and schizophrenia, characterised by delusions and hallucinations. What are these? In essence, inferences about states of affairs in the world that aren't there: the delusion that some government agent is compelling me to act against my will, or the voices I hear that aren't there — and yet I do not know that that's the case.

Another example is the agnosias, whether from functional medical symptoms or true hemi-neglect syndromes in unilateral brain damage — inferring something is not there when it is. "I can't feel my arm, doc." Or this side of peripersonal space just doesn't exist for me. Or, "I don't know who you are." There are psychiatric syndromes where you have a sense of familiarity but can't explicitly identify who you're talking to, or the converse — where a loved one has, you're convinced, been replaced by an impostor. These are all examples of false inference.

So if you want to understand belief-updating in the brain in the context of a condition like autism, schizophrenia, Parkinson's disease, or a dissociative syndrome of a neurological or functional sort, you need to understand the neuroanatomy, the neurochemistry and the neurophysiology of this belief-updating process that has gone wrong. Nearly everything explainable at this formal level can be reduced to the process of belief-updating.

And in that sense your consciousness should be no different. Then you get into the interesting worlds of altered states of consciousness — from the normal physiology of "why do we sleep?" or the therapeutic interaction between mindfulness training and a serotonin reuptake inhibitor, right through to locked-in syndromes, coma, and the altered states you might see in severe depression or psychosis. In particular, abnormalities of consciousness in the lay or folk-psychological sense, which is usually read as a sense of self, of selfhood. You need to understand what kind of inference process — what evolution of beliefs — underwrites a particular construction of the world where you cease to be yourself. That's what's happening in depersonalisation, or delusional mood in psychiatry, and possibly things you've experienced taking recreational drugs. A lot of our grip on the world is only revealed in terms of the inferences we bring to understand it — when, transiently, we come off the rails and experience an alternative hypothesis, like "I'm not me" or "the world isn't real." When these endure, due to abnormal neurochemistry or psychopathology, they can become incredibly clinically burdensome. And if you want to address that with behavioural therapies or pharmacotherapy, you really have to know the nature of the process you're contending with.

That's a good question, and I'm not sure — you'd have to ask a philosopher who deals with states of consciousness how they'd deal with this person in a coma, or that person suffering from visual hallucinosis. I don't think the academic and philosophical treatments of states of consciousness really have as an explanatory target what you have in mind as a burgeoning clinician. I don't think they'd even go there. They probably stick to an understanding of sentience in terms of qualitative experience — what it's like to be a bat, and all those great thought experiments that underlie a lot of philosophy.

It's an interesting question that's actually starting to have practical relevance. In the past few months various people have been trying to fund adversarial consciousness research — taking groups and disciplines that hold opposing or complementary views (consciousness as a particular state of information or message-passing, versus the view that that message-passing is in the service of belief-updating, of keeping a grip on the world and our body) and asking them, adversarially, to jointly propose experiments that would falsify one or other approach. That's extremely difficult, but also extremely important if you want to take the basic research and translate it into academic psychiatry. So I think it's still an open question.

Just as a worked example of the importance of fields you might not think are relevant — the philosophy of consciousness for your work — there's a current focus on the potential use of 5-HT2 serotonin receptor agonists and partial agonists, like psychedelics, and their therapeutic role. They clearly speak to this notion of consciousness as a form of belief-updating and making sense of the world, because they can have profound effects on your perceptual synthesis and the way you handle things. There's been particularly interesting work in end-of-life care, changing the way people conceive of and model themselves when they know they're going to die — using psychedelics as a way to revisit some of those hypotheses and put a different perspective on that part of their life, with remarkably effective outcomes. One session can have a tangible impact for many months.

Another great hope is that these drugs, which affect the modulatory neurotransmitter systems in the brain, may be therapeutically very relevant in conditions like depression — right through to the motor side. Things like Parkinson's disease rest not only on dopamine but on a number of modulatory neurotransmitters, and understanding the physiology that underwrites how the brain makes sense of its inputs from the body and world is absolutely essential before you develop hypotheses to test. But I repeat: finding the definitive, killer adversarial experiment that will tie your consciousness down to a process of belief-updating about your lived experiences is still slightly outside our grasp, I'm afraid. Probably we come back in five years and they may have found one.

That's an excellent question. My first instinct is slightly deflationary, and I motivate that cautionary position by appreciating the importance of what's known as the enactivist turn. In the brain and life sciences, and certainly in cognitive neuroscience, at the turn of the century — and indeed the millennium — there was a move towards enactivism, embodiment, situated cognition: an appreciation that your body matters, that the way you sample the world actively with your eyes and muscles — when you look around, it can be seen as a form of visual palpation — is actually where a lot of the heavy lifting of understanding and living lies.

If that's true, if that emphasis on the body as the way of interfacing with the world is the right way to conceive of self-organisation in sentient creatures like you and me, then it suggests the notion of a brain-computer interface may not necessarily work — in the sense that we are already the product of evolution, where selective pressure has been specifically focused on providing the best interface between the brain and the world. You are your embodied BCI, minus the computer part. To re-engineer that is going to be extremely difficult.

It's not inconceivable — there are some beautiful experiments using sensory substitution. I don't know if you've come across the work of my friend Peter König in Germany. Several years ago he equipped people with a belt of magnetic sensors that stimulated the tummy. After a few weeks they were able to feel the electromagnetic fields around them, and to perceive through this new modality — a sensation of a magnetic field — integrating it into their world model of where they were in a building or a city in relation to north, south, east or west. The brain is certainly plastic enough to interface in a new way with a new kind of sensory input, given the right sensors. If that capacity is preserved to the extent that you can plug microelectrodes into the visual cortex and feed it from an artificial retina, then in principle you can imagine a world where a very, very fast BCI of that kind is feasible.

But it does go against the embodied perspective — that most of my brain is really engineered, and has grown up through evolutionary time and through neurodevelopment, to handle the deployment of my sensory organs. Which means that if you substitute those sensory organs, they crucially have to be as manipulable and deployable as my own. This comes back to the active aspect of, for example, vision. It's not just having a good camera and processing all the pixelated information that impinges on the photoreceptors — that's the easy part. The real problem is knowing how to control a camera, where to point it, where to look, and how to control that as a physical object you use to navigate your world. So if any BCI is really going to be effective, it will have to be sensitive to these constraints. Advances in neuronal prosthetics will have to be informed as much by the careful engineering of a prosthetic that can be deployed and used physically, online, as by all the computer science and neuroplasticity that lets the brain make use of it.

Yes, I think so, absolutely. One of the things that consciousness — mind as a process — brings to the table is that it's just one of two dual aspects of neurophysiology and neuronal dynamics, instantiated in the wetware of our brains. What the enactivist view adds is the way that is physically coupled to its sensory organs and actuators, both inside and outside. We're not just talking about eyes; we're talking about autonomic reflexes that subtend literally gut feelings. All of these things come along as an inseparable bundle — an inseparable, self-organising mechanics you can't pull apart — but at least you can understand it as a conscious process of belief-updating on the basis of sampling new information from within or outside the body. If a particular brain-computer interface doesn't appreciate that, and there's a particular focus from either the mathematical or the fabrication side of the engineering, then it's unlikely to be successful.

You're right. Not only a nice science-fiction thing to contemplate, but clearly important for people with quadriplegia and paraplegia. There's an enormous amount of work on spinal cord injury and what kind of engineering and BCI is appropriate. The other obvious application is people with locked-in syndrome — trying to access what they're thinking, and at least ascertain whether they're minimally conscious, vegetative, or in a deep coma. So there are lots of clinically important applications.

But my scepticism inherits exactly from those examples that rest upon the body. What's gone wrong is the body — and can you re-engineer a body? In principle you could, but it would have to be the kind of body the brain has already learned to drive. I haven't thought about this example before, but let's use it as a metaphor. Imagine somebody took your car away, yet you had to drive to survive; you weren't allowed your own car and were given a substitute that wasn't drivable. You can see immediately it's going to be very difficult to live a normal life and interface with this non-drivable car.

So what do we mean by drivability? We mean you have the right kind of brain and the right experience-dependent configuration to manage the two-way exchange between you and the car automatically. I know which levers to push, exactly how to turn the wheel, how far in the future to anticipate when driving. Unless all of that is refabricated in a way that's become instilled into your nervous system — through years of experience, or if you're a young, good driver, months of learning, plus of course millennia of evolution — unless you can leverage what you are as an embodied brain in the physical apparatus that now constitutes the BCI, you're not going to get very far.

Which brings us back to my observation: you can deviate from your actual body, or your actual car, provided you don't deviate too much, and the brain can handle that through rewiring and neuroplasticity. But that puts special constraints on how far you can deviate, and on the kinds of patients these approaches will be useful for. Notoriously, children are much more plastic — they can acquire two, three, four languages contemporaneously in a way that would be impossible a few years down the line. So, going back to the car: if you're of an age where you can't relearn to drive or learn a new language, you're not going to be able to learn your new BCI device.

What do people generally do in that context? They normally try to increase the synchrony between the fabricated actuators and lift off very peripheral signals — using EMG, electromyographic signals from the forearms or upper arms, to drive actuators and little servos in an exoskeleton. You've seen wonderful examples of that on television in the past few months, enabling people to walk again. But they've been sensitive to the fact that the signals they use are the ones that would normally drive your actual legs — your actual biological car — as opposed to re-engineering some artefact that takes the role of a car you've never learned to drive.

Yes — that's inventive, and if I understand the question correctly, a different but exciting take on BCI. In a sense, are we not seeing that already? I'm mindful here of the notion my friend Andy Clark brought to philosophy and beyond — extended cognition. Very simply, my memory can now be downloaded into the environment and put in my iPhone. If I want to remember your phone number, I no longer have to rehearse it using my muscle memory; I know it's in my iPhone. Strictly speaking, you've downloaded your cognitive capabilities into the physical environment — an electronic device, a little computer. You've extended the reach of your cognition by encompassing a computer.

Take that idea and fast-forward to 2020 and the enthusiasm for robotic companions that emulate a lot of our interactions — creating a sense that this is something that cares for you and can relate to you, that there's a shared narrative between you and the computer, whether in caring for the aged or in the hospitality industry. I'd see that as another example of a computer interfaced with you in a meaningful way, requiring reciprocal exchange mediated by the good old-fashioned biological interface — speaking, looking, feeling, touching — but with a computer.

And there are principles you can bring to bear on that computer that inherit from the very inception of cybernetics. The architects of cybernetics, in England at least, were responsible for something called the good regulator theorem, which basically says that any system that manages to survive, self-assemble, or exist in an environment must at some level be a model of that environment. That means if you want to coexist with a computer in a meaningful way — you control it and it controls you — then that computer has to be a model of you, and you have to be a model of it. The simplest way to make that true is to recapitulate the structural form of the belief-updating, the message-passing, the physiology — at least as expressed through movement and communication between the two things being coupled. And then you get into a more general sense of why we have language: simply because we have a shared model of the world, a shared narrative, which makes us more mutually predictable.

A BCI that would cover compassionate robotic friends or pets means these things are now being engineered by start-ups with quite considerable venture capital funding — if they're not already on the market, they will be within months. So the brains of your computer have to have a formal similarity, in space and time, to your brain. That tells you quite a lot. We know our brains have a particular hierarchical, sparse connectivity structure — sometimes referred to as the connectome — that embeds a segregation and a particular kind of integration between different parts of the brain, of a hierarchical sort. People often talk about the visual cortical hierarchy. Which means your computer also has to have a hierarchical organisation.

And again — that's already happened. The very success of deep learning rests on the deep architectures of these neural networks. What does "deep" mean? Simply that they have more than one hierarchical level; it's hierarchical depth that defines a deep architecture. That's exactly the structural aspect of our embodied brains, which model the world in this hierarchical way. If you take that to its limit, you want deep-learning algorithms in your robotic pet that recapitulate, more and more, the deep generative models in our own brains — so they come to share a narrative, both in language and in the language of the body, the physiology, the autonomic responses.

This is not outlandish — people are working on these ideas right now. You get robots that blush, robots that have heart rates and synchronise with you, because you have a shared model of exchange. You are me and I am you, and it's just a question of working out whose turn it is to talk. That operates at all levels. If you did a Fourier transform of the audio stream of this exchange, you'd see me talking and you talking — probably me talking more than you — but at least there'd be a discernible rhythm to it, and also at the level of the thoughts and ideas I'm conjuring up, and you conjuring me through your questions.

These are both statements of the importance of a shared narrative — a consensus we can use to understand each other. If you can put that into a brain-computer interface, then you're talking about something that would be truly useful as a pet, or perhaps a clinician. And I'm not joking — many of these fundamentals are already in place. That synchronisation between patient and therapist goes beyond spoken language: right through to the way we move, nod and blink, the way our heart rates synchronise. All of these should be as synchronous and mutually responsive as possible when you want to build a good therapeutic relationship — certainly in dance therapy, cognitive therapy or psychotherapy, and one could argue a good interaction with your general practitioner too. I'm not so sure about a surgeon.

Well, I'm a Yorkshireman, and typically we prefer to "say nowt, hear all, see everything." The real reason is I'm quite shy, basically. But it's important to talk, and the most efficient way I find to talk is in a large group. That may inherit, again, from what you asked about earlier — my experience in the therapeutic community, where, if you remember, we had group therapy twice a day. That was a very formative experience. There's a vast literature on the power of groups at many levels, from social neuroscience through to politics, but certainly in therapy in the psychiatric domain, the dynamics of a group are incredibly important. I've found it's also the most effective way to spend social time in teaching, supervision or academic collaboration — which leaves a lot more time for thinking, and smoking, and cups of coffee, which is what I spend most of my time doing.

Well, just having big, malleable groups. I'm not talking about one group — I'm talking about a very big extended family of academics. In my job, most of the academics I speak to are either clinically qualified, aspiring to be, or related to healthcare fields in their interests. It's a network that spans from Melbourne through to Toronto, from Aberdeen through to Sussex, and it changes week by week. What's always in play is the notion that anybody can join at any time, depending on what's going on.

I take your point that groupthink may be constraining — that's an interesting argument, in terms of the eco-niche construction that comes out of a shared narrative — but for exploring new options and hypotheses it's obviously a bad thing. So the way we do it practically: you listen a lot to one person for an hour, you try to get inside their head, then you have a group discussion, and then I normally summarise what I've heard. Interestingly — I hadn't thought about this — it's almost exactly the same as my early training in psychotherapy, where you listen to the patient or client for one hour, you notice that the hour is up, and the important thing is to listen. You try to put nothing into the mix so you don't contaminate what's being put out there. When you're training in psychotherapy you then have a second, supervisory session where you effectively become the client and do the talking with your supervisor — that's how you learn to do psychotherapy formally and properly. I trained, though I'm not a practising psychotherapist, but the training was extremely useful: being able to listen for long periods and then, on the basis of what you've heard, unpack it. So when I say "a group," it's really a large number of people committing to listening to one person and then trying to unpack it from their different perspectives.

It's all play. Absolutely. If working is sitting in your conservatory, smoking your pipe, and thinking about how the world works or how you work, then it's absolutely play.

All the normal ones you're probably aware of. A certain degree of obsessive-compulsive neuroticism is always quite good — keeping things tidy, organised and crisp so you don't miss anything, and if you do, you remember where you put it. And always being very clear in your communication — which is another reason you have to learn maths: to write down your ideas mathematically so there's no ambiguity.

Something else that inherits from clinical training is unconditional positive regard. You make no judgments; you don't approach anybody, or a new idea, with any preconceptions, to the extent you're able — and it can be quite difficult to spot your own preconceptions. Sometimes you experience this as de-skilling. I'll give you an example: when you're dragged onto the ward and told you're going to be a nurse for 48 hours, the nurses will have to de-skill you immediately, because you'll have all sorts of fantasies about what you should do as a medical student or doctor which are completely unfit for purpose. You have to put those preconceptions aside, and that can be quite difficult at times. After a while you develop the skill of putting them to one side, and that usually comes along with this unconditional positive regard — everything that's going to happen to me, or that you're going to say in the next few minutes or hours, has to be taken very, very seriously. You're only doing it for the right reasons; I just have to work out what those reasons are.

Yes, I do. I'm quite vain in that regard. You can sometimes get quite tired and want a break, but what tends to happen is I feel quite anxious that I'm not using my life properly — and then I ask, well, what's the proper use of a life? It's to leave a legacy. So in that sense legacy is quite important, although I don't ruminate about it. It's a nice heuristic to organise your life around, provided it's not at the expense of interpersonal relationships. Generally speaking, if you keep your eye on your legacy — even if it's just being respected as a good citizen and a person of integrity by your friends and family — then that's the kind of legacy which is a good heuristic to fall back on when you ask yourself why you're doing this or that.