Consciousness: How Matter Becomes Imagination

"This book marked the start of a new chapter in my career—and indeed in my life. In 2001, I had the opportunity to move to the Neurosciences Institute in San Diego, California, to work with Gerald Edelman . Edelman won his Nobel Prize when he was 43 and capitalized on that to be one of the few people bringing a new legitimacy to the study of consciousness in neuroscientific circles. It had been more or less taboo for a long time, and it certainly wasn’t part of what was on offer when I was an undergraduate student at Cambridge. But Edelman, along with Francis Crick and a few other early pioneers—Wolf Singer in Europe and so on—had fought hard to remind people that consciousness is a central problem in neuroscience and that there are ways to address it reasonably, even if you don’t solve the whole thing. Now, one of the reasonable approaches to the scientific study of consciousness is the so-called ‘neural correlates of consciousness’ approach, pioneered by Francis Crick and Christof Koch. This approach was, and is, very deliberately pragmatic. I suppose the feeling was, ‘Okay, let’s just put aside these awkward philosophical and metaphysical problems and just remind ourselves that there are intimate relationships between what happens in brains and what happens in conscious experiences when you fall asleep, or if you undergo binocular rivalry, or whatever the manipulation is, so let’s just look for these correlations.’ This was a terrific idea, at least at the time. You can follow this approach and just remember to be modest about what can be concluded. It’s not a theory of consciousness, we’re just finding stuff out. The ability to do this had a massively salutary influence on the field, precisely for this reason. Of course, it is still methodologically problematic: for example, a key question is whether you’re studying the real correlates of consciousness, or the correlates of how people report about their experiences. Lots of debates rumble on about this to this day. But at least you could get on and do it. A lot of the empirical neuroscience of consciousness since the mid-1990s has followed this route, and productively so. This book takes a different route and that’s why I found it hugely influential and interesting when I read it back in 2000. It was co- authored by Edelman and Giulio Tononi. Tononi is now well-known for pioneering the integrated information theory of consciousness , which builds on ideas in this book. What struck me about it was that it tried to go beyond just establishing correlations, to attempt to explain properties of phenomenology in terms of mechanisms. This seemed midway between philosophy and neuroscience, and exemplified the benefits of bringing both together. The overall approach is something like, ‘Okay, we’re not going to propose some dramatic claim that directly solves Chalmers’s hard problem. Instead, we are going to identify characteristic properties of all conscious experiences, and then ask what properties the underlying mechanisms must have, to account for these properties of phenomenology. They came up with a couple of insights into phenomenology that may seem obvious, but which are actually quite deep. “How things seem is often quite an unreliable guide to how things are” The first is that every conscious experience is different from every other conscious experience you’ll ever have. This means that consciousness is hugely informative in a very technical, formal sense: every experience rules out many, many alternatives. The second is that every experience is unified. All of our conscious experiences are ‘all of a piece’, bound together—when we are conscious there is a single stream of experiences going on. These two properties seem to coexist and characterize every experience, whether it’s looking out of my window now and seeing the sea in the distance, or meditating, or visiting the dentist and feeling a sharp pain as she drills into my tooth. These two properties obtain in all these cases. The book then develops that and says, ‘Okay, well, if there’s a characteristic property of consciousness, that’s doesn’t obviously apply to other things in biology, at least not in the same way, then figuring out the mechanisms that could underlie this property should shed light on the brain basis of consciousness. Specifically, you can ask what kinds of systems co-express integration and informativeness (or differentiation) – and see if these kinds of systems exist in the brain. In the book they point out that the cortex has the right kind of anatomical structure for generating both information and integration, but the cerebellum, which has more neurons in it than the rest of the brain, doesn’t have these kinds of properties. This explains why the cerebellum may not be involved in consciousness, and we know as an empirical fact that it isn’t. I was inspired by this book because, at least for me, was the beginning of a form of consciousness science that had real explanatory potential. There were a few earlier papers that had been developing some of these formal measures of complexity that characterize co-expression of integration and information, but they hadn’t really applied them to consciousness. Reading this book made me realise that we had a real chance of developing a materialist story that takes phenomenology seriously and gets us further than correlation. And it influenced me in a very practical way too, since it was one of the main factors that encouraged me to go to San Diego. So, in 2001, I moved to America for six, seven years and worked on consciousness with Edelman at The Neurosciences Institute. Tononi had actually left San Diego before I arrived, and had moved to Wisconsin where he started to develop his integrated information theory (IIT) which builds on the core ideas I just mentioned, but takes them in a different direction. One of the main distinguishing features of IIT is that it makes a very strong claim about consciousness: IIT proposes that consciousness is integrated information, which is a very ambitious and provocative thing to say. The other thing that’s in Edelman and Tononi’s book is a summary of Edelman’s ideas about neural Darwinism—or, to give it its full name, the ‘theory of neuronal group selection’. This is one of the things Edelman is best known for, but also something that has been notoriously hard to understand. Put simply, it’s a theory of biological selection applied to the brain, during both development and everyday function. We all know from Darwin that natural selection led the evolution of distinct species. But there’s also selection within the body. Edelman’s Nobel Prize was in the field of immunology, where he was known for the idea of immunological selection. According to this idea, antibodies are able to fit to antigens, not because they somehow fold themselves around the invading antigens and then instruct other antibodies to do the same, but because of an internal process of somatic selection. The immune response is generated by diversity and selection within the immune system. It’s a beautiful insight. Edelman then applied the same idea to the brain and asked, ‘Okay, how do neural populations form?’ Broadly, yes. The brain is a vastly complex system in terms of the sheer number of neurons, but more so in the intricacy of their connectivity. It seems very implausible that single neurons end up doing particular things—implementing specific large-scale functions. Everything the brain does involves a network; it takes a village. How do these connections get sculpted? There are so many of them that it’s difficult to imagine they’re all precisely specified in the genome. There’s going to be—or at least this is Edelman’s argument, and I find it convincing—some internal variation and diversity that is then selected on through development and experience. We end up with the fine grained neuroanatomy that we each possess through this process of internal variation and selection. Yes. It’s very different. One of the disanalogies with Darwinian evolution is there’s no obvious mechanism of replication. In Darwinian evolution, genetic information is encoded in DNA, and DNA gets passed down through the generations. What happens during the lifetime doesn’t affect what’s passed down through the germline, at least to a first approximation. In the brain, it’s not clear there is any equivalent sort of encoding and inheritance, though there’s still variation and selection. Still, I think that selectionist principles offer a powerful way to think about brain development and its acquisition and function. Crick was down the road from Edelman and the Neurosciences Institute at the Salk Institute. They had a beautifully sparky relationships, two old grandees of biology, both with important Nobel prizes in their pockets—both deciding to study consciousness. And taking very different approaches: Crick with his reductive correlative approach, and Edelman with his grand theories. Crick infamously dubbed neural Darwinism ‘neural Edelmanism’ when it first came out. Edelman’s books are rarely easy reads. Working there for six years I was lucky to have the opportunity to have daily conversations with him. I read his books a few times and we discussed the ideas frequently, so I was able to put them into context. In general I think that Edelman’s ideas—especially those to do with neuronal group selection—may not have had the influence they deserved, in part because of the style of his writing, but in larger part because at the time there was a mismatch between his ideas and the availability of relevant data. These days, we finally have the technologies—like optogenetics—which allow us to observe the activity of very large populations of neurons in real time, which may deliver the data needed in order to see whether and, if so, how, population-level processes like selection are actually happening in the brain. No, there has to be weakening as well as strengthening, for all sorts of reasons. Very simply, your skull would run out of room if things only ever got strengthened. At a computational level, learning is just as much about pruning connections as it is about strengthening connections. One initially surprising fact about brain development is that synaptic density is at its peak at between 2 to 3 years old. From then on, we’re all losing connections. But this is a good thing because learning requires trimming away the stuff that’s not necessary. The principles of statistics and machine learning have taught us that pruning is necessary in order to enhance generalization—to avoid overfitting to the data on which algorithms are trained. There’s also a metabolic cost to having too many neurons and connections. There are all sorts of reasons why we need to select and finesse rather than merely reinforce. There’s one other book that I wanted to mention in connection with these ideas. This is Christof Koch’s recent The Feeling of Life Itself . Christof, after working very closely with Francis Crick until his death in 2004, then began collaborating with Giulio Tononi. In the last 15 years he has now become, along with Tononi himself, one of the main proponents of integrated information theory. In a way this seems to be a switch to the other side, but, alternatively, Koch’s trajectory can be thought of as an extension of the neural correlate approach, but now through a different theoretical lens. His book, The Feeling of Life Itself, is a very accessible, authoritative and up-to-date manifesto for IIT."