Thursday, 8 March 2018
Life, the universe and Markov blankets
Forty years ago today, on 8th March 1978, the Hitchhiker's Guide to the Galaxy was broadcast on BBC Radio for the very first time. I personally first heard it 12 years later, but since that time it has remained my favourite comedy ever. To celebrate the birthday of this remarkable show, book, film, computer game here is a section from my book. This is from a chapter on boredom, surprise and why babies are little scientists. It's also about why scientists are big kids.
Share and enjoy!
Life is an inevitable & emergent property of any (ergodic) random dynamical system that possesses a Markov blanket. Don't leave without it!
@FarlKriston, 12 Jan 2015, Anonymous Twitter parody of Prof Karl Friston, FRS
Professor Karl Friston is probably the most influential scientist you’ve never heard of. He works at the University College London’s world famous Functional Imaging Laboratory, affectionally known as the FIL. He has been there most of his career. Back in 1991 he invented Statistical Parametric Mapping. SPM is a statistical technique for analysing the data from brain imaging experiments. SPM is also a set of software that will do the analysis for you. The elegance of the method and the fact that Friston gave away his software for free has led to SPM being used in about 90% of all brain imaging studies. As a result, Karl Friston is the most widely cited neuroscientist alive.
Most of us would have been very happy with that but Friston’s contributions did not stop there. He worked with Chris Frith to develop a highly influential account of schizophrenia and invented something called Dynamic Causal Modelling. He is author or co-author of over a thousand scientific papers. This is a mind-boggling number. For comparison I have under 20. This is partly a consequence of his eminence but also evidence that he is a very practical researcher who stays involved in the nitty-gritty. Friends who work with him tell me he is an affable and generous colleague. Friston’s many, many awards include Fellowship of the Royal Society and something called Golden Brain Award. He is undoubtedly a worthy recipient. He is famous among brain scientists as perhaps the brainiest of them all.
Karl Friston’s most recent idea is his biggest yet. The ‘Free-Energy Principle’ tries to explain not only what brains do but possibly even life itself. But it is also making Friston infamous, as the nerdy but well-meaning mockery of @FarlKriston account shows. The trouble is that the free-energy principle is very hard to comprehend and Friston’s explanations and equations usually only make matter worse. I will do my best to translate. But if you found the last section bamboozling now might be a good time to reach for your security blanket.
Here’s a relatively tame example of Karl explaining it in his own words:
The free-energy principle says that any self-organizing system that is at equilibrium with its environment must minimize its free energy. The principle is essentially a mathematical formulation of how adaptive systems (that is, biological agents, like animals or brains) resist a natural tendency to disorder. (p.127, Friston, 2010)
This makes it sound important doesn’t it? Apparently it can explain everything from the existence of ‘Life as we know it’ (Friston, 2013) right up to Freudian theory and psychedelic drug experiences (Carhart-Harris & Friston, 2010).
The secret is nestled in our Markov blankets. As Karl helpfully explains “The term Markov blanket was introduced in the context of Bayesian networks or graphs and refers to the children of a set (the set of states that are influenced), its parents (the set of states that influence it) and the parents of its children” (Friston, 2013). Got that? In essence, Markov blankets are a supercharged version of Judea Pearl’s Bayesian networks. They provide a statistical way to represent the boundary between an organism and the world. The mathematics gets very complicated, combining Bayesian statistics, information theory and entropy to explain how life can survive in the face of the chaos of the universe. But, in some sense, the free energy principle states that life is about trying to avoid being too surprised by the future.
Knowing what might happen next sounds like a good survival strategy. For Karl Friston, life is anything that can predict its own future. From single cells to Sigmund Freud, he wraps us each in a Markov blanket and sends us out to do battle with the unknown. Describing organisms in this way has some useful features. Action, perception and learning all become mathematically well-defined properties of the system. Perception provides information to optimise future predictions, actions move us out of uncertain (dangerous) situations, and learning is about updating internal states and beliefs about the external world. This might seem a very abstract way of looking at things but it supporters see it as general framework that can be applied as easily to bees as to babies.
Critics of Friston’s theory say there is nothing easy about it. They view it as an interesting intellectual exercise but say that the Free Energy Principle and the closely related Bayesian brain hypothesis are too general to be useful in the real world. FEP is a lot more abstract that SPM. It is not easy to see how it can be used to predict how adult or baby brains will react to the world. But this style of reasoning is already being used to understand what happens in real brains.
My favourite experiment of this kind involved a group of ferrets who went to the cinema. In what was clearly a naked attempt to win an Ignobel Prize, Josef Fiser and Michael Weliky at the University of Rochester got ferrets to watch The Matrix on DVD. The choice was deliberate because like Neo and friends the ferrets had wires coming out the back of their heads. This allowed the scientists to watch what they were thinking. They would watch the film all day and then dream about it at night. In doing so they helped Fiser and colleagues worked out a lot more about how brains are Bayesian predictors (Fiser, Chiu, & Weliky, 2004).
Josef Fisher is the sort of scientist Hollywood might dream up. He is tall, charming, handsome and impeccably dressed. He is Hungarian but speaks English with an American accent. His work is in psychology and neuroscience but with a strong mathematical element. If they made a film about him there would be a whiteboard of equations in the background.
Using the Matrix for this research seems like a doubly prophetic choice. In grown-up ferrets, the scientists discovered patterns of neural activity which correlated significantly with the images on the screen, whilst baby ferrets were more confused. Moreover, the adults kept on thinking about the movie after it had finished. This was pretty cool but the most surprising thing was that ferrets brains were not just passively reproducing what they saw. They could dream about it and their dreams appeared to actively improve their models of the world. Like Neo in the film, they were bending their existing expectations to fit their new reality. Or in the language of Karl Friston, they were minimising their prediction errors via a Gibbs sampling over the probability space.
Still with me? Perhaps now might be a good time to mention Douglas Adams. In that wholly remarkable book the Hitchhikers Guide to the Galaxy, he relates that one the highest compliments intergalactic hitchhikers can pay to one another is to say that they are “someone who knows where their towel is.” There are, of course, many practical uses for a towel when travelling the universe but a towel’s greatest value is psychological. As the book recounts:
“any man who can hitch the length and breadth of the galaxy, rough it, slum it, struggle against terrible odds, win through, and still knows where his towel is, is clearly a man to be reckoned with” (Adams, 1979).
Babies and young children are more likely to carry a security blanket or soft toy on their adventures. Some estimates suggest that up to 70% of children have a strong attachment to a particular object. This seems to be largely a Western phenomenon, possibly a result of many more children sleeping separately from their parents than in Eastern cultures. The standard explanation was that these ‘attachment objects’ are a substitute for the original ‘object’, the mother and her breast. This does not hold up. The objects can provide security in novel situations but it seems to be an independent a child’s attachment to their mother (Donate-Bartfield & Passman, 2004).
Donald Winnicott thought attachment objects were reminder of security and love. I think this is correct. However, Karl Friston and the ferrets lets us see security blankets and velveteen rabbits in a wider context. Babies and children use these objects because we need security as we build our worlds. The ultimate aim of life is to explore enough of the world so that you can survive surprises. We will never be able to expect the unexpected but we can and we must reduce its scope and its impact. To survive we have to change our minds many times. Like the ferrets, we do this by improving in a Bayesian way, updating our beliefs to better fit our experience.
Babies are surprised every day and must continually confront uncertainty and explore the unknown. This is exhilarating and exhausting. It is not enough to add knowledge about the world, they must change their expectations, a new existential crisis every day. Mummy, teddy or a security blanket is a reassuring element of continuity and predictability. If babies know where that is, they know where they are. Perhaps this prevents the Markov blanket from unravelling?
It is hard to appreciate what a wild ride this must be for them. Adults do not change our beliefs very often. We have worked our whole lives to feel like we are right about most things. After all, in Friston’s theory that is the whole point of life; being less surprised over time. The best analogy I can suggest is to imagine your home planet gets destroyed and your best friend turns out be an alien who takes you hitchhiking across the remainder of the galaxy.
I might be biased here. Not only I am keen to meet aliens, I also still have a security blanket. I had one when I was little. I was so attached to it that my mother knitted a large elephant and sewed the tatty cotton rag onto his back. This was meant to deter me but inevitably I ended up dragging the elephant around everywhere. The original blanket disintegrated decades ago. It was replaced in my affections by a cellular blanket my mother bought home one day from hospital. The blanket wrapped a parcel containing my baby sister. No doubt I seemed alien and friendly to her. I still find cotton cellular blankets very soothing. May you do too? Around 30 percent of adults keep an old teddy or similar childhood memento. By my estimates, Charlie Brown’s friend Linus van Pelt must be into his seventies by now. I imagine he still carries his blanket some of the time.
As luck would have it just as I was finishing writing this section, Karl Friston came to my university to give a lecture. This was my chance to hear him in his own words. Would I be bamboozled? His opening was not promising, “I can give very good lectures. This is not one of them.” But he was wrong. He had us imagining ourselves as hungry owls and playing a logical guessing games. There were some of his infamous equations but he guided us through gently them. His reassuring, unhurried delivery no doubt acquired during his early training as a psychiatrist.
His summary of the purpose of life fits very well with the ambitions of our babies. When confronted with the big mysteries of the universe we are compelled to explain them (even if only to explain them away.) “The brain is in the game of explaining the sensory impression at hand.” This is a game of chance and we are all born with an intrinsic motivation to play it. We all keep score with the information gain measured in “Bayesian surprise” (or minimized free energy). Afterwards, I told him about this book and asked if he any insights into how this must feel for babies who play the game so much more intensely than adults? He could not much improve upon the hitchhiking analogy. And, wise man that he is, he also turned the question around. “This is why we are scientists isn’t it? To keep exploring and to try and recapture that feeling of joy?”
At the end of his famous 4 page paper Judea Pearl stated that he hoped his Bayesian networks would become “a standard point of departure for more sophisticated models of belief maintenance and inexact reasoning”. Be careful what you wish for. I do not think he could have predicted that ferrets would be watching The Matrix or that Karl Friston would drape Markov blankets over all life on Earth. “In truth it's blankets all the way down. So cuddle close and keep that free energy minimal.” (@FarlKriston, 8 Dec 2017)