20121120_SE
Source: Nesta
URL: http://www.nesta.org.uk/assets/events/nature_and_necessity_of_a_scientific_revolution
Date: 20/11/2012
Event: In Conversation with Stephen Emmott
Credit: Nesta, also to Geoff Chambers for transcribing this
People:
- Stephen Emmott: Head of Computational Science, Microsoft
- Geoff Mulger: Chief Executive, Nesta
Geoff Mulger: Good evening and welcome to Nesta on a cold November Tuesday. I hope you managed to get something to eat and drink. My name is Geoff Mulger, I'm chief executive here at Nesta, I'm just going to do a couple of minutes of introductions before handing over to Stephen and Jessica. So Stephen Emmott who is our main attraction this evening - many of you will know his work - he's had quite a bit of involvement here at Nesta as a trustee and distinguished fellow. He is head of the computational science laboratory at Cambridge. I've had the benefit of seeing his office where he works, but also very unusually, he's a scientist who has had his office recreated on the stage. I just wonder if we can have a show of hands, how many people in this room saw Stephen's play at the Royal Court theatre last summer? Nobody.
Stephen Emmott: Thank God.
Geoff Mulger: It was sold out. So you probably all tried to get tickets but failed. A very fascinating play really about the future of the human race and rather modest topics of that kind. Stephen as you'll hear works at the boundary of computing data, natural science, and i think potentially in the future social science as well, the understanding of human systems. I wanted to give one plug before I lose my voice for a Nesta report coming out next week which may be of interest to some of you on a neighbouring topic to Stephen's which is looking at the use of data in the economy in the society and it's a publication on the rise of what we call the "datavores" - those organisations which are the most avid users of data. And the implications of a society and an economy much more shaped by often realtime feedback using large scale data.
What that report raises is I think two sets of issues which I think Stephen's going to cover, one is the question of whether when you have availability of far more data, does this mean you no longer need theories, hypotheses, ideas, because all you need to do is look at the data, and they will jump out, the ways of understanding the world? Very different views on whether that is correct. And also the question of how the availability of data and modelling of data actually forces attention to clarity of thinking about causation. And some of what Stephen is doing is raising all sorts of questions about natural sciences, and just how solidly grounded are their underlying theories of causation, but of course the same applies to everything from business models to social science to understanding human psychology.
Anyway, that's enough from me. We're going to have a presentation of fifteen minutes or so from Stephen, and then there'll be a conversation with Jessica Bland who is now at Nesta which will explore some of the topics and then there'll be hopefully plenty of time for you to ask incredibly probing, difficult questions and to lead us towards enlightenment. So Stephen, over to you.
Stephen Emmott: Thanks Geoff. Despite what Geoff says I'm not entirely certain why I've been invited to give this talk and I'm fairly certain that by the end of it there'll be at least half of you asking yourselves the same question. So but I, and indeed, some of the issues that Geoff just mentioned are probably ones that would be best covered in the conversation with Jessica about data models and suchlike.
I have as Geoff said about 15 minutes to give a talk and there's quite a lot to get through, so I will just - my apologies in advance, sort of errors and emissions excepted.
Its going to be a very high level, very quick fly-through, quite a lot of issues to sort of tie together quite a lot of issues about why I think we need transformation of science and I'll just finish off very quickly with a couple of examples from my lab of our attempt to overcome some important barriers in science, to address some important scientific questions.
So oh, there you are Geoff. And there's me. No content yet? And in fact it might even go downhill from here.
So, another title. That's my own. So, "Need and nature of a new scientific revolution or transformation", it's what we're talking about, and I want to start off by just a very quick summary of what I think are the most important questions of our times, and certainly in my view the important questions for us as a society this century.
So the first is, you know, how is the climate going to change and more important, what's going to be the impact, what's going to be the consequences?
How we are going to feed a global population of ten, at least ten billion - it might be a lot more? How are we going to - sustainably - power a planet of ten billion or more? We could easily power the planet if we just continue to use oil, coal and gas but that would almost certainly finish us off.
Have we embarked on the sixth mass extinction of life on earth? the answer is almost certainly yes. In geological times that's certainly true, almost certainly the case. If you look at what's happening with the loss of amphibians, birds, mammals, non-vertebrates and indeed related to that question is the bigger question of what's the future of life on earth particularly for global ecosystems and ourselves, given all the other questions.
Some other interesting questions: Can we predict, prevent, or manage a global pandemic? Most epidemiologists are of the view that a global pandemic is a matter of when, not if. It could be tomorrow. It could be in 83 years time. The last one was the Spanish flu pandemic which was estimated to have killed 50 million people and is now estimated to have killed at least a hundred million people, and most epidemiologists now agree, the rate we travel around the planet, that the next one could kill up to a billion people. So It's a big deal.
How does the immune system work? That's also an important question, because right now, the only thing between us all in this room and the cemetery is our immune systems, because we're all breathing in all sorts of pathogens.
How do stem cells build us? And a topic of great interest at the moment. What is it the brain actually does? an interesting question not least because we have have apparently had 2 decades of spectacular success in neuroscience.
And I suppose finally, related to the sort of last batch of medical and purely biological questions is: are we going to be able to cure, prevent diseases such as Alzheimer's, motor neurone diseases or cancer.
The interesting thing about all these questions is that they're all related to science. Science is at the centre of them all and actually its biological sciences not the physical sciences that's at the centre of I would argue all of them, and we're unable to answer any of these questions, which is quite remarkable in itself.
In fact in terms of biology, we don't even knows how a cell works. In 2012.
So, just to quickly underline what I mean by that, in the area of climate, after 50 years of climate modelling in places like the Hadley Centre, uncertainty in climate models is still a critical issue, the key issue in climate modelling.
In ecology, despite 200 years of data collection in ecology we still understand very little about how species interact, about ecosystem structure and function and about extinction rates, so we're unable to ask the question, unable to fully answer the question about whether we've embarked on the sixth mass extinction of life on earth and what the future of life on earth is.
In biology and medicine as I mentioned earlier, despite 50 years of spectacular success, which are really about advances in techniques, in molecular biology and neuroscience, genetics, we literally don't know how, don't understand how a cell works, even though we can describe most of the parts.
I just want to underline the last part. I could have done, I could have ridiculed ecology just as much as biology but I wanted as a biologist I feel as though I'm at least somewhat qualified to ridicule my own discipline. But this is how I would summarise biological science in 2012.
And it's probably best way to say: How would a biologist go about fixing a radio? First of all biologists would spend decades of taxpayers' money buying tens of thousands of radios to experiment on, and then having spent ten years in figuring out how to get the back off of radios to see what's inside, they would spend another ten years on classifying all the parts inside according to their size, their shape, their colour, their position, that's basically the transition from anatomy to experimental biology. When experimental biology was fully flourishing let's call it, or as I like to call it "poke and measure", all sorts of poke and measure went on for decades until one sort of lucky post-doc in someone's lab figured out that if you take this particular part of this radio out, it stops working, and discovered that, and claimed that they'd discovered the music gene, or at least some component implicated in the music gene - I should say that terms like "implicated in", "involved with" are terms that one, is littered throughout the biological scientific literature, even now, which is quite, quite astonishing...
And then along came imaging about ten years ago, and in particular functional imaging, and so biologists would have put the radio in a magnetic resonance imager to look at the functional areas on the circuit board measured by heat - functional magnetic resonance imaging measures blood flow in the brain, which is a remarkable measure of information processing, or lack of it - but if the analogy here would be heat, and a biologist would have concluded that the power supply is where most information processing occurs because that's where most of the heat is, and let's not forget that the pictures look lovely in Nature.
And then also in the last decade along came the "-omics" genomics metabolomics, proteomics, so what a biologist would have done, having discovered that the "-omics" era is among, is upon us is put a thousand radios, put them all in a carrier bag, crush them all, and take the statistics of the final state, and reveal that a radio is an interaction network - this is state of the art in biology at the moment - involving silicon copper and plastic, and unquestionably a Nobel Prize awaits for the team that, the team that did that, and indeed, they got it.
So, the question is, given that that's the state of biology, and I could have done something similar for ecology, the question is: Now what? Given the fact that we don't even know how a cell works, and we don't know how ecosystems function, and we don't know whether we can feed a population of ten billion, and we don't know how we're going to be able to power the planet, we have no way of predicting or preventing a global pandemic, what now in biological science, the natural sciences more generally?
Well I would argue that we need a fundamental transformation of both how science is done, and what science is done. And in particular, you know transformations in science don't happen, fundamental transformations in science don't happen, very often. Arguably the last one occurred at the end of the sixteenth, at the end of the seventeenth century, and, er, there have been very little, very few revolutions in science. But they occur, fundamental revolutions in science occur when prevailing views of the day tend to be overturned by fundamentally new ideas, by fundamentally new ways of thinking, and by formal, what we want to call languages for thinking about those ideas and representing them, so the discipline of physics came about when, was enabled I think it's a stretch to say it came about but the formalisation of the discipline of physics came about or was enabled because of a formal language called calculus. So we need that kind of new kinds of conceptual methods in science, in particular in biological science, ‘cos all we've got at the moment in terms of formal methods are the methods from physics, so we're looking at things like rates of change, and biological systems are not defined by rates of change, they're defined by processes. And so we need fundamentally new - not only fundamentally new ways of thinking and new ideas, but new ways of formalising those thinking in terms of new conceptual and computational languages, and I'll come to it right at the end, but we also need new kinds of scientists to actually pioneer that transformation in science.
What I want to do now is just give you two simple examples from my lab about how we're thinking about, how we might be able to bring about such transformation or think about bringing together new ideas, new kinds of scientists and new kinds of formal languages to think very differently about how biological systems work.
So the first example is the brain...
[There follows an illustrated discussion of brain research.]
*******
Example 2, and I'm nearly done, which is pretty bold, which is the future of life on Earth. As I mentioned earlier, we have no idea about the rate at which we're losing, no good idea about the rate at which we're losing ecosystems. We have no idea about ecosystem structure and function, and about how ecosystems provide for us what are known as ecosystem services. So, food and water are two important ecosystem services, and that's an important question because we are losing species. We're not clear about how much, it could be between, it could be a rate of extinction a hundred times greater than we would expect from normal evolutionary processes, or it could be a rate of over a thousand times that we would expect from normal evolutionary processes, but we're losing them at an alarming rate.
Given the fact that we're actually unable to characterise or understand how many species there are on earth, most scientists agree that the simple fact of that means that we're losing species at a far faster rate than we currently think we are, so it's almost certainly at least a thousand times - the rate of species extinction on earth is at least almost a thousand times greater than it should be through normal evolutionary processes, and that's a big deal, because we depend upon ecosystem structure and function for our own well-being on this planet.
So, the other interesting thing is that there are something like thirty classes of global climate models from the Hadley Centre, from the Lawrence Livermore Laboratory, from Potsdam, elsewhere, and remarkably, there isn't a single global ecosystem model, and the reason is is because everyone in the ecology community says its too difficult to do and so we tried to do it. And the reason why the ecology community understandably say its too difficult to do is there are a number of problems in modelling the global ecosystem. First is we don't know how ecosystems work. Second problem is, as I mentioned earlier, we don't know how many species there actually are, which is from a recent paper with people in my lab and at Dalhousie University in Canada. Third problem is it's infeasible to model every individual, which people think you need to do if you're going to build a global model of ecosystems, so in terms of autotrophs, so everything at the bottom of the food chain that relies on sunlight for survival, you know there are billions of them per litre, all the way for example in birds, you know there are thousands of them per kilometre, you can't, you know it is infeasible to model every individual across the planet
so solution one is not to model every individual. Solution one we think or this is what we've started out with is to model functional groups into these sorts of things herbivores carnivores omnivores ecotherms, things that are sessile, don't move, things that are mobile, like birds, things that just have a big bang of reproduction versus things that reproduce every year and then it's autotrophs, for example, this is just an example, model cohorts, because many of these things can be categorised as basically belonging to the same group; model processes - an interesting thing is attempts to model ecosystems is that they tend to, previous models have tended, not entirely, but have tended to model phenomenon, observable phenomenon about where species are and what you can see them doing, but that's actually not the right way to go about it. What you need to do is actually model biological processes and here are some of them like where they move to metabolism predation mortality and one thing and another, and then solution 4 is to, if you're going to bring all of that together, is you need to effectively build what you might want to crudely call an LHC for the planet rather than LHC to look for Higgs' Boson.
So this is a new set of computational methods which instead of colliding particles, collide models and data, thousands of models with petabytes of data using this sampling method for parameterising models and selecting the most, the most, let's call it the best model, for want of a better term, so code name for this method is called Philsbac it's based on Monte Carlo microchain sampling, a particular robust version of it called Metropolis-Hastings, and then from that you get an ideal model structure, you get parameter distribution, and you get these predictions, so what we've built here is a prediction machine for complex natural systems but one which enables models to be built that are fully data-constrained, which has simply not been possible before.
So what we get, sorry about the quality of this, this is completely in the wrong colours, but what we get if we build this global ecosystem model, which we have done, is we can basically model the last few million of years of biomass distribution of biodiversity distribution, and remarkably, if we model that, even though you know we've just got these functional types in there like autotrophs and herbivores and carnivores, if we model the previous few million years, we get a remarkably faithful model of where ecosystems are now and how we think that they're distributed.
Our next step is to model what we think is going to happen to ecosystems, so we think we'e got a fairly good idea for a first step of having a reliable model of ecosystem structure and function and a global ecosystem model, our next step - and I'll show you just one or two a couple of examples - is to model where - so that's postdiction, for a few million years, we're going to do prediction for the next probably 200 years, and this sort of model is enabling us to ask fundamentally new questions about what ecosystems do, ecosystems behaviour, our impact on them, and the impact of our impact on these ecosystems on our wellbeing.
The interesting thing about this model is that - there was an important report came out in 2011 which was this Living Planet report. What it showed was, if you looked at ecosystem health - I won't bother going into what that meant in this report - in the 1970s the report showed that there's been a significant reduction, from one, as in the state of global ecosystems in 1970, up to 2007 when they stopped collecting data for this report, a degradation of about 30%, like 0.7 of, so a significant decline. What this model shows is that actually the problem is far worse than the policy community, policy-making conservation community and indeed the climate community think - ‘cos the climate is related to ecosystems structure and function - is far worse than we think, so the dotted line is the output from our model.
So that's two very quick examples of what I mean by an attempt to kind of have a transformation of the way science is done and what science is done by asking totally new questions about what the brain does, what cells do, thinking about cells as computation and living software about thinking about can we actually do things which everyone thinks is impossible, like build a global ecosystem model.
The important thing about this I think that I really want to underline is the last slide is that none of this is going to be possible without an entirely new generation of entirely new kinds of scientists, of scientists that have a very different way of thinking about biology and natural science, scientists who are scientifically first rate, not just in one discipline, but are genuinely interdisciplinary rather than working in interdisciplinary projects, which tends to just mean that people still work in silos, and it requires people who are computationally first rate, and I don't mean people who know where the on button is on their Mackintosh, I mean conceptually and mathematically computationally first rate. And those kinds of scientists are only just emerging, you know arguably the first generation of scientifically first rate, computationally first rate natural scientists are only just beginning to emerge.
And what's interesting is, no-one would think about being able to become a physicist if you weren't mathematically first rate, and its going to be impossible to do ecology or biology over the next decade unless you're computationally first rate, and I don't think the research councils or the general university system has grasped this problem yet, so the idea of new kinds of scientists is one that I really want to hammer home fundamentally as critical to this transformation, an urgently needed transformation of science, if we're going to solve some of these massive important and unprecedented global challenges that we face, of which science is at the centre of all of them. And that's it. Thank you very much.