Imagine, for a moment, the intricacies and complexity of the human brain. More powerful than any computer that has ever been invented and with the capacity to hold somewhere in the region of 2.5 petabytes of information (enough, says Paul Reber, professor of psychology at Northwestern University, to store three million hours of TV shows), it has perplexed and fascinated scientists in equal measure.
For years, people have been studying what makes us tick and how the brain works, but such studies are about to get even more serious. Millions of pounds have been pumped into the most advanced neuroscience project in the world, a series of studies that are vast in ambition and scope. For the next decade, the Human Brain Project aims to better understand the operation of the grey matter that lurks between the lugholes of every one of us. And if things go according to plan, it will ultimately lead to fresh knowledge in areas of vast importance, including mental health.
One of the key figures in all of this is Steve Furber, ICL professor of computer engineering from the School of Computer Science at the University of Manchester. As the brains behind the inner workings of the BBC Micro and the genius co-inventor of the ARM chip, he is using his vast knowledge to spearhead the SpiNNaker Project, a novel computer architecture that is inspired by the working of the human brain. It forms part of the Advanced Processor Technologies Research Group at the University and the Human Brain Project. Prof Furber hopes it will spark some major breakthroughs.
At its heart are a million bespoke ARM processors which give the SpiNNaker team considerable power. “The system is ultimately capable of modelling up to a scale of one per cent of the human brain,” Prof Furber reveals. But although the Human Brain Project is hoping to eventually recreate an entire brain, such an achievement is still some way off.
“For a full-scale computer model of the brain, we’d be looking at exascale [1018, or a million million million operations per second] and the industry does not know how to build exascale computers with economic energy consumption,” he explains. “Even if we did crack it, we’d be looking at a computer that consumes tens of megawatts. You won’t be building that into your walking, talking Android robot any time in the foreseeable future – it would need a nuclear power station in its head just to keep the computer going.”
Recreating one per cent is therefore enough for now, even though the Human Brain Project wants the whole mouse brain to be reconstructed and simulated by 2020 and for an entire human brain four years later. A mouse brain is 0.1 per cent of a human brain so SpiNNaker will be able to model ten whole mice brains. Recreating one per cent of a human brain is certainly more than capable of opening up the possibilities for serious study: the SpiNNaker project works by mimicking the brain’s processes and it will give scientists a good indication of how it will work.
In the brain, axons carry electrochemical signals, or spikes, between neurons – the components which hold information – allowing them to communicate with each other. The neurons will be replaced by specially designed processors. Packets of information will be exchanged between these instead. “We’re modelling at a reasonably simplified level so our neural models are not highly detailed biological models,” admits Prof Furber. “They’re simplified models that have the expected dynamics of biological neurons. In general we abstract away most of the complex structure of the biological neurone and treat it as acting as a single point. The high-performance computing folk will build highly detailed biological models which we hope we will be able to extract data from to support these abstractions.”
Prof Furber’s contribution is one of a number of projects around Europe. His team is among 135 research institutions in 26 countries co-ordinated through the École polytechnique fédérale de Lausanne (EPFL). There are computer scientists, roboticists and doctors and neuroscientists looking into high-performance computing, medical information, neuromorphic computing, neuroinformatics, brain simulation and neurorobotics. The neurorobotics platform will exploit the 3D modelling capabilities provided by commercial and open source gaming platforms. Each of these areas will be tested by scientists within the Human Brain Project and by researchers across the world from 2016.
One of the fundamental aspects of the Human Brain Project, which is being collated and pulled together in Geneva, Switzerland, is the need to open it up. The research platforms, of which the work in Manchester is one, will provide services to researchers, clinicians and technology developers outside of the project. Neuroscience data generated by the project will also be deposited in what are termed ‘brain atlases’ so that they can be freely available to the scientific community. “Software for neuroinformatics should be released in open source format with a licence allowing free use for academic research,” according to a report handed to the European Commission about the project.
“The Human Brain Project is so big that almost everything is used somewhere – the major platforms will have operating system-agnostic accessibility,” says Prof Furber, who first put the plan in place for SpiNNaker in 2005 and was granted money from the Engineering and Physical Sciences Research Council a year later. Work has been able to progress over the past couple of years, since the chips were delivered – much of the early graft was put into designing them with their 18 individual processors.
In trying to better understand the human brain and its disorders, this IT-based research approach will look at the structure and function of the brain; the causes, diagnosis and treatment of brain diseases; and, as with SpiNNaker, the development of new computing technologies such as, for example, low-energy brain-like computing systems. To aid things, a ten-year plan has been put together that says the IT platforms will be ready within 30 months. It is hoped that computer- generated hypotheses about neural behaviour will be good enough for scientists to be able to make higher-level hypotheses about the emergent structures of the brain.
“The ultimate goal is to gain some understanding of the fundamental principles of the operation of the brain, which at the moment, in terms of information processing systems, is still a complete mystery,” explains Prof Furber. “A great deal is known about the technology from which it’s built and we can use MRI scans to see activity moving around, but all of the interesting stuff is at intermediate levels where we have no instrument to see what is going on and the only tools we can bring to bear are computer models.”
It hasn’t been without its problems. Large numbers of computer processors are needed to run in parallel and there needs to be an efficient way of sending signals between them. The neurons themselves were easy to model, but the connections were not. Prof Furber says the answer was to represent the ‘pinging’ of a neuron as a small packet in an electronic communication network and then move it quickly around the system to many different places.
“Our ambition, if you like, with SpiNNaker is to remove the computational constraints which have limited the scales of computer models that neuroscientists have been comfortable building so far,” Prof Furber tells us. “They have built big models that take months to run, but they are still constrained by computer resources and we effectively want to remove that constraint so they can model things on an appropriate scale and hopefully gain new insights that way.”
SpiNNaker will have two main uses. It will look at the development of treatments for mental disorders. Neuroscientists and psychologists will be able to use the project to understand certain conditions, including Alzheimer’s, dyslexia and depression. Prof Furber also wants computer scientists to use the SpiNNaker machine to work out the best way of producing electronics more tolerant of system faults.
It will also have applications in robotics. Prof Jörg Conradt at Technische Universität München has already taken 48 of the APT group’s chips and placed them on the back of a robot. It can recognise two shapes via its silicon retina. Meanwhile, Professor Henry Markram, director of the Human Brain Project, says we could one day be using neuromorphic computers that learn like the brain. He say computer chips could be produced with specialised cognitive skills mimicking the brain and allowing for decision making and crowd analysis.
Prof Markram has long had an interest in simulating the human brain on computers. He set up the Blue Brain Project in 2005 and its research involved studying slices of living brain tissue using microscopes and path clamp electrodes. The data about the different neuron types are used to build biologically realistic models of neurons and networks of neurons in the cerebral cortex. Underpinning this is a software package called Neuron, which was developed from the 1990s onwards by Michael Hines at Yale University and John Moore at Duke University. It is written in C, C++ and Fortran. Now at version 7.2, it is free and open source software with the code and the binaries freely available online.
“For us there is the biological motivation that understanding the brain is an interesting scientific challenge in its own right and, if we can build that understanding and build computer models, we should then be able to offer a lot of assistance for people trying to develop important drugs for mental diseases,” states Prof Furber. “But also, from a personal perspective, we’re interested in the grand AI challenge. There has been a lack of progress in this area because we have not really understood what natural intelligence is. We struggle to emulate it in machines because we still don’t understand what we’re emulating, so I think the right approach is to go back to the biology and try and understand that better and apply it.”
Using technology to better understand the brain is of vital importance, Prof Furber adds, because the big pharmaceutical companies have “pretty much stopped investing in drugs for mental diseases”. He says the costs grew too high because there is a lack of basic understanding about the human brain. “In most cases big pharma now designs drugs,” he notes. “They understand a problem and they design a drug to address it directly and then they have to do a fairly constrained search to find the right biochemistry to treat the problem. With brains, because they don’t understand the problem, effectively the search for drugs is entirely empirical so they try everything and see what happens. That’s just become far too expensive so they’ve pulled out. Mental disease is still economically a bigger problem than heart disease and cancer put together and yet it is largely ignored because we don’t have the tools to do anything about it.”
Prof Furber knows there is a lot of money riding on the entire project. The total costs are some €1.19 billion and, of that, €555 million is earmarked for personnel. Sources suggest that the entire project will take 7,148 person- years of effort. Being open will make it worth the money, though. Usually it is difficult to collect neurological research data because of the unsystematic nature of the information collected from previous brain research. Here the data is being collated and utilised via open source software. A database of brain research from the tens of thousands of neuroscience papers published each year will also be built.
By the end of it, there will be better understanding of the structure and function of the human brain in terms of gene expression, cell numbers and morphology, long-range connectivity, and cognitive function.
“At some point, and I hope it will be in this project but who knows, there will be a major breakthrough,” Prof Furber insists. “We’ll say, ‘ah yes, this is how the brain represents information and this is how the different regions interact and cohere, operating independently some of the time and operating in synchrony at other times’. But at the moment we don’t even have that fundamental key insight into the framework around which to wrap all of the details. As I’ve said, we’d need computers the size of aircraft hangars to mimic an entire human brain, but there is hope that that understanding will lead to important applicable systems, such as vision systems that make driverless cars a practical reality, and also lead to some major mental health advances.”
To find out more about the Human Brain Project, visit the official website.