The next “space race” might be the race to develop a synthetic model of the human brain – one that Google and Microsoft will participate in, if a report is true.
And instead of trying to beat the Russians, this time the Americans will be racing against the Europeans, who have already announced their plans.
The New York Timesreported Monday that the Obama Administration is close to announcing the Brain Activity Map, which scientists quoted by the paper say could be on the scale of the The Human Genome Project, a $3.8 billion project to map the human genome that, the Times reported, returned $800 billion in jobs and other benefits.
The Brain Activity Map would attempt to document how the brain works, from the tiniest neurons up through how possibly the different regions of the brain communicate with one another. If the project succeeds, the Brain Activity Map might give us an understanding of how the human brain “computes” data through its complex web of neurons. It might also help scientists solve brain-related diseases like Alzheimer’s.
Modelling the human brain, and figuring out how it works, has long been one of the Holy Grails of supercomputing, prompting fears of a “technological singularity,” where successively advanced artificial intelligences design ever more refined versions of themselves, leading to a future where humans become increasingly irrelevant.
On a more realistic scale, learning how people think could allow services to begin anticipating their needs, a problem companies like Google and Microsoft would be interested in solving. The Times reported that a Jan. 17 meeting at CalTech was attended by the National Institutes of Health (NIH), the Defense Advanced Research Projects Agency and National Science Foundation, plus Google, Microsoft, and Qualcomm.
Google representatives did not return an emailed request for comment, possibly because of the U.S. President’s Day holiday. A Microsoft Research representative said that the company declined to comment.
Two of the foundations of the Times report were public statements: a tweet by NIH director Francis S. Collins, and a mention of the efforts to map the brain by President Obama in his State of the Union address:
“Every dollar we invested to map the human genome returned $140 to our economy,” Obama said, according to a transcript of the speech. “Every dollar. Today, our scientists are mapping the human brain to unlock the answers to Alzheimer’s. We’re developing drugs to regenerate damaged organs, devising new materials to make batteries 10 times more powerful. Now is not the time to gut these job-creating investments in science and innovation. Now is the time to reach a level of research and development not seen since the height of the space race. We need to make those investments.”
Collins then tweeted: “Obama mentions the #NIH Brain Activity Map in #SOTU”.
The Other Horses in the Race: the EU
Funding for the U.S. effort could last as long as 10 years, and possibly top $3 billion over that time. But the bar was set earlier by a massive collaboration among more than 80 European research agencies, which won an award from the EU of one billion euros ($1.34 billion) to develop a computer simulation of the human brain, known as The Human Brain Project.
That will partly cover the intriguingly named “Neuropolis,” a building dedicated to ”in silico life science” that will serve, at least in part, as the computer infrastructure behind the effort. The Swiss Confederation, the Rolex Group, and various third-party sponsors are backing this part of the effort.
“The HBP will build new platforms for “neuromorphic computing” and “neurorobotics,” enabling researchers to develop “new computing systems and robots based on the architecture and circuitry of the brain,” according to the The Human Brain Project.
Other Horses: IBM’s/DARPA SYNAPSE
The Defense Advanced Research Projects Agency, responsible for the initial funding and challenges to design self-driving cars and other public-private partnerships, has worked with IBM to develop SYNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics), whose ultimate goal is to build a “build a cognitive computing architecture with 1010 and 106 synapses” – not a biologically realistic simulation of the human brain, but one where computation (“neurons”), memory (“synapses”), and communication (“axons,” “dendrites”) are mathematically abstracted away from biological detail.
Using 96 Blue Gene/Q racks at the Lawrence Livermore National Laboratory, the most powerful supercomputer in the world, the team achieved 2.084 billion neurosynaptic cores containing 5310 neurons and 1.37×1014 synapses, according to the blog of Dharmendra Mohda, the leader of IBM’s Cognitive Computing division. That’s only 1,542 times slower than real time.
IBM assembled its diagram of the interconnections inside the cerebral cortex of the macaque, a small monkey, as an early model of how the brain works.
IBM’s Watson, of course, is another example of how a computer can interact with humans, absorbing the reams of unstructured data and winning Jeopardy, among other things.
Google itself last year sat down to try and develop its own neural network, and then presented it with data from its own network. The result, as was somewhat widely publicized, was that the network ended up constructing an internal image of a cat, and then spent its computational efforts deciding which YouTube videos were and were not cats.
So how could Google or Microsoft benefit from a federal partnership? On the surface, they might receive federal funding for research. Cognitive computing on the order of what IBM is hoping to achieve, for example, can take millions and millions of dollars, even if the computing resources are already available. (The Times reported that the CalTech meeting was designed to determine if sufficient computing resources were indeed available; the answer is yes, the paper reported.)
Thinking the way that humans think would allow Google or Microsoft to anticipate even more what their users want, and to provide them with that data. Both companies can do that to some extent through data accumulated from millions of users; if the most common “t” word I search for is Twitter.com, Google can start pre-loading the page in the background. But thinking like a human thinks, and making the seemingly random associations that humans make thousands of times faster than we make, could mean everything from artificially-crafted memes to pre-processed sound bites for politicians.
Cognitive computing might not be the singularity, but it’s an important first step. And the race is on.
Image Source: DARPA