From a mouse, a better machine
Baylor team to map rodent’s brain for AI study
An international team of scientists is hoping that the key to building smarter computers and artificial intelligence may be locked inside a tiny portion of a mouse’s brain that they plan on mapping to ultimately allow machines to learn as we do.
The U.S. government’s Intelligence Advanced Research Projects Activity recently announced that it has awarded a $21 million contract to a team of neuroscientists, mathematicians and computer experts to conduct the research, an effort comparable in scope to the wildly successful Human Genome Project.
That initiative was initially met with some skepticism, but it has since allowed scientists to unravel the mysteries of DNA, which in turn transformed medicine and forensic science, and advanced our understanding of human evolution.
Much in the same way, scientists involved in the brain mapping project believe it could revolutionize the fields of neuroscience and artificial intelligence.
Dr. Andreas Tolias, associate professor of neuroscience at Baylor College of Medicine, is one of three principal investigators on the five-year project, which is part of a broader initiative launched by President Barack Obama in 2013 to better understand brain diseases and develop new technology, treatment and cures.
“Our ultimate goal with this project is to study the activity of 100,000 neurons in the brain, to see how they talk to each other, to see how they process information,” Tolias said Thursday. “It’s very complex work.”
Tolias and others involved in the project believe their research might ultimately revolutionize computing, which has already evolved to the point that some programs can recognize facial features, respond to voice commands and operate selfdriving cars.
Just this week, an artificial intelligence system beat the world’s best player of Go — a game that’s significantly more complicated than chess — in a feat that astounded its creators who predicted it would take 10 years for a machine to outwit man.
But computers do have limitations. They rely on artificial neural networks and simplistic algorithms that must analyze information for patterns, in order to learn.
And while computers can analyze volumes of information very quickly, they pale in comparison to the human brain and what it can do, scientists say.
“These artificial neural networks are still incredibly primitive compared to biological neural networks, and don’t learn the way real brains do,” said Xaq Pitkow, co-principal investigator on the project and an assistant professor at both Baylor and Rice University.
“By modeling the brain’s computations and extracting their key features, we think we can give computers the ability to do much better.”
To accomplish such a feat, scientists will map the circuitry between neurons in one cubic millimeter of a mouse’s brain using various techniques.
It will be an immensely challenging job that, up until very recently, wasn’t possible.
“Technologies in both physics and molecular biology have advanced so much that we can now record from many hundreds of neurons at a time with even more extensive recordings on the horizon,” said Jacob Reimer, assistant professor of neuroscience at Baylor. “This lets us analyze neural circuits in ways that we couldn’t dream of just a few years ago.”
Three teams will attempt to map the rodent’s brain: One led by a biologist and computer scientist from Harvard, another led by a computer scientist based at Carnegie Mellon University and Tolias’ colleagues at Baylor.
His team will use a highpowered microscope to look at neuron activity in all six layers of the cortex of the brain. The Harvard team is looking at the top layers of the cortex, while the Carnegie Mellon team will be assigning each neuron a “barcode” and then drawing chemical connections between those barcodes to reconstruct the neural circuitry.
With each of the mapping strategies, scientists hope to record the neural activity, decode it, and ultimately transform the brain’s biological algorithms into improved artificial neural networks for a computer.
In other words, scientists think they can show a computer how to use the same kind of “brain power” humans use when they learn a new task.
So does this mean computers could ultimately develop the same kind of artificial intelligence that Professor Stephen Hawking says he worries about? The same kind of artificial intelligence that wipes out humanity in science fiction movies?
“No, no, no,” said Tolias. “Our ultimate goal is to help humanity. We are talking about using computers in a way that will make people’s lives better.”