Harvard University is set to develop new artificial intelligence systems that work faster, smarter and perhaps better than the human brain. The Intelligence Advanced Research Projects Activity (IARPA) has awarded $28 million to fund this project.
The challenge is to figure out why the brain is so adept at learning, and then design computer systems able to interpret, analyze and absorb information the way humans do.
Researchers will then record brain activity, particularly activity in the visual cortex, in a fashion never done before. They will map "connections at a scale never before attempted" and reverse-engineer the data to create better computer algorithms.
Such algorithms will be used to come up with a detailed 3D neural map. The venture is a moonshot challenge similar to the Human Genome Project in scope, says project leader David Cox, assistant professor of computer scince and molecular and cellular biology.
“The scientific value of recording the activity of so many neurons and mapping their connections alone is enormous, but that is only the first half of the project,” he says in a statement.
Cox adds his team will also probe the basic principles of how the human brain learns and design computers matching or outperforming its ability.
Many computers today demonstrate storage power similar to the brain, but they remain inferior to the latter’s ability to recognize patterns and learn information.
Experts estimate that the brain can store from 10 to 100 terabytes of information, with some pegging it at almost 2.5 petabytes. Just over one petabyte of data is equivalent to around 1.6 million CDs' worth of information.
While a human only needs to look at a vehicle a couple of times to recognize one, an AI system needs to process samples by the hundreds or thousands to retain those details.
The multi-phase project will start in Cox’s laboratory, where they will train rats to visually recognize different items on a computer screen and then record activity inside the visual cortex. A portion of the rat’s brain will then be studied at the lab of project member Jeff Lichtman, where it will be sliced into ultra-thin pieces and imaged.
“We are very excited to get started but have no illusions that this will be easy,” Lichtman says.
These AI systems are projected to be useful for detecting network invasions, reading MRI images, driving cars and – if humanity gets lucky – just about any human task imaginable.
