Best listening experience is on Chrome, Firefox or Safari. Subscribe to Federal Drive’s daily audio interviews on Apple Podcasts or PodcastOne.
“Ai” used to be a crossword puzzle clue for three-toed sloth, but today it more likely stands for artificial intelligence. In information technology circles, AI is tossed around as a panacea for everything from disease to employee boredom.
If you define AI as the application of algorithms that can learn and evolve as they’re exposed to new data, the use cases do indeed seem endless. The ultimate goal of AI is machinery that acts autonomously but within bounds — such as federal policy on this or that. AI is often and inaccurately described as computers that can think like a human brain. But according to a human brain expert I spoke to recently, don’t hold your breath — that could take another century.
I don’t think serious researchers in AI see the creation of a “brain” as the goal per se. Machines that can make auditable decisions in specific domains are more to the point. Even that gets complicated because of how data source selection can bias even the most capable algorithms.
Dr. Walter Koroshetz, the director of the National Institute of Neurological Disorders and Stroke, pointed out the brain has 85 billion neurons and maybe a trillion interconnections. He said he chuckles at the idea of a computer equaling a human brain in our lifetimes. Even with the level of brain mapping accomplished so far, Koroshetz said, “We may know something about some of the letters the brain uses. But we don’t know the words, we don’t know the sentence structures.”
Until the brain structure and ways it processes information are exponentially better understood, how could anyone build an artificial one?
It turns out the brain research and AI research complement one another in a field called neuropathic computing. Its goal is not simply highly gifted machines, but rather machines which can do their work at power levels approaching those of the brain. Given what it does, the brain is a very low power device, maybe 20 watts. Fun fact: The brain is only 2 percent of the body’s weight but it consumes 20 percent of your metabolic output. Supercomputers that can perform at petaflops (a unit of computing speed equal to 1,000,000,000 million or 1015 floating-point operations per second) practically require their own electrical generating stations.
Cut to the chase: How to get fast results in artificial intelligence
The other day I spoke with Dhireesha Kudithipudi, one of the nation’s leading AI researchers and a computer engineering professor at Rochester Institute of Technology, also my alma mater. She directs the school’s Neuropathic Artificial Intelligence Lab. Among its projects is development of machines running complex algorithms at real-brain power levels. The work involves both software and low-power circuits using memristors. These tiny, nonvolatile components retain memory without using power. Kudithipudi said that brain-inspired computing, the informal term for neuropathic, depends on knowing more about the brain itself and how its pathways work.
Many federal grant-making agencies are pursuing some piece of neuropathic research. Besides NIH, they include the National Science Foundation, National Security Agency and the Air Force.
If the world depended on me for the cutting edge of science and technology, we’d still be wrapped in skins around a fire put, barbecuing goat bones. So I find all of this both interesting and exciting. A convergence of real brain and artificial brain research — what a thought.