In an announcement that had to come from a big tech company sometime this decade, Samsung says it wants to “copy and paste” the human brain onto “neuromorphic” computer chips. The attempt at taking our precious wetware and transferring it to hardware seems to have an unspecified timeline. But if the South Korean conglomerate can figure it out there may be a paradigm shift in the field of artificial intelligence.
As Samsung notes, the company wants to realize a concept that has been in the AI zeitgeist since the ’80s. The idea of neuromorphic computer chips that mimic the way the human brain functions is simultaneously complex and simple. As Samsung’s “copy and paste” description implies, the goal is to take the neurons and synapses of the human brain and embed them in silico—that is, on computer chips. But implementation of the silicon thinking system seems staggering.
The easiest way to understand the tech is as a reverse engineering of the brain. Just as the brain stores its memories in a decentralized way, across 100 billion or so neurons, so will Samsung’s neuromorphic chips. (Hopefully.) This is a nontraditional way of using computer power. Unlike conventional computers, Samsung’s copy and pasted brain wouldn’t divvy up memory and processing tasks; it would perform its tasks asynchronously just as the human brain does. Selecting specific silicon “neurons and synapses” to perform a task at maximum efficiency.
If Samsung can implement these unique computer chips the company would cut down on the power costs of computing. Existing computer systems use far more electricity than a human brain does. A typical desktop can use anywhere from 200 to 400 watts per hour. And a supercomputer? More than two million. A human brain on the other hand only needs 12 watts, or approximately a fifth of what a standard lightbulb uses. And functional neuromorphic chips would be on par with the thinking organ.
Unfortunately, it doesn’t seem like Samsung has laid out a clear path forward for the public yet. And it definitely seems like the company has its work cut out for it. Developing neuromorphic chips will require engineers to eschew current machine-learning algorithms that require enormous amounts of data to work. The neuromorphic chips will also need to mimic the “activity” spikes in the human brain. That requires a whole new type of chip architecture. (Immediately above is an unrelated image of what a neuromorphic chip could look like on the micrometer scale.)
“The vision we present is highly ambitious,” Donhee Ham, a Fellow of Samsung Advanced Institute of Technology and professor at Harvard, said in Samsung’s press release. “But working toward such a heroic goal will push the boundaries of machine intelligence, neuroscience, and semiconductor technology.” Indeed, it sounds like this endeavor is also going to push a lot of human brains to their breaking points.