To outmatch human smarts requires more than speed and power.
Cross posted at The Fallible Mind, my Psychology Today blog.
What makes humans human? Is it our skill at building things? Certainly not. Even ants design and build complex architectures. Perhaps it is language or culture? But then again, birdsong is language, and the so-called sacred rituals of chimpanzees, or elephant burials aren’t so different from our own. If we ever succeed at creating artificial intelligence in the way that popular culture imagines, should we worry that human cognition will no longer stand at the top?
“We know that there are things that the human mind achieves that even the most powerful supercomputer cannot, but we don’t know why,” says Sir Roger Penrose, a leading authority on consciousness research at the Mathematical Institute of Oxford. Take the chessboard pictured above. When a computer approaches it, it calculates probabilities based on the layout and open positions to conclude that black must win no matter what white does.
But anyone who has played chess knows that the conclusion is untrue. White can win, and often does. Why a computer can’t understand a proposition that is obvious to us isn’t due to a lack of computing power. It’s due to a lack of consciousness.
The high-tech world is abuzz about neural networks, artificial groups of calculating nodes that work in concert, as neurons do, to “learn” from whatever data is fed into them. We can thank neural networks for today’s prosthetic limbs that read electrical impulses from the brain and translate them into movement of the artificial limb. This feat used to belong in the realm of science fiction, but even as impressive as the accomplishment is, neural networks still cannot solve Sir Roger’s chessboard challenge.
Chess computers, even IBM’s Deep Blue that beat world champion Garry Kasparov, are not conscious. And therein lies the problem: Computer scientists model neural networking on our understanding of how brains network, but we understand only so much.
An adult human brain makes up only 2 percent of our total body weight. But it consumes 20 percent of the energy we use, a far greater ratio than in any other animal on the planet. Our energy-hog brain also contains 40 percent more neurons (86 billion) than our closest chimpanzee relative (53 billion). It takes most of that energy just to keep all our neurons alive, so the brain has a built-in energy efficiency plan. For maximum efficiency only 1 percent to 16 percent of neurons can fire at any given moment. This doesn’t mean that nothing is happening elsewhere; all brain areas have their turn at firing.
The seemingly silent areas that appear shut off and not doing anything are actually the parts that make us most human. They underlie executive functions and integrative abilities. They are crucial for reasoning, problem solving, and decision-making. They are the reason why any of us can see in the chessboard what a supercomputer cannot: nuance, strategy, surprise.
Neurons turn out to be only part of the story about intelligence. Just this month British researchers published evidence that as humans evolved, brain blood flow increased more rapidly than brain volume did. Blood carries oxygen, glucose, and other nutrients that support its structure and keep it working. If true, then an increase in fuel may have had more to do with our intellectual advancement than advances in the brain’s structure itself.
We are still in the dark about what happens in the mid ground between a single neuron discharging and the emergent properties that arise from many groups firing simultaneously, such as consciousness and overt behavior. Until we understand that connective middle, AI may be at a standstill.
Don’t worry yet about a Terminator takeover, or even one by IBM’s Watson and its offshoots. For now, consciousness still remains uniquely our own, even if we don’t understand why.