I was sitting in New York’s Bryant Park one recent fall afternoon and admiring an unheralded triumph of human science and communication: the crowded sidewalk. At the intersection of 6th Avenue and 42nd Street, people were streaming in all directions, darting this way and that, and yet they almost never bumped into each other.
Sidewalk traffic is improvised choreography, and its magic starts with math. Each person, like a dancer in a Broadway musical, unconsciously calculates his or her own speed and velocity, and those of people coming from other directions. A map of these paths, or vectors, is quietly drawn in each person’s brain. It highlights coming crashes, and that’s where communication comes in. Two people on a collision course look at each other. With slight glances and gestures they send signals: I’ll go here, you go there. It’s subtle and sophisticated. But because we have been tuned into each other through millennia, we pull it off with just a fraction of our brainpower, often while listening to music or talking on the phone.
These small miracles are the ones that baffle the machines pushing into our cognitive realm. I spent a year with scientists at IBM as they developed Watson, the Jeopardy-playing computer. And while Watson can dig through mountains of data and answer questions about Zoroastrian priests or the chemical components of aspirin, it does not understand people. It has never had a body or felt an emotion, and it doesn’t know what it is to communicate with another person. In fact, it doesn’t “know” anything. Believe me, when it comes to intelligence, Watson is far closer to its fellow tools, the thermostat and the calculator, than it is to cats or monkeys, much less to human beings.
Yet Watson operates in the realm of language. So does Apple’s new digital assistant, Siri. And because these software programs appear to converse and often provide correct answers, they’re routinely compared to people. Their humanity also fits into their marketing, since it draws attention to them and seems magical. Yet the false promise of human intelligence leads to the following responses, each dangerous in its own way:
1) Fear: Machines are going to surpass us in intelligence and, perhaps, eventually run our lives.
2) Dismissal: The machines are idiotic. They don’t know anything and only search for statistical correlations among lines of ones and zeros. They don’t pose any challenge to us.
Both responses are misguided. Fear leads people to resist the advance of technology, which has the power to make our lives safer, cleaner, healthier and more prosperous. And dismissal of the machine’s “intelligence” distracts people from the transformations ahead. It’s akin to pooh-poohing the work of an engineer who builds a tool infinitely clumsier than the human hand, and failing to see that his machine–the bulldozer–boasts extraordinary powers of its own.
When I was starting research on my book about the Watson project, Final Jeopardy, I believed that scientists, using the latest brain research and ever more powerful computers, might be able to build a truly smart machine within a decade or two. It saw a brain machine as a cognitive moon shot.
But then I went to a Singularity conference in San Francisco. One of the speakers, a Cambridge biologist named Dennis Bray, spent an hour describing the phenomenal complexity within a single cell. Considering that the each brain houses an estimated 100 billion neurons, and quadrillions of connections among them, how can scientists, he asked, be expected to decode intelligence? In investigating the most intricately known piece of circuitry in the universe, they’re sure to find layer upon layer of complexity.
A few weeks later, I discussed this with Joshua Tenenbaum, an MIT professor in computational cognitive science. The 1969 moon shot, he said, was the work of engineers who were building on five centuries of scientific research into chemistry, physics, astronomy and metallurgy. When it comes to the brain, he said, “we’re just now at Galileo.”
This doesn’t mean that engineers won’t be building phenomenally powerful computers. But like Watson, these machines are going to be optimizing transport systems, predicting the path of epidemics and answering our questions through their own statistical methods, not ours.
When it comes to teaching them to “think” like humans, even small steps require extensive research. One example. The U.S. Defense Department want their electronic monitors around the world to piece together patterns of data and formulate their own conclusions–as a human would. Say that a machine “sees” a man going into a building with a package. Ten minutes later the same person emerges without it. A four-year-old child would instantly conclude that he left the package there, or perhaps gave it to someone. But teaching a computer to generate such hypotheses based on observations will take years of work. This is the focus on one project, Mind’s Eye, sponsored by the DARPA, the Pentagon’s research wing. It will take years and millions of dollars for researchers at 12 leading universities to lift the visual intelligence of a computer to that of a small child.
This isn’t necessarily bad news. In a time of high unemployment, computers are leaving some jobs for people. And at least for the next decade or two, they’ll be working for us, and not vice versa.
Top image: Crowded New York street. Courtesy Flickr user Pixagraphic.
Stephen Baker is the author of Final Jeopardy: Man Vs. Machine and the Quest to Know Everything (2011: HMH). He was formerly a senior writer covering technology for BusinessWeek.