Monday, October 11, 2010

Google Cars Drive Themselves, in Traffic

Anyone driving the twists of Highway 1 between San Francisco and Los Angeles recently may have glimpsed a Toyota Prius with a curious funnel-like cylinder on the roof. Harder to notice was that the person at the wheel was not actually driving.

The car is a project of Google, which has been working in secret but in plain view on vehicles that can drive themselves, using artificial-intelligence software that can sense anything near the car and mimic the decisions made by a human driver.

With someone behind the wheel to take control if something goes awry and a technician in the passenger seat to monitor the navigation system, seven test cars have driven 1,000 miles without human intervention and more than 140,000 miles with only occasional human control. One even drove itself down Lombard Street in San Francisco, one of the steepest and curviest streets in the nation. The only accident, engineers said, was when one Google car was rear-ended while stopped at a traffic light.

Autonomous cars are years from mass production, but technologists who have long dreamed of them believe that they can transform society as profoundly as the Internet has.

Robot drivers react faster than humans, have 360-degree perception and do not get distracted, sleepy or intoxicated, the engineers argue. They speak in terms of lives saved and injuries avoided — more than 37,000 people died in car accidents in the United States in 2008. The engineers say the technology could double the capacity of roads by allowing cars to drive more safely while closer together. Because the robot cars would eventually be less likely to crash, they could be built lighter, reducing fuel consumption. But of course, to be truly safer, the cars must be far more reliable than, say, today’s personal computers, which crash on occasion and are frequently infected.

The Google research program using artificial intelligence to revolutionize the automobile is proof that the company’s ambitions reach beyond the search engine business. The program is also a departure from the mainstream of innovation in Silicon Valley, which has veered toward social networks and Hollywood-style digital media.

During a half-hour drive beginning on Google’s campus 35 miles south of San Francisco last Wednesday, a Prius equipped with a variety of sensors and following a route programmed into the GPS navigation system nimbly accelerated in the entrance lane and merged into fast-moving traffic on Highway 101, the freeway through Silicon Valley.

It drove at the speed limit, which it knew because the limit for every road is included in its database, and left the freeway several exits later. The device atop the car produced a detailed map of the environment.

The car then drove in city traffic through Mountain View, stopping for lights and stop signs, as well as making announcements like “approaching a crosswalk” (to warn the human at the wheel) or “turn ahead” in a pleasant female voice. This same pleasant voice would, engineers said, alert the driver if a master control system detected anything amiss with the various sensors.

The car can be programmed for different driving personalities — from cautious, in which it is more likely to yield to another car, to aggressive, where it is more likely to go first.

Christopher Urmson, a Carnegie Mellon University robotics scientist, was behind the wheel but not using it. To gain control of the car he has to do one of three things: hit a red button near his right hand, touch the brake or turn the steering wheel. He did so twice, once when a bicyclist ran a red light and again when a car in front stopped and began to back into a parking space. But the car seemed likely to have prevented an accident itself.

When he returned to automated “cruise” mode, the car gave a little “whir” meant to evoke going into warp drive on “Star Trek,” and Dr. Urmson was able to rest his hands by his sides or gesticulate when talking to a passenger in the back seat. He said the cars did attract attention, but people seem to think they are just the next generation of the Street View cars that Google uses to take photographs and collect data for its maps.

The project is the brainchild of Sebastian Thrun, the 43-year-old director of the Stanford Artificial Intelligence Laboratory, a Google engineer and the co-inventor of the Street View mapping service.

In 2005, he led a team of Stanford students and faculty members in designing the Stanley robot car, winning the second Grand Challenge of the Defense Advanced Research Projects Agency, a $2 million Pentagon prize for driving autonomously over 132 miles in the desert.

Besides the team of 15 engineers working on the current project, Google hired more than a dozen people, each with a spotless driving record, to sit in the driver’s seat, paying $15 an hour or more. Google is using six Priuses and an Audi TT in the project.

The Google researchers said the company did not yet have a clear plan to create a business from the experiments. Dr. Thrun is known as a passionate promoter of the potential to use robotic vehicles to make highways safer and lower the nation’s energy costs. It is a commitment shared by Larry Page, Google’s co-founder, according to several people familiar with the project.

The self-driving car initiative is an example of Google’s willingness to gamble on technology that may not pay off for years, Dr. Thrun said. Even the most optimistic predictions put the deployment of the technology more than eight years away.

One way Google might be able to profit is to provide information and navigation services for makers of autonomous vehicles. Or, it might sell or give away the navigation technology itself, much as it offers its Android smart phone system to cellphone companies.

But the advent of autonomous vehicles poses thorny legal issues, the Google researchers acknowledged. Under current law, a human must be in control of a car at all times, but what does that mean if the human is not really paying attention as the car crosses through, say, a school zone, figuring that the robot is driving more safely than he would?

And in the event of an accident, who would be liable — the person behind the wheel or the maker of the software?

“The technology is ahead of the law in many areas,” said Bernard Lu, senior staff counsel for the California Department of Motor Vehicles. “If you look at the vehicle code, there are dozens of laws pertaining to the driver of a vehicle, and they all presume to have a human being operating the vehicle.”

The Google researchers said they had carefully examined California’s motor vehicle regulations and determined that because a human driver can override any error, the experimental cars are legal. Mr. Lu agreed.

Scientists and engineers have been designing autonomous vehicles since the mid-1960s, but crucial innovation happened in 2004 when the Pentagon’s research arm began its Grand Challenge.

The first contest ended in failure, but in 2005, Dr. Thrun’s Stanford team built the car that won a race with a rival vehicle built by a team from Carnegie Mellon University. Less than two years later, another event proved that autonomous vehicles could drive safely in urban settings.

Advances have been so encouraging that Dr. Thrun sounds like an evangelist when he speaks of robot cars. There is their potential to reduce fuel consumption by eliminating heavy-footed stop-and-go drivers and, given the reduced possibility of accidents, to ultimately build more lightweight vehicles.

There is even the farther-off prospect of cars that do not need anyone behind the wheel. That would allow the cars to be summoned electronically, so that people could share them. Fewer cars would then be needed, reducing the need for parking spaces, which consume valuable land.

And, of course, the cars could save humans from themselves. “Can we text twice as much while driving, without the guilt?” Dr. Thrun said in a recent talk. “Yes, we can, if only cars will drive themselves.”

Sunday, October 10, 2010

Aiming to Learn as We Do, a Machine Teaches Itself: NELL, the Never-Ending Language Learning system

Give a computer a task that can be crisply defined — win at chess, predict the weather — and the machine bests humans nearly every time. Yet when problems are nuanced or ambiguous, or require combining varied sources of information, computers are no match for human intelligence.

Few challenges in computing loom larger than unraveling semantics, understanding the meaning of language. One reason is that the meaning of words and phrases hinges not only on their context, but also on background knowledge that humans learn over years, day after day.

Since the start of the year, a team of researchers atCarnegie Mellon University — supported by grants from the Defense Advanced Research Projects Agency andGoogle, and tapping into a research supercomputing cluster provided by Yahoo — has been fine-tuning a computer system that is trying to master semantics by learning more like a human. Its beating hardware heart is a sleek, silver-gray computer — calculating 24 hours a day, seven days a week — that resides in a basement computer center at the university, in Pittsburgh. The computer was primed by the researchers with some basic knowledge in various categories and set loose on the Web with a mission to teach itself.

“For all the advances in computer science, we still don’t have a computer that can learn as humans do, cumulatively, over the long term,” said the team’s leader, Tom M. Mitchell, a computer scientist and chairman of the machine learning department.

The Never-Ending Language Learning system, or NELL, has made an impressive showing so far. NELL scans hundreds of millions of Web pages for text patterns that it uses to learn facts, 390,000 to date, with an estimated accuracy of 87 percent. These facts are grouped into semantic categories — cities, companies, sports teams, actors, universities, plants and 274 others. The category facts are things like “San Francisco is a city” and “sunflower is a plant.”

NELL also learns facts that are relations between members of two categories. For example, Peyton Manning is a football player (category). The Indianapolis Colts is a football team (category). By scanning text patterns, NELL can infer with a high probability that Peyton Manning plays for the Indianapolis Colts — even if it has never read that Mr. Manning plays for the Colts. “Plays for” is a relation, and there are 280 kinds of relations. The number of categories and relations has more than doubled since earlier this year, and will steadily expand.

The learned facts are continuously added to NELL’s growing database, which the researchers call a “knowledge base.” A larger pool of facts, Dr. Mitchell says, will help refine NELL’s learning algorithms so that it finds facts on the Web more accurately and more efficiently over time.

NELL is one project in a widening field of research and investment aimed at enabling computers to better understand the meaning of language. Many of these efforts tap the Web as a rich trove of text to assemble structured ontologies — formal descriptions of concepts and relationships — to help computers mimic human understanding. The ideal has been discussed for years, and more than a decade ago Sir Tim Berners-Lee, who invented the underlying software for the World Wide Web, sketched his vision of a “semantic Web.”

Today, ever-faster computers, an explosion of Web data and improved software techniques are opening the door to rapid progress. Scientists at universities, government labs, Google, Microsoft, I.B.M. and elsewhere are pursuing breakthroughs, along somewhat different paths.

For example, I.B.M.’s “question answering” machine, Watson, shows remarkable semantic understanding in fields like history, literature and sports as it plays the quiz show “Jeopardy!” Google Squared, a research project at the Internet search giant, demonstrates ample grasp of semantic categories as it finds and presents information from around the Web on search topics like “U.S. presidents” and “cheeses.”

Still, artificial intelligence experts agree that the Carnegie Mellon approach is innovative. Many semantic learning systems, they note, are more passive learners, largely hand-crafted by human programmers, while NELL is highly automated. “What’s exciting and significant about it is the continuous learning, as if NELL is exercising curiosity on its own, with little human help,” said Oren Etzioni, a computer scientist at the University of Washington, who leads a project called TextRunner, which reads the Web to extract facts.

Computers that understand language, experts say, promise a big payoff someday. The potential applications range from smarter search (supplying natural-language answers to search queries, not just links to Web pages) to virtual personal assistants that can reply to questions in specific disciplines or activities like health, education, travel and shopping.

“The technology is really maturing, and will increasingly be used to gain understanding,” said Alfred Spector, vice president of research for Google. “We’re on the verge now in this semantic world.”

With NELL, the researchers built a base of knowledge, seeding each kind of category or relation with 10 to 15 examples that are true. In the category for emotions, for example: “Anger is an emotion.” “Bliss is an emotion.” And about a dozen more.

Then NELL gets to work. Its tools include programs that extract and classify text phrases from the Web, programs that look for patterns and correlations, and programs that learn rules. For example, when the computer system reads the phrase “Pikes Peak,” it studies the structure — two words, each beginning with a capital letter, and the last word is Peak. That structure alone might make it probable that Pikes Peak is a mountain. But NELL also reads in several ways. It will mine for text phrases that surround Pikes Peak and similar noun phrases repeatedly. For example, “I climbed XXX.”

NELL, Dr. Mitchell explains, is designed to be able to grapple with words in different contexts, by deploying a hierarchy of rules to resolve ambiguity. This kind of nuanced judgment tends to flummox computers. “But as it turns out, a system like this works much better if you force it to learn many things, hundreds at once,” he said.

For example, the text-phrase structure “I climbed XXX” very often occurs with a mountain. But when NELL reads, “I climbed stairs,” it has previously learned with great certainty that “stairs” belongs to the category “building part.” “It self-corrects when it has more information, as it learns more,” Dr. Mitchell explained.

NELL, he says, is just getting under way, and its growing knowledge base of facts and relations is intended as a foundation for improving machine intelligence. Dr. Mitchell offers an example of the kind of knowledge NELL cannot manage today, but may someday. Take two similar sentences, he said. “The girl caught the butterfly with the spots.” And, “The girl caught the butterfly with the net.”

A human reader, he noted, inherently understands that girls hold nets, and girls are not usually spotted. So, in the first sentence, “spots” is associated with “butterfly,” and in the second, “net” with “girl.”

“That’s obvious to a person, but it’s not obvious to a computer,” Dr. Mitchell said. “So much of human language is background knowledge, knowledge accumulated over time. That’s where NELL is headed, and the challenge is how to get that knowledge.”

A helping hand from humans, occasionally, will be part of the answer. For the first six months, NELL ran unassisted. But the research team noticed that while it did well with most categories and relations, its accuracy on about one-fourth of them trailed well behind. Starting in June, the researchers began scanning each category and relation for about five minutes every two weeks. When they find blatant errors, they label and correct them, putting NELL’s learning engine back on track.

When Dr. Mitchell scanned the “baked goods” category recently, he noticed a clear pattern. NELL was at first quite accurate, easily identifying all kinds of pies, breads, cakes and cookies as baked goods. But things went awry after NELL’s noun-phrase classifier decided “Internet cookies” was a baked good. (Its database related to baked goods or the Internet apparently lacked the knowledge to correct the mistake.)

NELL had read the sentence “I deleted my Internet cookies.” So when it read “I deleted my files,” it decided “files” was probably a baked good, too. “It started this whole avalanche of mistakes,” Dr. Mitchell said. He corrected the Internet cookies error and restarted NELL’s bakery education.

His ideal, Dr. Mitchell said, was a computer system that could learn continuously with no need for human assistance. “We’re not there yet,” he said. “But you and I don’t learn in isolation either.”