'People do not walk on their heads.'' The assertion comes about 900 statements deep into the 527,308 items that comprise the Open Mind common sense database. It's after ''Laws are the rules of society'' and before ''The sky is blue during the day.''
This collection of mundane facts, which would take more than 20,000 pages to print out, consists entirely of statements so unremarkable they are barely worth stating. Most of us would correctly dismiss them as common sense.
Which is exactly the point, according to Push Singh, a graduate student at MIT who works out of a tiny office in the university's Media Laboratory.
''After spending years researching artificial intelligence, we still can't build machines that can think about ordinary things the way any 3-year-old can,'' Singh said. This glaring shortcoming is easy to describe: Computers lack common sense. As sophisticated as computing has become, today's programs lack that basic breadth of knowledge that most people share -- the stuff we understand without even thinking about it.
Filling this knowledge gap, however, is an enormous undertaking. Although creating computers with broad-based human knowledge was a well-publicized aim of the research into artificial intelligence in the early 1960s, many scientists in the field have abandoned the project. Not only is it too difficult, but the people who fund research clearly favor more narrowly focused practical AI projects like expert systems and robots, which can perform a narrow range of tasks extremely well.
But now there are signs that ''common sense'' artificial intelligence research may be making a comeback, sparked by projects like Singh's Open Mind database. For the first time, after decades of theoretical research, researchers and programmers have begun using a freely distributed, natural language common sense database to start the process of building common sense into products, programs, and applications.
In fact, as Singh sits in his cramped office in the Media Lab, he's able to point in the direction of a number of MIT researchers using his database for applications that may soon bring common sense AI to consumers.
A few doors down to the right, Barbara Barry, a graduate student in the Media Lab's Interactive Cinema group, is working with Singh to build common sense into video cameras. On the other side of the Media Lab, Henry Lieberman, a research scientist who works with the Software Agents Group, is using common sense to enhance e-mail programs, language translation software, even a search engine. And just outside Singh's office, the Media Lab's ''wearable computing'' group is building common sense into the devices and sensors they believe many of us will be wearing in the future.
''Now that we're at the point where other researchers are building common sense into their applications -- that's when it starts to turn the corner,'' Singh said.
Walter Bender, the executive director of the Media Laboratory, agreed. ''I think the common sense project is a screamer,'' he said. ''We think it's ready to take off as a research area.''
Common sense is still a relatively small area of research. MIT's Artificial Intelligence Laboratory, which dates to the late 1950s and which will soon merge with MIT's Laboratory for Computer Science, is now celebrated for its more practical work with robots rather than for general ''human intelligence'' programming.
In fact, Marvin Minsky, one of the cofounders of the Lab (and Singh's adviser at MIT), recently caused a stir in the field when Wired News reported that he told a Boston University audience that ''AI has been brain-dead since the 1970s.'' The article went on to quote Minsky attacking the current artificial intelligence ''fad'' of making ''stupid little robots.''
Minsky, who is now less actively involved with the field after working at the intersection of computing and human intelligence since the early 1950s, said his remarks sounded more extreme taken out of context. But he did not temper his frustration with the ascendency of ''expert systems'' over common sense research.
''People have made all sorts of specialized systems, but none of them approach the ability to think,'' he said.
''Some researchers feel that they have to make simple things first,'' Minsky said. ''That means there are lots of dumb robots out there. . . . You make a little robot that can smile and the media pays attention, but there's really nothing behind it in terms of intelligence.''
Rodney Brooks, the current director of the Artificial Intelligence Laboratory, who is known for his work with robots, deflected Minsky's criticism.
''I know Marvin believes that we're neglecting common sense, but there's plenty of work for everyone in this field,'' Brooks said. ''It so happens that we're approaching artificial intelligence from the point of view of a real robot with arms pushing things around in the real world. We're studying intelligence by embodying it in a robot playing in the world like a baby. And others are pursuing the same subject by studying vision.
''But none of us are going to get there any time soon, and none of us are going to get there alone.'' Brooks continued. ''We're all approaching the problem of human intelligence from different directions.''
The direction of the Open Mind approach, specifically, is an attempt to tap into the shared common sense of the general public, inviting contributions of common sense statements in natural language, to a website (http:
//openmind.media.mit.edu). Although the project has managed to collect more than half a million statements, Singh believes that perhaps 100 million statements will be required before the group closes in on codifying common sense. (A similar but for-profit project based in Austin, Texas, run by Cycorp Inc., has been collecting common sense statements for nearly two decades; it has a database of more than a million pieces of common sense knowledge.) Singh is also approaching the common sense challenge from a number of other angles. A parallel ''Open Experiences'' database, which hasn't been publicly released, is designed to harvest the common sense contained in stories. Singh is also developing an architecture to structure these statements into something that resembles human cognition.
Whether any of these overlapping initiatives succeed is still in doubt, of course. Ray Kurzweil, the Massachusetts-based inventor, entrepreneur, and author who has been applying artificial intelligence to fields as various as music, medical learning, and the stock market, appreciates the difficulty of the undertaking.
''Common sense sounds like one simple skill, but actually it requires the full range and depth of human intelligence,'' he said.
Kurzweil has written a number of books on the topic (''The Age of Intelligent Machines'' and ''The Age of Spiritual Machines'' are two) and is editor in chief of an online artificial intelligence publication, KurzweilAI.net. He said most of the advances in the field recently have been in ''narrow AI,'' including medical diagnosis and financial decision-making.
Common sense capabilities will come with improved pattern-recognition abilities in computers, rather than simply faster processing speed, Kurzweil said, adding: ''Our brains are massively parallel in their processing, which presents a challenge to computer hardware and software.''
Kurzweil also believes that ''we have to understand the human brain better'' before we will be able to make significant advances in understanding, or replicating, common sense.
''We have to almost reverse-engineer the human brain,'' Kurzweil said. ''In that sense, we're kind of where the genome project was 10 years ago.''
Still, Kurzweil remains optimistic about the long run. He even has a bet with Lotus Development Corp. founder Mitch Kapor that artificial intelligence will develop to the point that, by 2029, a computer will be able to pass the so-called ''Turing Test,'' fooling an interviewer into believing that he or she is interacting with a human. Kurzweil has $10,000 riding on his belief that computers will have common sense in less than 30 years.
In the meantime, it appears that incremental advances will continue and continue to be applied.
''The common sense problem is horribly hard to solve completely, but that doesn't mean we can't already start using what we're learning'' said Lieberman, the Media Lab research scientist.
In fact, Lieberman is already using the Open Mind database to help e-mailers quickly find relevant digital photographs to drop into messages. Sitting in his office, he demonstrates how his program would know that a reference to the ''bride's sister'' in an e-mail message could be illustrated, perhaps, with a photo of the bridesmaids, ''because it's common sense that the bride's sister is often bridesmaid,'' he said.
Lieberman demonstrated a number of other applications that contained common sense. The key, he said, was that the projects are ''fail soft,'' meaning that ''if it works, great . . . if it doesn't, it's not a big deal.''
Which provides a path for common sense to start making its way into the mainstream.
''We're a long way from any kind of complete solution,'' Lieberman said, ''but that doesn't mean that we can't make progress. After all, a little bit of common sense is always better than nothing.''
D.C. Denison can be reached at denison@globe.com.
