David Shamah September 22, 2008
The way we search the web is all wrong, according to Danny Fine of Haifa-based BrainDamage. "When we search for information, we are the ones doing all the work, inefficiently inputting keywords and narrowing down the results until we find what we want. We're supposed to be the masters, not the slaves," he says. "So why are we doing all the work?" Right now, there isn't much option, but when Fine gets through with the Internet, he asserts, it's going to be a whole different place.
There are billions - maybe even trillions of pieces of data on the web, most of which consist of "units" of ideas, eight words in length or less. Nearly all data search engines use a variation of keywords, also known as Latent Semantic Analysis or Indexing (LSA/LSI).
It's a form of artificial intelligence, based in large part on the work of linguist Noam Chomsky, who pioneered the application of mathematical principles to language. The system analyzes documents, creating a map of keywords and the "distance" (in definition) between them.
"The search engine doesn't really understand what you're asking, of course - it's just a dumb computer, after all," Fine tells ISRAEL21c. "The way it figures out what you're looking for is by comparing your request to a long list of keywords that are indexed in a database with other terms that could really be what you're looking for."
That's why most searches produce a few relevant and more irrelevant results; the search engine starts to narrow things down when you click on a link.
Leading questions
That's also why successful searches usually don't contain too many words. "The idea of asking a question of your search engine is almost unimaginable to most people, because of the search method and results we've been taught to accept," Fine says, with the search extremely fast - but often very inaccurate.
But BrainDamage has a different idea in mind. Instead of what he calls the failed linguistic methods used by Google and the rest, Fine proposes a different system to communicate with computers and databases; "Natural thinking technology," which will put the burden of "understanding" on the search engine, enabling it to return far more accurate results than are currently possible.
The BD system does this by assembling a huge database of texts and, using its proprietary and patented system, reassembling the information into logical constructs and ideas with definitions and meanings attached to them. A part of those data constructs is supplying contexts for terms and ideas, so in a case where the question being asked can apply to different situations, the BD engine will seek to clarify the question by asking for more information.
"Our system gathers information and develops it, guided by the user, to reach a conclusion - using the same patterns of logic and ideas human beings do," says Fine.
Take, for example, the sentence "My son was terrorizing us until he got his toys," says Eli Abir, who designed the BrainDamage system and is the company's CTO. Terrorism in this context, of course, means misbehaving, not an acolyte of Bin Laden. Abir says that search engines have no way of knowing this, and as a result give many "false positives. But because BrainDamage's system relies on contextual logic, we can produce much more accurate results every time."
BrainDamage's first application is called Noesis, and is geared to improving search results. But BD's technology, which in essence will teach machines to figure out what humans have in mind when they make a request, can be adapted to almost any other computer-driven operation.
Artificial intelligence at a new level
"Our system advances artificial intelligence far beyond where it is today, enabling computers to truly understand what is being asked of them - and to respond appropriately," Fine says.
Eventually, it could be installed in consumer items like washing machines, or integrated into the phone system, to enable far more complex operations than are currently possible. And, BD's technology, because it relies on contextual logic, will work with any language, with no need for endless sets of keywords in multiple languages.
While BrainDamage's technology is revolutionary, says Fine, he realizes getting the rest of the IT world on board will be a hard sell. "BD's technology was developed by a unique individual, Eli Abir, and it frankly flies in the face of the accepted formulas for artificial intelligence." With BD, Abir has chosen to go up against Chomsky, called by pundits "the most quoted man alive." But Abir - and Fine - say they're up to the challenge.
"When you examine the current body of literature on artificial intelligence, you realize that researchers have hit a brick wall - that there seems to be no way to build the 'intelligent robots' we were told would be doing all the work for us by now at the dawn of the AI era, three decades ago. With BrainDamage, the possibility of machines that can actually understand and 'think,' based on what we tell them to do, becomes a reality," Fine says.
And Fine is logical enough to realize that he needs to give BrainDamage time to blossom as a company. "We're not actively seeking VC money right now, because we realize we have to re-educate the investors as well," he says, adding that BD is not in a hurry to bring in investors, who would likely seek an exit by selling the technology to an Internet giant.
"We really have something revolutionary here, and we intend to see BrainDamage through - until it becomes the standard for communication with computers." However, he adds, BrainDamage has shown its prototype to several major companies, "and to say they were very impressed would definitely be an understatement." The first Internet application based on BD technology should be available to the general public within a year, Fine adds.
In the end, Fine says, the technology world will have to adopt BrainDamage, or something very similar. "A new English word is invented every 90 minutes. There is no way the keepers of the keyword lists will be able to keep up and produce accurate results with that daily volume of new information," he explains.
"The current AI implementation of 'talking' to computers has reached its limit. Once Internet users see the difference between the current method of searching and the one we're implementing, they'll be sold," he adds.
No comments:
Post a Comment