A pioneering New York company that once hoped to develop the first artificial intelligence is preparing to declare bankruptcy.
Intelligenesis Corp., which was creating the Webmind software, has been evicted from its Broadway office suite and plans to file for Chapter 7 bankruptcy next week.
"We stopped paychecks at the end of March," said 34-year-old Ben Goertzel, the company's chairman and chief technical officer. "Nobody's been paid in April. But we have 35 people who are still doing work."
Intelligenesis was founded in 1997 and burned through $20 million raised from investors who believed that Goertzel, a former computer science instructor, and his team of researchers could devise a machine intelligence capable of forecasting stock market trends more accurately than a human.
What the company ended up with was about 750,000 lines of Webmind Java code, a fledgling effort to rewrite key components in the more capable C programming language, and multiple rounds of layoffs that demoralized the office and allegedly drove some ex-employees to alcohol and prostitutes.
In many ways, Intelligenesis' tale is the same as any other never-made-a-profit dot-com that has given, say, fuckedcompany.com so much to write about during the last year. (Even the company's website was offline on Wednesday as it moved from the company's former offices to the apartment of a loyal ex-employee.)
But the search for artificial intelligence involves far more than paychecks and mortgages: The creation of a non-human intelligence will be a historic moment, and Webmind attracted more than its share of enthusiasts hoping to be a part of it -- never mind that academics generally believe that such a feat is at least decades away.
That's why Goertzel said he's talking to investors who might want to save his company by writing checks for the sake of science.
"We're trying to raise funds in two directions," he says. "The AI project and not the business angle at all, from individuals who are interested in science. Then there's large corporations who are interested (in stock market predictions)."
He had some successes. In June 2000, for instance, Intelligenesis announced it had partnered with NetCurrents to provide a service that lets corporations monitor what's being said about them in chat rooms, mailing lists and newsgroups. Intelligenesis had 105 employees at the time.
But predicting market movements based on news articles and public statements by officials was just one step toward Webmind's goal of a worldwide, self-organizing machine consciousness -- in other words, it was a way to make money until the AI was done.
During a more optimistic time, Goertzel told the Christian Science Monitor that it was difficult to overemphasize how important his project was. "We are on the verge of a transition equal in magnitude to the advent of intelligence, or the emergence of language," he said in 1998.
Now he's busy calculating what went wrong.
In a fascinating essay about the demise of Intelligenesis, Goertzel said that his mistakes include underestimating how much time Webmind would take, focusing more on research than business, and hiring too many people in too many different locations around the globe.
The essay, titled "Waking Up From the Economy of Dreams," notes the difference between the AI programmers and everyone else.
"I've noticed a rather extreme difference in reaction to the current situation, between people closely involved in engineering the AI Engine and others. Some folks not too close to the AI Engine are rather annoyed about all the time, money and effort spent trying to build real AI. But the people working on the AI Engine have a different feeling.... They can feel the success, the momentum, the power in the work that's been done."
The Webmind codebase currently includes the AI Engine, the Market Predictor, the Classification System, and the Recommendation System, all in varying degrees of completion. Finishing the modules will take $2 million and one to two years of work, Goertzel estimates.
Part of the problem is that the code is far too sluggish to be used as anticipated, due largely to its size and the company's decision to use Java instead of a speedy compiled language like C. Java is an interpreted language that makes development simpler -- but it wasn't intended for large projects and has trouble handling multiple-gigabyte databases.
"The complete knowledge of all words in English and their representations took up 8 GB in RAM," Goertzel says. "In the C version we can get all that into half a gig." He says that preliminary testing shows the C version can deliver a 1,000- to 10,000-times speed increase, and reduce memory usage by a factor of 30.
If Intelligenesis fails to find an investor to save it from bankruptcy, portions of the Webmind seem likely to survive. The New Zealand programmers want to recreate the classification portion of the Webmind project, while the Brazilians are still working on the AI itself.
Goertzel, who is a Brazilian citizen, says he might join that team. "My wife isn't too happy about having to move to Brazil," he says. "My kids are, though."
