Web users are being invited to help create the first model of human thought.
By collecting "atoms of information" a Canadian scientist hopes to be able to teach a computer what it means to be human.
Once created, the body of knowledge will be made available to other artificial intelligence researchers trying to make machines smarter or easier to interact with.
The project is expected to run for ten years or however long it takes to collect one billion nuggets of knowledge.
Collecting common sense
The Generic Artificial Consciousness project is the brainchild of researcher Chris McKinstry who wants to make computers think about the world like people do so that they become more useful.
McKinstry says the inspiration for Gac (pronounced Jack) came from work he was doing on a computer that could pass the Turing test.
The test was first suggested by British mathematician Alan Turing as a way to check if a machine was intelligent. Turing thought a test for intelligence would be more useful than a definition that computers had to live up to.
The test was intended to take the form of a conversation conducted via computer screens. Judges have to decide whether they are chatting to a machine or a human.
A machine is judged to have passed the test, and be able to think, when the judges cannot distinguish between its responses and those of a real person.
In 1990, the Turing Test stopped being theoretical with the creation of the Loebner Prize in which computer programs compete to see which one is most human-like.
Tackling Turing
McKinstry says the Turing Test is more an examination of human experience than it is a test of reasoning. But he adds that a computer that sees the world like we do and shares our knowledge of it is likely to be more useful than one that can think.
With Gac he is attempting to create the computer that knows everything that people know about the world and can use it to reason.
McKinstry knows he cannot codify common-sense consensus knowledge all by himself, there is just too much of it. So he is turning to the internet for help and asking anyone who is interested to contribute atoms of knowledge - or "mindpixels" as he calls them - to start building Gac's database.
Each mindpixel takes the form of a statement that can be answered yes or no. The answer must not change from person to person, location or time.
Every time a person submits a mindpixel they are asked to validate at least 20 others that have already been submitted. McKinstry says people should answer questions as the average person would, avoid being pedantic and not to use their expertise.
Knowledge workers
"What we end up doing is building a high resolution digital model of everything that is shared between our heads without all the personal stuff," McKinstry said.
He estimates that around one billion mindpixels will be needed to capture this consensus information. He says Gac could take 10 years to complete because three times as many mindpixels will have to be gathered and winnowed down to the subset that everyone agrees on.
Once complete, the database will be used to train a neural network. "We will reward and punish the design dependent on how it performs against this image of humanness," he said.
He suspects that the finished artificial consciousness will be very good at revealing the patterns and regularities in human thought, and be very useful for anyone designing friendlier computers.
Everyone who submits enough valid mindpixels will be given shares in any company that emerges out of the research project.
He is hoping that the Gac project captures the imagination of the internet population like Seti@home has. That project used idle computers to search for signs of alien life.
"Gac is a distributed project like Seti@home," McKinstry said, "except that I'm not after your computer equipment and your electricity, I'm after your humanity."
Gac has some similarities with another exercise in artificial intelligence known as the Cyc Project. The researchers behind that project are also trying to codify common sense knowledge but are doing so with a much smaller team who have to be conversant with programming and knowledge representation.