AI project

[malaka]

Many years ago (late 80s, early 90s?), I saw a PBS special on Artificial Intelligence. This new (at the time) project had taken a different angle on the traditional approach. They began by entering thousands of words (adjectives, nouns, verbs, etc.) into a database. Then they taught it sentence structure and some basic "truths". Then they would allow the machine to churn at night and develop its own "ideas". The next day, they would examine each "idea" and tell the program if it was correct or incorrect. For example, the program came up with this hypothesis: If a man holds an electric razor, his entire arm becomes electric. The technician would then instruct the program that this "idea" was incorrect. However, the program was also able to construct correct "ideas" of its own, such as: If Bob is Harry's father, and Sam is Bobs father, then Sam must be Harry's grandfather.

I may have some of the project details incorrect. It was a while ago.

ANYWAY...I was wondering if anyone knew the name of this project or what ever became of it. Any info would be appreciated.

[malaka]

Anyone? Anyone? Bueller?

[JoeyBlades]

Perhaps you're thinking of Doug Lenat's "Cyc".

Here's their home page: http://www.cyc.com/

Here's a page with more Cyc links: http://www.robotwisdom.com/ai/cycresources.html

It's not exactly as you describe, but it sounds similar enough to be what you're remembering...

I know of a few, similar projects, but in none that I know of can the AI spontaneously derive 'ideas'. There are some that can read stories and derive a synopsis of the story, test logical conclusions about the content, etc., but in the end, it still merely number crunching...

[malaka]

Thank you!!! That was exactly what I was looking for! The name "Douglas Lenat" seemed to ring a bell.

As far as machines deriving "ideas", the text below describes what I was thinking about (I think :) ).


From "Hal's Legacy: 2001 as Dream and Reality", by Doug Lenat:

We're now in a position to specify the steps required to bring a HAL-like being into existence.

1. Prime the pump with the millions of everyday terms, concepts, facts, and rules of thumb that comprise human consensus reality -- that is, common sense.

2. On top of this base, construct the ability to communicate in a natural language, such as English. Let the HAL-to-be use that ability to vastly enlarge its knowledge base.

3. Eventually, as it reaches the frontier of human knowledge in some area, there will be no one left to talk to about it, so it will need to perform experiments to make further headway in that area.

...

The goal was to give CYC enough knowledge by the late 1990s to enable it to learn more by means of natural language conversations and reading (step 2). Soon thereafter, say by 2001, we planned to have it learning on its own, by automated-discovery methods guided by models or minitheories of the real world (step 3).


WHat other similar projects are you aware of? Exciting stuff, to be sure.