Where did Dr. Eliza come from?

In 1966, Joseph Weizenbaum wrote a computer program that he called Eliza, which could carry on a brief and very limited conversation. Without anyone realizing it at the time, this incorporated the good idea that statements of symptoms could easily be recognized using shallow parsing methods; and the bad idea that a useful conversation could be carried on by simply manipulating words.

In 1976, still without realizing the two separate but merged ideas that were involved, Joseph Weizenbaum wrote Computer Power and Human Reason that erroneously "debunked" the idea that simple programs could do really powerful real-world things in natural language. This effectively killed further research into shallow parsing methods.

In 2001, Steve Richfield engineered a new experimental cure for his idiopathic atrial fibrillation, requiring 4 months of intensive engineering effort to complete (plus a year of recovery). Steve then realized that his 4 months of effort should have been possible in just a few minutes on a much more enlightened Internet.

Steve then got a contract to repair some C-130H military aircraft simulators where other experts had failed. This not only required using everything Steve knew, but Steve had to develop new methods to utilize the scant available information to find subtle intermittent problems buried in circuit boards with 300 chips on them, 60 of which contained custom programming. Further, most of the documentation had been lost.

Meanwhile, Eleanor Richfield, Steve's daughter, took on the project of finding a cure for a familial metabolic condition that had been killing her family members for generations. Using Steve's approaches, Eleanor was able to find her cure in about 4 months. Soon, everyone in the family tree with this condition was cured.

In 2002, Steve brought Eleanor in as his apprentice to assist in repairing the aircraft simulators. After a couple of weeks, Eleanor remarked that repairing people and repairing electronics involves basically the same processes, and that only the tools and details are different. After lengthy discussions, it became obvious that a relatively simple computer program should be able to solve most of mankind's most difficult problems, and do so much better than most human "experts".

During the 2003-2004 timeframe, first experimental code, then demo code was written to bring logical rigor to difficult problem solving. The goal evolved to "Create Eliza, as it might have evolved in 40 years, if only Weizenbaum hadn't trashed the concept". This included many discussions with various experts on the Internet and in person at conferences.

In 2005 at the Int'l Conference on Artificial Intelligence WORLDCOMP'05, Eleanor Richfield presented the demo code for Dr. Eliza written by Steve Richfield.

In 2006, Steve created Gracie to interface speech recognition and synthesis products to Dr. Eliza. This made it possible to discuss problems in a back-and-forth unstructured conversation in real time. However, this proved to be a step in the wrong direction, because people with problems serious enough to write them up were not about to trust their solution to the (believed) unreliability of speech I/O products, even though the actual error rate was at about that of humans.

In 2008, Dr. Eliza was put up on the Internet for live processing of problem statements. However, Dr. Eliza's knowledge base was still quite limited, so while Dr. Eliza is clearly on a track to greatly exceeding human problem solving skills, this will require many months more work to become routine.

If you want to keep up to date on Dr. Eliza's development, then email us at News@DrEliza.com.