Biomedical Engineering Blog

Biomedical Engineering

The Biomedical Engineering blog is the place for conversation and discussion about topics related to engineering principles of the medical field. Here, you'll find everything from discussions about emerging medical technologies to advances in medical research. The blog's owner, Chelsey H, is a graduate of Rensselaer Polytechnic Institute (RPI) with a degree in Biomedical Engineering.

Previous in Blog: Engineering Zombies (Part 1)   Next in Blog: A Bigger Belly May Make You Forget

Engineering Synthetic Minds (Part 2)

Posted March 25, 2008 12:01 AM by Steve Melito

The English language contains over 250,000 distinct words, excluding inflections. Yet it's far too simple for artificial intelligence (AI). According to RPI professor Selmer Bringsjord, past AI projects have failed because of the complexity involved in developing character descriptions, an important component of The Turing Test. In order to "pass" this test, a machine must convince a human that it is a person. This natural-language conversation takes the form of a "story", in which the machine (a form of AI) relates aspects of its "human" existence. To impart something as simple as information about a high school attended, that high school must be a predicate attached to buildings in a geographic location, famous alumni, and other data. Multiply this predicate-based requirement across thousands of other pieces of information, and your puny processor may grind to a halt.

The Star Trek Holodeck

Selmer Bringsjord, director of the Rensselaer Artificial Intelligence and Reasoning Laboratory (RAIR), describes his work at the Computational Center for Nanotechnology Innovations (CCNI) as "engineering synthetic characters with minds." He likens the AI avatar he plans to create to the synthetic characters on the Star Trek holodeck, a fictional room which hosts interactive, holographic projections for entertainment and training purposes. Today, Professor Bringsjord is filling his synthetic character with the memories of a graduate student who has agreed to share what EE Times calls "all sorts of facts, figures, family trivia and personal beliefs". Bringsjord's toolbox includes a number of existing algorithms, each of which uses a formula or theorem for an input. If the input is a proof, then the output is "yes". If the input is not, then the output is "no" and a counterexample is provided.

Algorithms and Screenplays

Unfortunately, some of the existing algorithms are not particularly useful in industry. Others require modifications so that they can be used in what Bringsjord calls "a parallel way". Once "parallelized", these new AI algorithms require the use of a supercomputer. But that's not where the story ends. Google the phrase "creating three-dimensional characters" and your search will display hundreds of books about writing screenplays. "These books are the how-to manuals for synthetic characters", Dr. Bringsjord explains. "They tell you what to ask". Although some writers worry that AI researchers will "turn literary creations into a rational, logical end", Bringsjord's goal of a "three-dimensional, cognitively robust narrative" isn't so simple. "The missing link," the RPI professor explains is "leveraging what humans do so well".

But humans aren't always good. Nor are they always rational. That's where CR4's interview with Selmer Bringsjord will continue.

Editor's Note: Part 1, Part 3, and Part 4 of this interview are also available.

Steve Melito - The Y Files

Add to Technorati Favorites


Interested in this topic? By joining CR4 you can "subscribe" to
this discussion and receive notification when new comments are added.

Previous in Blog: Engineering Zombies (Part 1)   Next in Blog: A Bigger Belly May Make You Forget