Rensselaer Research Review
Home Page Awards & Grants Patents Accolades Back Issues RPI Research
*
*
*

Building a Better Brain

Rensselaer’s Cognitive Science Department conducts research aimed at understanding how humans think. Ultimately, that knowledge will be used to build better artificial intelligence systems and to engineer improved interfaces between artificial and natural cognitive systems.

By Sheila Nason     Printer-friendly PDF version

*
*
*
*
*
*
* Bettina Schimanski and Selmer Bringsjord with PERI

Intelligence-Tested: Bettina Schimanski and Selmer Bringsjord with PERI. Photo by Mark McCarty.
*
* *
Robots aren’t very smart, according to Selmer Bringsjord, Rensselaer professor of cognitive science. A clever toddler can conduct a more understandable conversation than the best of today’s artificial intelligence (AI) devices. A computer may be able to win at chess by quickly testing vast numbers of possible moves, but if someone changes the rules of the game, the computer will be stymied, and the human will win.

“We are still the most intelligent entities in the known universe,” says Bringsjord, who chairs Rensselaer’s Department of Cognitive Science.

But humans want smarter AI systems to do everything from entertaining us in the newest interactive game to simulating battlefield scenarios as an aid to combat planning.

Rensselaer’s cognitive science researchers are working on the problem. They conduct research aimed at understanding how humans think and then make use of that knowledge to build better artificial intelligence systems and to engineer improved interfaces between artificial and natural cognitive systems.

An Irresistible Subject

Cognitive science — understanding why we’re so smart — is an irresistible subject to humans, says Bringsjord. Studies go back, arguably, to Aristotle and certainly to Hobbes, who speculated that all human intelligence consists of mechanical computations.

On a more practical level, cognitive science occupies a central place in an interactive entertainment industry that brings in billions of dollars. From movies without human characters to video games to displays at the Universal Studios theme park, there is a demand for more realistic synthetic characters and better interfaces. Cognitive science is also of great interest to national defense agencies for, among other things, improved intelligence analysis and more effective personnel training.

Rensselaer’s Cognitive Science Department was formed in 2001 from the former Philosophy, Psychology, and Cognitive Sciences Departments. The highly interdisciplinary program brings together computer engineering and such social sciences as psychology, a combination that is increasingly necessary in a world in which humans must constantly interact with ever-more-complex thinking machines. Today the department consists of 12 tenured faculty members, two research professors, four clinical faculty members, one adjunct professor, and two post-docs.

Yingrui Yang, an assistant professor who specializes in human reasoning and decision-making, was the first faculty hire. He collaborates with Bringsjord on war-gaming work. Assistant Professor Brett Fajen also joined the faculty in 2001. His expertise is in perception and action, dynamical systems modeling, and virtual environments.

Professors Wayne Gray and Ron Sun brought their expertise in computational cognitive modeling to the department. The newest faculty addition is Assistant Professor Nick Cassimatis, who had been working at the Naval Research Laboratory’s AI Center to improve the conversational abilities of robots.

Members of Rensselaer’s Cognitive Science Department conduct research in next-generation AI, next-generation computational cognitive modeling, and cognitive engineering. Many projects involve aspects of all three research areas.

AI Systems That Can Pass IQ Tests

*
*
Click image to start movie

The goal of the robot is to track the orange robot it sees to the right of the scene in the beginning. Notice how it revises its plan, in the middle of executing it, because of new sensor information and physical reasoning. (requires QuickTime)

*
* *
A cognitive system — human or artificial — can perceive information in the environment, analyze that information, and act upon it, according to Bringsjord, who directs the Rensselaer Artificial Intelligence and Reasoning (RAIR) Laboratory. Bringsjord argues in favor of the term “Psychometric Artificial Intelligence,” which is taken from psychometrics, the science of measuring and testing psychological properties. Under this definition, an artificial agent is intelligent only if it excels at all established, validated tests of intelligence.

Bringsjord led the team that developed PERI (Psychometric Experimental Robotic Intelligence), a 3-foot tall robot with a vision system and a dexterous manipulator. The AI used in PERI is based on logic instead of on patterns, mathematics, and numerical computing. Using these talents, PERI became the first robot to pass part of an IQ test. The robot achieved a perfect score on a section of the Wechsler Adult Intelligence Scale-Revised, a timed test in which colored blocks are arranged to match card designs.

At present, Bringsjord’s team is working to help PERI pass two harder sections of the IQ test. In one, examinees are given pieces of an object. They must figure out what it is and put the pieces together. In the other, examinees are given a group of snapshots of people, and they must assemble the pieces to make a coherent story.

Bringsjord, who is a consultant to gaming, multimedia, and e-learning companies, has also built Brutus, a creative AI system that can write short stories of up to 500 words based on the concepts of betrayal, deception, and evil.

RAIR lab researchers are interested in the entire concept of advanced synthetic characters, whether they appear as robots or as characters in the advanced computer games developed in the RAIR lab. At present, the group is working on “E,” an evil character who is very good at lying.

Bringsjord and Research Professor Konstantine Arkoudas are funded by the Defense Advanced Research Projects Agency (DARPA) to build an intelligent system that can learn by reading. The system will be based on MARMML (Multi-Agent Reasoning and Mental Metalogic), which was developed by Bringsjord and others in earlier work.

Unlike current systems, the intelligent machine envisioned by Bringsjord and Arkoudas would be able to read about a new topic and create mental models of the material, to weave the new knowledge with what it already knows, and to understand context issues such as who wrote the article and why. It would also be able to make inferences from the material and to communicate about the topic in English. Finally, it would be able to work cooperatively as part of a team of machines.

Cassimatis, who works on integrated models of language and physical reasoning, says most people don’t realize how many aspects there are to human conversation. The ear records sound waves, and the brain knows what syllables they represent. The mind understands how the words fit together linguistically. The listener fits the new information into what is already known and, at the same time, gets into the speaker’s mind, reaching conclusions about what the speaker is trying to say. All of these processes are currently modeled using different computational techniques. He is working to integrate those processes to simulate the process humans perform naturally and instantaneously.

*
*
*
*
* Continued
“Building a Better Brain”  Page:  1 | 2     Next Page
*
*
*

Home | Awards & Grants | Recent Patents | Accolades | Back Issues | RPI Research

   Rensselaer Polytechnic Institute
   
110 8th Street, Troy, NY USA 12180
Copyright ©2004 Rensselaer Polytechnic Institute.
Published quarterly by the Rensselaer Office of Communications in collaboration with the Office of Research.

Molecularium(SM) and associated characters are service marks of Rensselaer Polytechnic Institute. All rights are reserved.