Rensselaer Department of Cognitive Science Department of Computer Science
Rensselaer Artificial Intelligence and Reasoning (RAIR) Laboratory
Home Projects People Lectures Symposia/Conferences Sponsors Tour
Psychometric Artificial Intelligence and PERI

PERI
Abstract(+) from the overview paper:

We propose an answer to the "What is AI?" question, namely, that AI is really (or at least really ought in significant part to be) Psychometric AI (PAI). Psychometric AI is the field devoted to building information processing entities capable of at least solid performance on all established, validated tests of intelligence and mental ability, a class of tests that includes IQ tests, tests of reasoning, of creativity, mechanical ability, and so on. Along the way, we: set out and rebut some objections to PAI; describe PERI, a robot in our lab who exemplifies PAI; and briefly treat the future of Psychometric AI, first by pointing toward some promising PAI-based applications, and then by raising some of the "big" philosophical questions the success of Psychometric AI will raise.


Beginnings of PAI:

We have begun our research with the WAIS-R (Wechsler Adult Intelligent Scale - Revised) and the first task of this test has already been surpassed successfully. For reasons of legality we cannot mention the specifics of this subtest, the Block Design Task, but we discuss another similar puzzle which PERI can also solve successfully. This puzzle (shown in snapshots below; compliments of the Binary Arts Corp) is described in more detail in our IJCAI overview paper (see above).




Current work on PERI:
  1. Some of PERI's knowledge is being represented in first-order logic to lay a foundation for natural language understanding and generation.
  2. The Vision System is being finalized.
  3. Introducing Language Capability (NLP/NLG/NLU).
  4. Initiating work on next task: Picture Arrangement (i.e. Narrotological Reasoning).

Building Block Method:


When faced with the challenge to build a complicated system, it's almost always smart to decompose the system into parts, and build them first. PAI is based on the hypothesis that human-level intelligence can be engineered by first building "blocks" able to successfully negotiate various tests of mental ability. A visual respresentation of this is shown above. We seek to test this hypothesis -- e.g., by seeing whether building blocks corresponding to agents able to solve sub-tests on the WAIS ("Arithmetic", "Digital Span", "Information", "Similarities" and "Comprehension") can be weaved together to produce an agent capable of solving problems in elementary algebra.


Background Information:

Research on Psychometric AI and PERI began in December 2001. At present the following software/hardware make up the robotic system:

The 5 DOF vertically articulated mechanical arm is the Intelitek Corporation's Scorbot-ER IX model. Originally attached was a pneumatic two-finger parallel gripper which could be either only completely open or closed around an object. Our new addition includes a three fingered BH8-260 BarrettHand from Barrett Technology Inc. (as shown below). The hand uses servo motors, and can lift a total of 13.2 pounds (6.0 kg), or 4.4 pounds per finger (2.0 kg).

The Vision System is made up of a Cognex MVS-8100M Frame Grabber and a Sony Black-and-White XC55 Video Camera.

The Hearing is made possible by Dragon Naturally-Speaking Professional Solutions Version 6.0 Software, soon to be combined with a NLP/NLG/NLU.

At the core of PERI resides its brain and nervous system - a complex Common Lisp program and associated Scorbot Advanced Control Language Library.


Present and Future Application:

Psychometric AI is quickly becoming a well-known and well-entrenched way to pursue AI. PAI presents a definitive goal for researchers in the field to strive for. In fact, there is considerable evidence that PAI has already taken root, Phase I of Project Halo being just one example. In the Winter 2004 issue of the AI Magazine, the creators of the Project Halo initiative admitted to the appeal of using a standardized test as a very useful evaluation:

It was important... to adapt an existing, well known evaluation methodology with easily understood and objective standards... In the end, a 70-page subset of introductory college-level advanced placement (AP) chemistry was selected because it was reasonably self-contained... (AI Magazine, Winter 2004 Issue, p. 30).
Furthermore there is every reason to think that the future of AI is bound up with PAI. A DARPA/IPTO workshop was held in January 2005 to discuss the (10 year) future topics for the next Grand Challenge, a Cognitive Grand Challenge. A potential topic being considered is building a machine capable of scoring well on the SAT (Scholastic Aptitude Test).
PAI/PERI Project Links
Psychometric AI Home
Recent News
Presentations and Demos
Publications
Media Coverage
Online Library
Restricted Content