Forum for Artificial Intelligence


About FAI

The Forum for Artificial Intelligence meets every other Friday at 3pm in the ACES Auditorium to discuss topics in artificial intelligence. After the formal talk, we continue our conversation at the Crown and Anchor. All are welcome to attend.

Please send questions or comments to Harold Chaput or Tal Tversky.

Full Schedule: 9/2210/610/2011/311/1712/1

September 22nd

David Quinto [email]
Department of Linguistics

Then There's Modality: What the Deaf and Deaf-Blind have to say about language

Over the last three decades, linguists have examined language in the visual-gestural modality. American Sign Language and European sign languages have received the most attention, but studies of Asian, Middle-Eastern, and Latin American sign languages have also been conducted. Striking similarities to spoken language have become evident through this research, but interesting differences have also been pointed out.

In addition, recent work on sign language as it is used by Deaf and Blind people in this country has revealed some differences between tactile signed language and American Sign Language as it is used by sighted Deaf people.

Based on this preliminary work on tactile signed language, we will discuss some of the differences and similarities between the structure and form of language in the visual-gestural and tactual-gestural modalities. Of course, we will also make comparisons to auditory-oral language (i.e., spoken language) throughout our discussion.

By examining language in different modalities--paying attention to what is similar as well as what is different across modalities--we can hope to come closer to a general theory of language and the brain.

October 6th

Adrian Agogino [email] [web]
Electrical and Computer Engineering

Collective Intelligence

There are many problems that can only be solved by the joint action of large communities of computers each running a sophisticated machine learning algorithm, where those algorithms are not subject to centralized, global control. Examples are routing of air traffic, control of swarms of spacecraft, routing of packets across the Internet, control of multiple Mars rovers and communication between the multiple processors in a modern computer. The mathematics of "COllective Intelligence" (COINs) is concerned with the design of multi-agent systems in order to optimize an overall global utility function when those systems lack centralized communication and control.

Typically in COINs each agent runs a distinct Reinforcement Learning (RL) algorithm so that much of the design problem reduces to how best to initialize/update each agent's private utility function, so as to avoid their working at cross purposes as far as the global utility is concerned. Traditional "team game" solutions to this problem assign to each agent the global utility as its private utility function. In our work we used the COIN framework to derive the alternative "Wonderful Life Utility" (WLU), and experimentally established that having the agents use it induces global utility performance up to orders of magnitude superior to that induced by use of the team game utility.

October 20th

Dr. Gerhard Werner [email] [web]
Department of Biomedical Engineering

The Return of Simon's Ant

Starting from the premise that all cognitive systems (human and artefactual) are dynamical systems, I will sketch briefly the trajectory of evolution of the various stages of the symbolicist and the connectionist hypotheses. This is to set the stage for the recent challenge by the dynamicist theory of cognition, which I will discuss in some detail. The principle illustrated by the brief parable of "Simon's ant" will then serve to illustrate the role of complexity for Cognition: I will consider it as a promising, though essentially neglected, extension of the dynamicist theory to view cognizers and environment jointly as complex interacting systems. I will suggest that this aspect has not received the attention it merits, despite the promise it holds for a more realistic understanding of cognition, both human and artefactual.

November 3rd

Prof. Brad Love [email] [web]
Department of Psychology

Modeling Human Category Learning

Current models and findings in human category learning research will be considered. Human learning data from a variety of learning modes (including inference-based and classification learning) will be overviewed. Two models (ALCOVE and SUSTAIN) will be applied to the data. ALCOVE is an exemplar based model that performs abstraction by interpolating among stored examples. In ALCOVE, all abstraction is indirect. SUSTAIN is a clustering model that recruits clusters in response to prediction errors. In SUSTAIN, all abstraction is direct. Although both models are very different, they tackle the human learning data in the same fashion. Still, the complete pattern of results favors SUSTAIN's account of human learning -- it appears that what is stored in memory depends critically on the structure of the learning problem and the learning mode engaged.

November 17th

UT Robotics Laboratory [web]
Patrick Beeson [email] [web]
Joseph Modayil [email] [web]
Jefferson Provost [email] [web]
Department of Computer Sciences

R2D2, Where Are You?

Robot navigation requires a cognitive map. The Spatial Semantic Hierarchy is a multilevel representation of the cognitive map. We will discuss our research in implementing the SSH on physical and simulated robots. The Intelligent Wheelchair Project provides a test bed for these ideas, as well as other interesting ideas which are independent of the SSH domain. Control, topology, learning, attention, vision, and speech are some of the topics which will be covered in this presentation.

December 1st

Prof. Michael Benedikt [email]
Robert Turknett [email]
Center for American Architecture and Design
UT Department of Architecture

The Omega Files: Complexity, Organization, Evolution, and Preference

This talk will report mostly theoretical work out of my forthcoming book A General Theory of Value. Beginning with the work of theoretical biologists Daniel Brooks and E.O. Wiley (based in turn on physicist David Layzer's ideas of cosmogenesis), I will outline a way of thinking about the "proper" growth and balance of complexity, C, and organization, R, that keeps natural systems alive as well as artificial life systems interesting and self-perpetuating. The significant and to-be-optimized variable, I will argue, is what I call "omega," the geometric mean of C and R formulated in Shannon-and-Weaver-like terms.

Rob Turknett will demonstrate TokenTrade, his quasi-ALife economic system based on this theory, live and online. Other sorts of evidence for Omega-optimality will be presented too, from DNA codon-sequence statistics to LIFE and Sugarscape, to operant conditioning schedules, to preference for melodies, to human information processing limits...which the lecture itself will hope to breach.

Past Schedules

Spring 2000

fai (fA) n. Archaic. [Middle English]: Faith.