FAI meets every other Friday to discuss current topics in artificial intelligence.

After the formal talk has completed, we continue our conversation at the Crown & Anchor.

All are welcome to attend.

Please send questions or comments to: Harold Chaput (chaput@cs.utexas.edu) or Tal Tversky (tal@cs.utexas.edu)

Would you like to receive announcements about FAI? Click here.

 Talks for Spring 2000

January 28th

Harold Chaput & Tal Tversky
UT Dept. of Computer Sciences

Alternate Essences of Intelligence

Abstract from the paper: We present a novel methodology for building human-like artificially intelligent systems. We take as a model the only existing systems which are universally accepted as intelligent: humans. We emphasize building intelligent systems which are not masters of a single domain, but, like humans, are adept at performing a variety of complex tasks in the real world. Using evidence from cognitive science and neuroscience, we suggest four alternative essences of intelligence to those held by classical AI. These are the parallel themes of development, social interaction, embodiment, and integration. Following a methodology based on these themes, we have built a physical humanoid robot. In this paper we present our methodology and the insights it affords for facilitating learning, simplifying the computation underlying rich behavior, and building systems that can scale to more complex tasks in more challenging environments.

Paper: Brooks, R.A., C. Breazeal (Ferrell), R. Irie, C. Kemp, M. Marjanovic, B. Scassellat and M. Williamson, "Alternate Essences of Intelligence," to appear AAAI-98. [Available for download in Compressed Postscript and PDF.]

February 11th

Esra Erdem
UT Dept. of Computer Sciences

Missionaries and Cannibals in the Causal Calculator

Abstract: A knowledge representation formalism is "elaboration tolerant" to the extent that it is convenient to modify a set of facts expressed in the formalism to take into account new phenomena or changed circumstances. John McCarthy illustrated this idea by defining 19 elaborations of the Missionaries and Cannibals Problem. We argue that, to a certain degree, the goal of elaboration tolerance is met by the input language of Norman McCain's Causal Calculator. We present formal descriptions of the basic Missionaries and Cannibals Problem and of ten of McCarthy's enhancements as input files accepted by the Causal Calculator. Each enhancement is obtained from the basic formulation by the simplest kind of elaboration--adding postulates.

Paper: Lifschitz, Vladimir. "Missionaries and cannibals in the causal calculator, " to appear in Principles of Knowledge Representation and Reasoning: Proceedings of the Seventh International Conference, 2000. [Available for download in Postscript]

February 25th

Prof. Les Cohen
Childrens' Research Laboratory
UT Dept. of Psychology

How Infants and Experts are Alike

Abstract: I plan to summarize a considerable amount of empirical research from our laboratory and from other laboratories that all lead to the same conclusion about the development of infant perception and cognition. The research topics range from something as "simple" as infants' perception of a 45-degree angle to something as complicated as their understanding of events involving a sequence of causal chains. In all of these cases development seems to follow the same set of constructive, information processing principles that resemble a form of Hebbian learning. To the extent time allows, I shall also show how these domain general principles can be used to explain aspects of infant cognition that some argue can only be based upon domain specific, innate modules. Finally I will propose that these information processing principles are not peculiar to development, but may in fact be the same principles that apply to anyone learning to become proficient or expert on some new task. Hopefully, in the question and answer period we also can discuss how computational modeling can be used to make these information processing principles more explicit.

March 10th

Kenneth Stanley
UT Dept. of Computer Sciences

Neuro-Evolution: Can artificial brains be evolved?

Abstract: It is possible that abstractions of certain neural mechanisms can be combined to create neural networks analogous to animal or even human minds. If such abstractions exist, then processes based on natural selection can guide a search through various configurations of neural mechanisms. In my talk, I will discuss what I think these abstractions might be, how these genetic and neural building blocks can be programmed into a computer, and how evolution can find a neural structure that supports intelligent thought.

Paper: Köhn, Phillip (1996). "Genetic Encoding Strategies for Neural Networks" [Available for download in Postscript.]

March 24th

Prof. Bill Geisler
Center for Vision and Image Sciences
UT Dept. of Psychology

Perceptual Grouping and the Statistics of Natural Images

Abstract: The human brain manages to correctly interpret almost every visual image it receives from the environment. Underlying this ability are a number of sophisticated perceptual grouping mechanisms. These include contour grouping mechanisms that are able to appropriately link locally-detected edge elements into global contours. Although a general view of how the brain achieves effective contour grouping has emerged, there have been a number of different specific proposals and few successes at quantitatively predicting performance. These previous proposals have been developed largely by intuition and computational trial and error. A more principled approach is to begin with an examination of the statistical properties of contours that exist in natural images, because it is these statistics that drive the evolution and development of the grouping mechanisms. I will describe our measurements of both simple and Bayesian edge co-occurrence statistics in natural images, as well as human performance for detecting natural-shaped contours in complex backgrounds. I will show that contour detection performance is quantitatively predicted by a local grouping rule derived directly from the co-occurrence statistics, in combination with a very simple integration rule (a transitivity rule) that links the locally grouped contour elements into longer contours. Finally, I will described how this general approach, based upon natural-image statistics, can be extended to the study of other perceptual grouping phenomena.

April 7th

Prof. Art Markman
Similarity and Cognition Lab
UT Department of Psychology

In Defense of Representation

Abstract: The computational paradigm, which has dominated psychology and artifical intelligence since the cognitive revolution, has been a source of intense debate. Recently, several cognitive scientists have argued against this paradigm, not by objecting to computation, but rather by objecting to the notion of representation. Our analysis of these objections reveals that it is not the notion of representation per se that is causing the problem, but rather specific properties of representations as they are used in various psychological theories. Our analysis suggests that all theorists accept the idea that cognitive processing involves internal information-carrying states that mediate cognitive processing. These mediating states are a superordinate category of representations. We discuss five properties that can be added to mediating states and examine their importance in various cognitive models. Finally, three methodological lessons are drawn from our analysis and discussion.

Paper: Markman, A.B., & Dietrich, E. (in press). "In defense of representation." Cognitive Psychology.

April 21st

Prof. Mary Dee Harris
Speech Recognition Group
UT Department of Computer Sciences

A Spoken Natural Language Interface to the Intelligent Wheelchair

Abstract: Building an Intelligent Wheelchair in a research environment entails research topics in vision, reasoning, and spatial representation. From a practical point of view, though, an Intelligent Wheelchair should give a disabled individual more freedom and flexibility of movement and therefore must include a workable user interface. While a toggle or joystick will suffice for manipulating the chair in many situations, real intelligence in this robotic device should include spoken language capability for those users who can tell the chair their destination and provide some direction while enroute. The Speech Recognition component of the Intelligent Wheelchair attempts to provide such capability.

Using a headset microphone for input and sophisticated speech recognition software installed on a PC, the Wheelchair accepts verbal commands in natural language (English) and execute such commands to transport the user as requested. Initially the commands are quite simple: "Take me to the elevator" or "Go to Painter Hall," requiring fairly direct actions by the Wheelchair. Eventually more complex sequences of interaction are envisioned. At such time, the Wheelchair could understand that a statement, "I forgot my bookbag," means that the user needs to return to the previous location to get that bag. Such interactions require the system to understand the user's intentions as well as the ability to identify the words spoken.

For the simplest statements, the software acquired, the Truetalk GrapHvite System (from Entropic which has been bought out by Microsoft) identifies the words and their grammatical sequence based on a pre-defined vocabulary and a lattice of possible statements, producing a text transcription of the words spoken. Our system then employs the Franklin Sentence Parsing Program to produce a syntactic parse of the text transcription. Finally, the system semantically parses the result to produce a conceptual meaning representation that the wheelchair will use to execute the required actions.

May 5th

Prof. Sandy Stone
Advanced Communication Technologies Lab
UT Department of Communications

A Perceptron Reveals the Face of Sex

Abstract: Of what are minds made? Experience has shown the answer to, unfortunately, be a big no. It makes little sense to force the burden of communication on a single medium, restricted to just one of spoken or written text, gestures, diagrams, or graphical animation, when in many situations information is only communicated effectively through combinations of media. Sensor abuse is a serious and debilitating condition. We outline plausible future work along these lines which can lead to vastly more ambitious systems.

The power of sexuality is discussed and it is shown that this power does not mainly stem from the 'combination' of 'good properties' of the mates (building block hypothesis) but rather from genetic repair diminishing the influence of harmful mutations. The agent should be capable of operating with minimal intervention from humans, and it should be able to cope with uncertainty. Yet biology lacks a theory of organization. Evolution to the edge of chaos is demonstrated in an imitation game. Continued growth of the global norn population, to figures measured in tens of millions, is quite likely.

Does life essentially concern a living entity's material composition or its form? This particular subject was lucky enough to pull himself out of his pitiful condition, but others are not so lucky. My arguments for this are very weak but let's pretend, for brevity, that things are clearer than they are. These justifications may be examined by the user to gain insight into the operation of the set of rules as they apply to a problem.

Actual computer output is included.

 

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