biography

Biography

My work

I grew up in the Midwest, and went through useful gifted and honors programs as a result of the excellent public education systems I encountered. I studied signal processing and neurobiology in my later education, and overall have been interested in what intelligence is, and how it operates.

A study of intelligent systems can be made on three useful levels, as David Marr instructs: 1) what is the physical implementation of the system? (e.g., neurons, semiconductors, &c.), 2) what is the algorithm, or mathematical description, of the system's operation? (e.g., adaptive function-fitting, optimization, search, &c.), and finally, 3) what is the computational goal or purpose of the system? (e.g., control of the environment, competitiveness in ecological or market contests, survival, and so on).

Artificial intelligence, as a discipline, has misunderstood the nature of the problem to be solved for decades, with debilitating consequences for effective research into the nature of intelligence. The problem with historical cognitive science and AI, as I see it, comes from the assumption that language is a sufficient underlying representation for the kinds of cognitive function we see in humans and many other animals. Because humans use language to communicate, we tend to make the mistake of thinking that cognition itself is, primarily and only, rooted in the structure of language. This was expressed in the physical symbol system hypothesis by researchers Newell and Simon (1963), a theory proposing that symbolic processing is both necessary and sufficient for intelligent behavior.

But imagine that you wanted to make a robot that could replicate the amazing feats of sensorimotor control that Michael Jordan performed over and over again on the basketball court. How much of the functioning of the robot's nervous system should be devoted to linguistic-style symbol processing? Can a nervous system get sensory input, transform it into 'words', and use those word-tokens to control coordinated motor behaviors? Does 'object recognition' alone govern how to spin body and limbs past intelligent and embodied opponents, and slam a basketball into a hoop? It seems ridiculous to think so.

So then, what is needed for intelligent behavior as we experience it?

Our senses give us limited and partial information about what's going on around us in the moment. It would seem silly to try to directly control the activity of our finger muscles using pixel intensities measured on the retina. Pixel intensities, by themselves, don't tell an explicit story about what motor behaviors need to know to do their job. It becomes necessary to transform sensory signals into an alternate set of descriptors that are more informative for behavior. How is the current sensory input indicative of how the nearby environment is likely to behave in the future, either by itself, and/or in response to a range of my possible behaviors? How do I attach value (preference, aversion, cost, benefit, utility, and so on) to predicted sensory results of the possible range of my behaviors? How does the environment tend to appear over time, including in its response to my actions within that environment?

Many of these sensorimotor control problems are continuous-valued and ordinal in their organization, in multiple dimensions (in humans, there are millions of dimensions of sensor inputs and motor outputs). A truly intelligent embodied system must make as accurate a model as possible of the experienced tendencies of the sensorimotor environment, and use that world model to plan and execute behaviors that are likely to make conditions more favorable for it.

This is my research program.

Media

Non-fiction books I've read this year:

Freakonomics: A Rogue Scientist Explores the Hidden Side of Everything - Steven D. Levitt and Stephen J. Dubner (2006)

The Fabric of the Cosmos - Brian Greene (2005)

Collapse: How Societies Choose to Fail or Succeed - Jared Diamond (2005)

The Botany of Desire: A Plant's-Eye View of the World - Michael Pollan (2002)

The End of Faith: Religion, Terror and the Future of Reason - Sam Harris (2005)

Sperm Wars: Infidelity, Sexual Conflict, and Other Bedroom Battles - Robin Baker (1995)

Fiction books I've read this year:

The Company: A Novel of the CIA - Robert Littell (2002)

Woken Furies - Richard K. Morgan (2007)