[Back to PhARG Research][PhARG Home]

Objective and Subjective Neural Representation in Simulated Autonomous Agents

Pete Mandik

In my original proposal for the McDonnell project, I embarked on a neurophilosophical investigation of two broad categories of mental representation. On the one hand are what I call "objective representations": representations that an organism uses to represent aspects of its environment that would, in some sense, "be there anyways"--aspects that don't themselves reflect anything about the organism. On the other hand are what I call "subjective representations": representations used by the organism to represent stuff that depends on the organism, stuff that "isn't there anyways". For some quick examples of the distinction, consider the following.
1. the objective representation of something as being 190 degrees Fahrenheit as opposed to the subjective representation of something as being "too hot for me"
2. the objective representation of something as being made of chlorine as opposed to the subjective representation of something as being inedible, or poisonous
3. the objective representation of something as being 300 miles south of the equator as opposed to the subjective representation of something as being in the upper right quadrant of my visual field.
For further examples we may turn to cognitive neuroscience. Especially pertinent is research which concerns the ways in which spatial locations are represented in egocentric (self-centered) and allocentric (other-centered) frames of reference (neural representations postulated to be localized in parietal cortex and the hippocampus, respectively).

I have proposed to undertake an extended examination of cognitive neuroscientific accounts of the differences between egocentric and allocentric spatial representations and relate these accounts to philosophical accounts of the nature of representation.

When I originally embarked on my McDonnell project, I envisioned the ultimate philosophical payoff to concern issues in the theory of consciousness. I proposed a teleo-informational account of the egocentric/allocentric distinction, and employed it in account of subjectivity that provides a novel representational-physicalist response to Jackson's (1982)"Knowledge Argument" against physicalism. My preliminary work in this area is summarized in Mandik (2001)

Since embarking on the project, the focus of my work has shifted slightly toward more foundational issues in the theory of representation, (though I have not abandoned my interested in consciousness-related topics).

Among the kinds of question I am most eager to answer would be: on the assumption of a causal/informational theory of representation, what room could there be for a distinction between an egocentric representation of a spatial location and an allocentric representation of one and the same location? On the assumption that there is a valid distinction between egocentric and allocentric representations of space, what philosophical theory of representation is best able to accommodate it: information theories, isomorphism theories, functional role theories, or some other?

I have been thinking hard recently about how such questions could be addressed by striving to literally implement the various theoretical proposals. How would one go about building a system that could satisfy the threefold conjunction of (1) working in accordance with principles of neurobiological plausibility and (2) conforming to some philosophical account of how brain states represent environmental features (either egocentrically or allocentrically) and (3) actually capable of learning to run a maze? Suppose one were to build a robot that was "wired up" with a plausibly neural-like network and was capable of behaving interestingly with regard to spatial routes and locations (e.g. learning to run a maze). Could one then retrospectively "look under the hood" and tell whether (1) the robot was representing spatial locations egocentrically or allocentrically or (2) the robot was representing locations in accordance with a resemblance theory of representation or in accordance with an asymmetric dependence theory? Could one build a neural-network driven robot that implemented, say, Fodor's asymmetric dependence theory--e.g. had internal states that were asymmetrically dependent on spatial locations--that was actually able to navigate a maze in virtue of having such internal states?

I want to build and experiment with robots. Robots being relatively expensive, I propose to make do with software simulated robots: virtual autonomous agents. Virtual autonomous agents have the further advantage of being amenable to sending and dealing with over the internet. This aspect of my project integrates nicely with the educational initiatives of McDonnell project to start a web-based neurophilosophy course.

Among the philosophical accounts of the neural basis of representation that I would like to attempt to implement in virtual autonomous agents are accounts currently being developed by project members Chris Eliasmith, Rick Grush, and Jon Opie. Eliasmith is developing his statistical dependence account of representation. Opie is developing his isomorphism based account of representation. Grush is developing his motor-emulation based account of representation. In accordance with my original proposal, I continue to develop an account of objective (allocentric) and subjective (egocentric) neural representation. I propose to collaborate with Eliasmith, Grush, and Opie to see how each of their accounts allows for (1) a neuroscientifically plausible account of the distinction between egocentric and allocentric representations of space and (2) the implementation of this distinction in neuroscientifically plausible virtual autonomous agents.

Jackson, F. (1982), "Epiphenomenal Qualia", Philosophical Quarterly 32, pp. 127-36.

Mandik, P. 2001. Points of view from the brain's eye view: subjectivity and neural representation. Philosophy and the Neurosciences: A Reader. (Eds.) William Bechtel, Pete Mandik, Jennifer Mundale, and Robert Stufflebeam, Oxford: Basil Blackwell.