Letter from Hans Moravec

February 21, 2001

Hans Moravec Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213

Originally published March 25, 1999 at The New York Review of Books. Published on KurzweilAI.net February 22, 2001.

To the Editor:

In the April 8 NYRB review of Raymond Kurzweil’s new book, John Searle once again trots out his hoary “Chinese Room” argument. So doing, he illuminates a chasm between certain intuitions in traditional western Philosophy of Mind and conflicting understandings emerging from the new Sciences of Mind.

Searle’s argument imagines a human who blindly follows cleverly contrived rote rules to conduct an intelligent conversation without actually understanding a word of it. To Searle the scenario illustrates machine that exhibits understanding without actually having it. To computer scientists the argument merely shows Searle is looking for understanding in the wrong places. It would take a human maybe 50,000 years of rote work and billions of scratch notes to generate each second of genuinely intelligent conversation by this means, working as a cog in a vast paper machine. The understanding the machine exhibits would obviously not be encoded in the usual places in the human’s brain, as Searle would have it, but rather in the changing pattern of symbols in that paper mountain.

Searle seemingly cannot accept that real meaning can exist in mere patterns. But such attributions are essential to computer scientists and mathematicians, who daily work with mappings between different physical and symbolic structures. One day a computer memory pattern means a number, another it is a string of text or a snippet of sound or a patch of picture. When running a weather simulation it may be a pressure or a humidity, and in a robot program it may be a belief, a goal, a feeling or a state of alertness. Cognitive biologists, too, think this way as they accumulate evidence that sensations, feelings, beliefs, thoughts and other elements of consciousness are encoded as distributed patterns of activity in the nervous system. Scientifically-oriented philosophers like Daniel Dennett have built plausible theories of consciousness on the approach.

Searle is partway there in his discussion of extrinsic and intrinsic qualities, but fails to take a few additional steps that would make the situation much clearer, but reverse his conclusion. It is true that any machine can be viewed in a “mechanical” way, in terms of the interaction of its component parts. But also, as Alan Turing proposed and Searle acknowledges, a machine able to conduct an insightful conversation, or otherwise interact in a genuinely humanlike fashion, can usefully be viewed in a “psychological” way, wherein an observer attributes thoughts, feelings, understanding and consciousness. Searle claims such attributions to a machine are merely extrinsic, and not also intrinsic as in human beings, and suggests idiosyncratically that intrinsic feelings exude in some mysterious and undefined way from the unique physical substance of human brains.

Consider an alternative explanation for intrinsic experience. Among the psychological attributes we extrinsically attribute to people is the ability to make attributions. But with the ability to make attributions, an entity can attribute beliefs, feelings and consciousness to itself, independent of outside observers’ attributions! Self-attribution is the crowning flourish gives properly constituted cognitive mechanisms, biological or electronic, an intrinsic life in their own mind’s eyes. So abstract a cause for intrinsic experience may be unpalatable to classically materialist thinkers like Searle, but it feels quite natural to computer scientists. It is also supported by biological observations linking particular patterns of brain activity with subjective mental states, and is a part of Dennett’s and others’ theories of consciousness.