Special Reading #1 - The Chinese Room

Comments:
Comment 1
Comment 2

References:
Title: Minds, Brains, and Programs.
Author: John Searle
Venue: Behavioral and Brain Sciences, 1980

Summary:
In this paper, John Searle makes the statement that the traditional method of making programs can never make a true AI that understands like humans do. He accomplishes this through a simple thought experiment, wherein an individual is given a book that has a series of directions of what to do when someone slips Chinese characters through a hole in the wall. The directions would look something like this image: (citation below)

The person inside the room reads the directions, looks at the slips of paper dropped in the door, and writes the supplied characters on sheets of paper and slides them another hole.

Now, a Chinese person outside the room, assuming the program in the books was written well enough, might think that this room understands Chinese quite well. However, the person inside has no understanding of what the Chinese characters mean and, no matter how many times this task is performed, will never know what they mean. This means that this construct, and by extension a program following the same principles, cannot truly be intelligent. He then posits that the only way to make a true AI would be to emulate the processes in the human mind, which we still do not understand.

Image courtesy of: http://www.mind.ilstu.edu/curriculum/searle_chinese_room/searle_chinese_room.php

Discussion:
I find this argument interesting because it destroyed my preconceptions of what AI could be. Especially in movies today, AIs are shown to have levels of intelligence rivaling humans, and this paper showed me convincingly that these portrayals are false. The argument that finally got to me was that even though a computer can do a great simulation of an explosion, you do not expect to get hit by shrapnel, so why would a simulation of intelligence create a true intelligence? In fact, as far as the argument is concerned, I can't think of a fault in this argument that does not discredit human intelligence to make it equivalent to machine intelligence.

Possible future work that could come out of this paper could include research into real brains to determine this quality that AIs are missing or research into AI programming methods to get around this limit.