Behavior vs. Function:
Behaviorism and Functionalism as Methods of Artificial Intelligence
By Nathan E. Pralle
Author's Note: This is a paper I wrote for my independent study in philosophy of mind and artificial intelligence under Professor Gereon Kopf while in college at Luther College
The mere idea of intelligent machines, artificial intelligence, or computers that react like humans often scares, intrigues, and amazes people.
However, with increased technology, advances in studies of the mind and mental processes, and a general increase in the awareness that humans may be able to replicate themselves in non-organic fashions, one can envision a time when such a fantasy will manifest itself in the world and begin to become a part of normal society.
In working on these 'machine marvels,' there are many philosophically inclined aspects to be considered.
One of the largest and often most frequently discussed is the concept of consciousness and the problems of the representation of a human mind within the constraints of a mechanized brain.
Problems exist once one fathoms the intricate constructs that must be formed to contain something as complicated as the human mind in a machine.
However, two separate philosophies can present the required viewpoint from which to address these issues and arrive at a holistic solution to the problem of minds in machines.
This problem is not a new one and has been considered ever since the first philosophers considered the mind and the various ways of trying to understand its complicated construction.
Previously the mind remained mysterious, and under Descartes it seemed that it would remain as such with his theory of dualism.
Dualism proclaimed that mind and body were separate entities that acted on each other in some unknown manner, but that mind was impossible to analyze because it was non-material.
"Indeed, one might fairly characterize Dualism as being virtually an official announcement that the mind/body problem is forever insoluble: O magnum mysterium! The mind unquestionably interacts causally with the body, but we could not in principle even begin to discover how." (Lycan 2).
Under this philosophy, the pursuit to place mind into a machine was ludicrous and wasteful.
One could not capture that which is not physical in something so bound to the material of reality.
Theories and philosophies of the mind progressed, but one essential property that all lacked was the ability to describe the mind in terms which could be duplicated without reconstructing it in a similar brain construct, nerve by nerve.
The brain, being an extremely complex organ, is not the practical model from which to build a computer system that would manifest a human mind.
The practicality is nearly zero because of the difference in mediums; construction of a mind in organic terms is best done with organic parts and organic design, while construction in a computer system is best done with computer parts, in a computer-styled design.
The philosophy must be one that only has computationally representable aspects of the mind.
The idea of "computationally representable" must be briefly explained so as to distinguish the extent to which this phrase is applicable towards mental phenomenon.
Computer technology is such that one can represent or encode information into various series of binary digits (zeros and ones) that can then be stored, manipulated, and transformed into other series.
These series can be understood to represent a larger, more abstract concept, such as an actual digit, character, word, sentence, color, intensity, or any number of other items.
Each of these can then be combined to form higher, more abstract concepts.
By using the binary system as the base and the simplest blocks, one can construct through this process of abstraction and addition any needed concept in a computational model.
Thus, any system of philosophy of the mind that is presented as a solution must allow for its constructs to be represented in formations of simple binary representation or some abstracted system of symbols that is derived from the simple binary base.
Un-computationally representable concepts exist in the mind-body problem of Cartesian dualism, as well as theories such as reductive materialism, which states that each process of the mind is reducible to individual brain processes.
These ideas are valid ways to think of the mind, but are next to useless when attempting to formulate a computer-mind strategy.
The first of two theories, functionalism is a theory that can be represented well in a computer context.
If an artificially constructed being were to act and perform in the same manner as a human it might very well be considered to be full representation of a human.
This is the basic tenant of functionalism in that the manner in which the mind works is not the issue, simply the ingestion of input and the production of an output is the key pursuit.
Any input, as long as it generates the correct output, is valid for this system.
"Other types of mental states (sensations, fears, beliefs, and so on) are also defined by their unique causal roles in a complex economy of
internal states mediating sensory inputs and behavioral outputs." (Churchland 36).
In a computer system, input and output are essential parts to the completion of a program and can be easily observed and controlled, thus making the pursuit of replicating a mind a plausible notion.
As a working theory, functionalism is very attractive because it detracts from the exact replication of the brain to a reproduction of the mental states and reactions of the brain.
Although the exact states of the mind which is being replicated are not necessarily occurring in the artificial mind, the results that are perceived are the same as what one would expect to get out of a human and the given input.
This also allows for many different kinds of reproductions based on various materials, not just computer technology.
In this case, even an alien's pain, as Churchland postulates, could be functionally identical to any human pain, and therefore, any other sensation.
"The alien may have an internal state that meets all the conditions for being a pain state, as outlined earlier.
That state, considered from a purely physical point of view, would have a very different makeup from a human pain state, but it could nevertheless be identical to a human pain state from a purely functional point of view.
And so for all of his functional states" (Churchland 36).
This theory is obviously a great boon to the idea that mind can be represented in objects other than human brains, such as computers.
Functionalism, of course, is not the perfect answer to the problem, and there are many arguments against it.
Many argue that functionalism disregards the internal qualia, or sensory experience, of a person.
When one person feels pain, he or she may react to it in whatever manner is appropriate for that particular pain.
Externally, the reaction is all that matters to the functionalist looking to prove that the pain occurred.
However, the part that is missing is the internal aspect, what the mind is actually feeling associated with that pain and whether or not there are other side effects of that pain occurring in the mind (such as remembering a similar pain from a time past.) The rebuttal to this argument is that in the problem of representing the mind, the idea of 'hidden qualia' is of no concern.
Whether or not such qualia exist in an artificial mind is of debate and ultimately, no great consequence, for if the artificial mind is acting exactly like the real one, what purpose does the qualia serve? It would be an argument at least to state that perhaps in order to represent the mind fully, in an artificial state, the mind would have to encompass all of the hidden qualia that humans normally possess and be able to represent that, whether shown or not.
However, since the purpose of the experiment is to represent the mind in a computer system in a functional way, the qualia problem is of little or no concern.
However, functionalism as currently known does possess one shortcoming that is of great consequence when discussing the goal of artificial representation.
The fact that functionalism states that one cannot create exact one-to-one mappings of the inputs of the system to the outputs of the system is a serious flaw considering today's computer technology.
"As he [the functionalist] sees it, the adequate characterization of almost any mental state involves an ineliminable reference to a variety of other mental states with which it is causally connected, and so a reductive definition solely in terms of publicly observable inputs and outputs is quite impossible." (Churchland 36).
With the current systems that are running, distinct inputs produce distinct outputs, and the results expected do not vary from the planned results.
If the output deviates from the expected, then it is a problem of the program that is running, not of the internal makeup of the computer and program acting in a rather random fashion.
Functionalism insists that the combination of internal states in the mind is too complex to expect to derive an exact match for the input to the output.
There are two main ways to tackle this lacking in functionalism.
The first is to change the current computer technology that is being used.
The system that would have to be developed is one that could output a wide range of results based upon one input, and also one that allowed the inner states of the machine to effectively change the 'normal' output into one that is different yet within the parameters of the possible outputs.
The essential problem with such a system is determining at what level the change would occur.
As the level of abstraction drops closer and closer to the binary level, the room and locations for change to occur grows smaller and harder to justify in any meaningful terms.
Since the meaning of the information contained within the system is dependent upon its abstraction from the binary level, the closer one is to the binary level the harder it is to create meaning out of any change that must occur.
Thus the change would necessarily have to occur high in the abstract levels where change would affect some meaning upon the symbols down the chain that represent the larger concept.
Such a computer system is plausible, but using present systems requires that the criteria for which we model the mind be changed.
The theory at hand then is one called behaviorism.
Akin to functionalism, behaviorism looks and acts very similar.
This theory is also concerned, as functionalism is, with the system's reactions to input.
"By 'behavior', the behaviorists mean the publicly observable, measurable, recordable activity of the subjects at issue: bodily movements, noises emitted, temperature changes, chemicals released, interactions with the environment, and so forth." (Churchland 88).
In comparison to functionalism, however, behaviorism insists that a one-to-one mapping can be effected in the system between the input and output, therefore creating an expected reaction to any given situation presented to the 'mind.
'
This, as previously discussed, is the main issue that is needed to be compatible with today's computer technology.
Such systems that could possibly support a behavioristic ideal of mind are already in use, albeit the power that is needed for the implementation of a human mind may still be lacking.
Since experiments with computer technology can be conducted to determine if a certain input is matching a certain output, behavioristic models of human minds happen to be very conducive to such testing.
Therefore it is much more plausible that a system could be developed, tested, and experimented with until it behaved in the 'correct' way.
Behaviorism's naysayers bring several arguments to bear when dealing with it.
First, they argue, to only state the nature of things according to their behavior and reactions is disregarding the internal workings of a system and the functions of internal mechanisms.
"If a restriction in favor of operational definitions were to be followed, therefore, most of theoretical physics would have to be dismissed as meaningless pseudoscience!" (Churchland 90).
Clearly, most philosophers wish for a mind-view that encompasses both aspects of the mind, the behavioral and the internal, which seem to be the antithesis of each other.
In this particular instance, behaviorism again shows a viewpoint that is useful for creating like minds in computer technology without being concerned with the internal workings and details of a system that isn't even of the same material as the human brain!
As with functionalism, behaviorism still denies the understanding of internal qualia and other mental states.
The fact that the internal is being ignored does not include the fact that the internal is not occurring.
In the case of the inverted spectrum argument, the actual events happening within the mind in question are not of consequence simply because the resulting behavior is what is expected.
The inverted spectrum problem states that while two people may be looking at a sheet that is colored green, one of the two is mechanically seeing the sheet as blue, the other as green.
Yet, since both were raised with the notion that the sensation they are receiving due to the green sheet is the color 'green,' there is no difference when you ask either of them what color they perceive.
Hence, behaviorism does not contradict the conditions in the inverted spectrum problem, as William Lycan would declare, but simply works around them as a matter of purpose in the implementation.
(Lycan 4-5)
Behaviorism, in fact, may very well ensure that such internal sensory events happen more like in the human brain.
To make a system that responds with one and only one output as opposed to a varied response range would narrow the way that internal functions must occur to produce the output.
The behavioristic system, although more suited to today's computer systems and their requirements, nonetheless is much more difficult to render in a modeled sense, since much more must be done to maintain consistency in the system.
Thus, qualia and other internal sensory events may have a chance at existing.
Whether or not these are of any use to the outside world in dealing with the particular artificial mind is a matter of a different debate.
Behaviorism then, as opposed to functionalism, seems to be the method to pursue in developing artificial minds based upon human minds.
But are these theories enough? This question is valid because although behaviorism and functionalism give a rose-colored viewpoint from which to approach the idea of artificial representation of the mind, their lacking is in the implementation and construction problems associated with making the actual machine mind and containing the whole of the human consciousness within it.
The details of actually creating the constructs and the hardware to handle the human mind is less than sufficient at the present time, regardless of the external philosophical system from which one takes a viewpoint.
Thus, much more work needs to be done in the area before the realization of human minds in computers can be manifested.
However, the ideas of behaviorism and functionalism give a solid foundation from which to build a mind-view and proceed to deal with the more intricate workings of the human mind.
The very idea of representing a human mind in computer technology may seem to be far-fetched and theoretically difficult, if not impossible, but by using the theories of behaviorism and functionalism one can conceivably construct a viewpoint of the mind that makes it easier to attempt this implementation.
Using functionalism as the basis for one's understanding and behaviorism as the implementation scheme for actually representing the mind, one can get the best of two worlds which are both geared towards the common goal of creating an artificially intelligent machine that responds to events like a human.
Perhaps in this way one can perceive of a time when machines will operate on a similar level with humans and not be the cause of nightmares.
Works Cited
Churchland, Paul M. Matter and Consciousness. Cambridge; The MIT Press. 1997. P. 36-95.
Lycan, William G. Consciousness. Cambridge; The MIT Press. 1995. P. 2-7.
This page and all content (C)2002-2004 Nathan E. Pralle.
www.nathanpralle.com
Unauthorized distribution is prohibited except with written permission of the author.