Why the current conception of AI is improbable.

5 May 2008

Why the current conception of AI is improbable.

Artificial Intelligence… what does it mean? The most available conception of AI is that of the smart robots we see in film and TV, pieces of machinery that think and act as if they had human minds. This is the ultimate goal of the proponents of artificial intelligence, to prove that a self-aware intelligence can be made using only computers, and it is this that objectors say is impossible, for various reasons. However, it is important to notice a failure of both sides to consider a matter of definition: what is truly meant by “artificial intelligence?” The easiest explanation is a machine that is conscious, but what does this consciousness constitute? Self-awareness is without a doubt the sole unifying concept of artificial intelligence, but this is conceived as the final step culminating in the birth of AI, and all other aspects of that consciousness would have to precede its awakening. There is, then, a need to understand what the aspects of intelligence are and their origins in the human mind, since it is human consciousness and intelligence that would be the model for artificial intelligence. So once it is established that the concerns facing the creation of a generic artificial intelligence are addressed, we must then address whether human intelligence can be translated into artificial intelligence. Referring to two influential points of view on the issue, Daniel Dennett’s “Consciousness Imagined” and John Searle’s “Can Computers Think?”, these two issues will be addressed: is artificial intelligence possible, and is the replication of human intelligence possible within a machine?

John Searle says AI is impossible because of a clear problem with the specifics of intelligence and the manipulation of symbols. His basis for this is that, “There is more to having a mind than having formal or syntactical processes. Our internal mental states, by definition, have certain sorts of contents” (Searle, 31), while computers, because of their solely formal/syntactical nature, do not. This is a flawed argument from the get-go: what constitutes these contents? He states that thoughts are strings of symbols with meaning, lines of syntax with semantics, “…symbols have no meaning; they have no semantic content; they are not about anything” (Searle, 31). But what are the physical manifestations of these semantics? Searle manipulates reality to fit his theories: computers are incapable of mind because they lack semantics, they lack meaning because all that computers are capable of is manipulation of symbols according to programmed rules. However, he applies this cold meaninglessness to the circuits of a computer while totally ignoring that the networks of human neurons have this same meaninglessness. By this association of human intelligence with “meaning,” he is essentially creating a mystical substance of meaning: “If my thoughts are to be about anything, then the strings must have a meaning which makes the thoughts about those things” (Searle, 31). He fails to define what exactly allows for meaning in the human brain, other than an allusion to generic “biology.” He states that intelligence “involves having an interpretation, or a meaning attached to those (perceived) symbols” (Searle, 33), but he completely disregards any explanation as to how or when this meaning exists in the human mind. In essence, his argument is “because computers are not biological, they can not think.” He relies on an apparent, almost mystical superiority of the biological brain, without any understanding as to how that brain actually works. In response, Dennett reiterates a question implied by Searle: “How could a complicated slew of information processing events in a bunch of silicon chips amount to conscious experiences?”, to which he replies: “How could a complicated slew of electrochemical interactions between billions of neurons amount to conscious experiences?” (Dennett, 433). If one truly broke down the functioning of the human nervous system, the events that transpire in a functioning neuron could be replicated in a computer microchip. It appears that Searle has an assumption that there is some fundamental functional difference between a firing circuit and a firing neuron, when in simple reality they are the same: on and off, 1 and 0. Whereas Searle stoops to implying strong AI supporters are closeted dualists, he fails to recognize that he is, with his declaration of the presence of “meaning” somewhere within the biological mind, necessitating a dualist frame of mind. Dennett writes, “…according to the Cartesian view, it takes an immaterial soul to pull of the miracle of understanding” (Dennett, 438), but misses the irony that Searle’s argument is analogous to the Cartesian view (maybe even equivalent to it): it takes a biological system to pull off the miracle of meaning.

Another problem with Searle’s position can then be found almost immediately after the above realization. He first establishes two viewpoints on how mental processes and the mind are most commonly viewed: one, as the results of the behaviors of structures of the brain, and two, as analogous to a program running on a computer. After laying out these two viewpoints, he then does something remarkable: he declares the two mutually incompatible by resorting to an appeal to ridicule. This is, in reality, a faulty assertion. Let us examine why.

The brain is an organ made up of many, many different structures that work together in an incredibly dynamic system to create consciousness; this is consistent with Searle’s “first point of view.” Each structure in itself is functionally interdependent on others for its input and sends outputs to many other structures. There are scores of these structures, which can further be broken down into smaller component structures; the brain has separate lobes, each has separate structures, which in turn have their own basic pieces. It is not unreasonable to think that if each structure were thoroughly and completely studied and cataloged, then a sufficiently complex structure of networked computer-structures could at least imitate the brain, and thus consciousness. Thus, if, instead of assuming that the two points of view are incompatible, we combine them, we come to see that consciousness is actually quite possible within a computer system.

Now, would this consciousness be comparable to human consciousness; to this the answer must be “Most likely not.” How does this make any sense at all? In fact, it is quite simple: the processes that achieve intelligence vary to an almost incredible extent. Immense variation occurs between individuals, such that while there are standards of how to measure intelligence and thought, no two individuals are quite the same in how they perceive, process, and respond to the world. This is significant because all of us have a brain. Just within humans, variation occurs even though we all have essentially the same structure in our heads, and point blank, circuit boards and microchips are not neurons and neurotransmitters. They can be made to imitate the processes, but they will never actually be able to duplicate human consciousness. It is almost inconceivable that an entirely different structure, in a completely different mode of organization, made of completely different materials will result in exactly the same phenomena, never mind that it will manifest in human consciousness; it should dare be said that it would manifest in mechanical consciousness, but not just for the above reasons, either.

Let us consider some factors in human consciousness. Maintenance of human consciousness is a complicated process, not only involving energy requirements but most importantly sleep, and specifically REM sleep. Conceivably, the mechanisms that make for REM sleep could be replicated somehow within a computer system, but the question is, why? Why would a machine need food nourishment or sleep? The answer is, it wouldn’t, and as such the experience of the mechanical consciousness would be significantly different than that of the human consciousness; different needs would lead to different consciousness.

The development of a consciousness shapes that consciousness. We as humans go through a complicated and lengthy process spanning approximately twenty years before our consciousness and intelligence are fully developed; an artificial intelligence would not experience this developmental stage. What would the effects of this be? One can only speculate, but considering how much of our “selves” is developed particularly during childhood, it only stands to reason that the consciousness of a machine would be quite different as a result.

Emotion, it should be noted, is another aspect of the human condition that can not be ignored, particularly the physiological effects of emotion. A machine will not experience the physical symptoms of stress, embarrassment, or love, but a human feels these constantly. This is neither beneficial nor harmful, simply different; this difference in experience will, again, create a crucially different manifestation of consciousness.

These particular experiences would, at minimum, differ significantly in machine or computer form, simply by nature of the difference between man and machine. These differences in experience would create a wholly new consciousness, that, while based off a human model, would necessarily differ. Thus, Searle’s skepticism as to the possibility of a computer duplicating a human consciousness is well-placed, though incorrectly justified.