Can a computer ever be conscious? To be clear, when we talk of a computer here, we are referring to a something analogous to a Turing Machine (if you are not familiar with Turing Machines, please allow me to pitch my short introduction to them here). That is to say, a machine that obeys clearly defined instructions. The list of instructions (algorithm) can be arbitrarily complicated and exhaustive, but will there still be something that a human could do that a machine could not? If not, then would the machine then said to be conscious?
Roger Penrose presented 4 possible views on this:
A) All thinking is computation. Conscious awareness is nothing more than running a particular computation. The physical entity conducting the computation is of secondary importance. Any machine thus executing the consciousness computation is therefore conscious and self-aware. This is called the strong AI approach.
B) Consciousness comes from physical actions in the brain. While these actions may be simulated in a computer, the simulation itself is not enough to create computation.
C) Consciousness comes from physical actions in the brain. But it is impossible even in principle to be able completely simulate these on a computer. In principle at least, it will always be possible to distinguish between the outcome of a human intelligence and a computer.
D) Consciousness cannot be explained physically, computationally or scientifically.
Of course, there are positions in between. Note for example, the idea of mind-body distinction is ignored in this classification. You could believe in the idea of a soul but still agree on the value of a scientific approach to understanding consciousness. But for the narrow confines of answering the question of whether or not AI consciousness is possible, most people should be able to adjust to the arguments evolved from one of these positions.
Let us start with A. Consider the fact that even for ourselves, the matter that makes our body and brain is replenished constantly. What persists is the pattern as stored in DNA and implemented through proteins. Thus when we speak of ourselves, we do not refer to the collection of matter that makes us but rather the enduring pattern. Human beings are thus patterns of information which according to the proponents of A can be simulated. At one extreme of this logic is the idea that the whole universe is a computation (or according to some, a simulation).
This is a highly operational view of science. If it looks like a duck, walks like a duck and quacks like a duck, then it probably is a duck. So, if a computer looks and behaves as if it is intelligent and conscious, we should consider it be conscious. There is after all, no reason to doubt it. This is the view behind the Turing Test for intelligence.
A person sits on a computer terminal and types questions to which he or she is given answers on the terminal. If after some time, the person is unable to reliably distinguish whether the responses are coming from a person or a machine, the machine can be said to be intelligent.
B argues against this and says that even if the machine’s behavior is indistinguishable from a conscious entity, it does not mean it is conscious. Consider Searle’s Chinese Room. John Searle proposed the idea of a non-Chinese speaker in a room. A Chinese speaker asks questions to this person by writing Chinese characters on a strip of paper and feeding it inside through a slot. The person inside responds with another strip of paper with Chinese characters.
But the person inside does not know Chinese. He has a big directory with all possible Chinese symbols anybody outside could ask and their corresponding answers in Chinese. He just looks up the symbols he received and copies out the answers as given in the book without understanding the meaning. But to the person outside, it seems as if the person inside the room understands Chinese.
Searle’s Chinese Room
Thus even if a computer behaves exactly as if it were conscious, it does not follow that it is conscious as it could just be following instructions without any comprehension of underlying meaning.
C is interesting because it does not directly deny A. It just says that it is impossible to ever completely simulate a living consciousness on a computer. Therefore the question of whether a computer that behaves exactly as a human would be considered conscious is irrelevant because it is in principle impossible to make a machine indistinguishable from a living consciousness.
Science fiction fans may remember Philip K Dick’s Do androids dream of electric sheep? (on which the film Blade Runner was based). There, androids were indistinguishable from humans except that they lacked empathy. Indeed the way to catch a robot was to give them a list of moral questions and gauge their responses.
C allows for the idea of an as yet undiscovered branch of science which could explain consciousness and would potentially be able to distinguish between a person and a machine of arbitrary complexity.
As for D, if someone believes that a scientific view is unhelpful when dealing with the subject of consciousness, then there is little more to be said about it. I guess if you have read this far, it is unlikely you hold that view.
Roger Penrose (1994), Shadows of the Mind, Oxford University Press
Philip K. Dick (1968), Do Androids Dream of Electric Sheep?, Doubleday