Can a Machine Be Conscious?
In the second presentation from the series, Thompson explores concepts identified with Alan Turing that consider human beings as material machines. Thompson suggests concerns arise when analyzing these issues in light of higher order constructs such as intelligence and consciousness.
TRANSCRIPT: Can a Machine Be Conscious? Vaisnava Ministries: What is Life? 2 – c. 1977-78 / (104)
Vaisnava Ministries presents “What is Life?” – a series of slideshows discussing modern and ancient views on the nature and origin of life and their implications for human society. In the second slide show in this series, we will discuss the question, “Can a machine be conscious?”
Since ancient times, people have speculated about whether or not a machine could be built that has thoughts and feelings comparable to those of a human being. The concept of an intelligent, sentient robot has exerted great fascination over the human mind. Could a structure of metal parts have perceptions and feelings like our own? Could we be enslaved by super-intelligent machines? Are we ourselves just sophisticated machines made of molecular components? If so, could our thought patterns be recreated by a computer and could we become immortal in this way?
The theory that life is reducible to chemical reactions has lent credence to the idea that a sentient robot is indeed possible, but with the development of the modern digital computer in the late 1940’s scientific thought on this subject has gone beyond mere speculation. Now an effort is being made to actually build intelligent machines and some apparent progress in this direction is being made. Here for example, we see the chess champion of Scotland playing against a robot chess player, a computer that can play chess well enough to beat fairly expert human players. In this case, the human being won, but many see it as just a matter of time before a truly intelligent or even super-intelligent computer is produced.
Their basic approach to machine intelligence was first outlined in the 1940’s by Alan Turing, one of the early pioneers in the development of computers. Turing maintained that any describable, measurable human action can also be performed by a machine. He took the ability to produce certain complex behavior as the definition of intelligence and concluded that intelligent machines are indeed possible. According to Turing, “The question, ‘Can machines think?’ I believe to be too meaningless to deserve discussion. Nevertheless, I believe that at the end of the century, the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted." Turing believed that as more and more sophisticated behavior was produced by machines, people would come to accept that machines are as capable of thought as they themselves.
In this slideshow, however, we will focus on an aspect of human nature that should be carefully distinguished from sequences of behavior, however complex – this is consciousness. Consider a man looking at an object such as a thermometer. What does it mean to say that he sees the thermometer? To be even more explicit, what does it mean to say that he perceives the color of the thermometer, the color red? We should note that an external description of the man's behavior does not say anything about the actual experience of seeing something red. Even a tape recorder hooked up to a photocell can be arranged to say, “I see the color red” in the presence of red light, but we are interested in the actual perception of red and we would not imagine this perception to be actually taking place in the arrangement of tape recorder and photocell. How then can we understand this perception? If a gross description of the man's behavior tells us nothing, and this will be especially true if the man says and does nothing in response to what he sees, then perhaps a closer look at the physical processes of vision will provide the answer.
Here, for example, are Maxwell's equations for electromagnetic radiation. These equations can be used to describe what takes place between the thermometer and the man's eye, but they actually tell us nothing about the perception of red itself. If we go further and examine the eye, we find that an image of the red object is focused on the retina. Going still further, we find in the retina itself that certain photochemical reactions are taking place in the rods and cones – the light receptive cells – and as a result, the red light is giving rise to electrical impulses in certain nerve cells. Yet this still gives us no insight into the perception of red color, and indeed scientists generally do not suppose that perception occurs at this stage.
[5:25]
If we go still further and pursue these electrical impulses into the brain, we will find a highly complex series of electrochemical actions and reactions. These phenomena of the brain are of course poorly understood at present, but even if they could be precisely enumerated, what would this tell us about the actual conscious perception of red? At best, a precise listing of brain states could only describe phenomena that somehow correlate with the perception of red and with the other contents of consciousness, such as thoughts, desires, feelings, and sensations of various types.
The contents of consciousness themselves, however, belong to a different category from descriptions of brain states, yet they must be related to brain states in some way. The question of how they are related is called the mind-body question, and it has given rise to many philosophical theories. We will not discuss these theories in detail, but we note three main alternatives: behaviorism, parallelism, and interactionism. We will discuss these alternatives in the context of modern research on computers and computer intelligence.
First, let us consider the approaches taken by researchers in artificial intelligence to the design of intelligent machines or intelligent computer programs, for all of this work has been carried out using programmable digital computers. This slide depicts a well-known example of such work. A program has been written at MIT that can answer simple questions written in English about arrangements of toy blocks, such as those depicted here. For example, one could type in the question, “Which cube is sitting on the table?” and received the typed response from the computer, “The large green one which supports the red pyramid.” Or one could ask, “Is there a large block behind a pyramid?” and receive the reply, “Yes, three of them – a large red one, a large green cube, and a blue one.”
In order to see what bearing this kind of performance has on the possibility of creating a sentient machine, let us briefly examine how programs of this kind work. Basically, computers function by carrying out a sequence of primitive instructions. Complex functions are broken down into combinations of simple operations that by themselves have little similarity to the complex function being performed.
Here we see an example of a computer program designed to calculate the square root of a number. The program has two basic features, a series of memory cells represented by the four labeled boxes and a list of instructions. To simulate the operation of the computer, simply place a number such as 4 in the box labeled 1. Then carry out the instructions in the list one after the other. As each instruction is carried out, the contents of the memory cells are modified.
For example, at this point we have written a 0 and 2, incremented 2 – that is added 1 to it – written a 0 and 3, and copied 2 into 4. Once the last of these instructions are carried out, the square root of the number in box 1 will be found in box 2. In an actual computer these instructions would be stored in a coded form in the computer's memory and executed automatically by special electronic circuitry built into the machine.
At first glance, it may be hard to see what these rather tedious instructions have to do with the calculation of a square root; actually, this is a typical feature of computer programs. The actual instructions carried out by the computer are normally quite far removed from the operations the computer is ultimately intended to perform. These intended operations are carried out by breaking them down into combinations of simpler operations. These, in turn, are broken down into still simpler operations and if necessary, these are broken down still further until finally the original operation is expressed in terms of a combination of elementary instructions.
[10:03]
For example, the overall goal of our program is to calculate square roots; this is indicated in statement 1 of this slide. This goal can be broken down into successive operations of squaring numbers and making comparisons; this is indicated in statement 2. This operation of squaring, in turn, can be broken down into the multiple calculation of sums; this is indicated in statement 3. Finally, the operation of calculating a sum can be broken down into the elementary operation of repeatedly incrementing a number by 1.
In this way, the operation of the program can be broken down into a hierarchy of levels of abstraction. The highest level represents the actual function that the program is intended to perform. The next level defines this operation in terms of a number of simpler operations and in general, the operations on each level are defined on the next level down. None of these levels actually exist as real entities in the computer, however. We may say that the computer is squaring numbers or computing sums, but actually these are only abstractions existing in our own minds. They are only convenient conceptualizations that we use to analyze the actions of the computer. In the computer itself, only the elementary operations are actually taking place. At any one moment in time, the computer is simply incrementing a number, copying a number from one box to another, and so on.
Our observations about the simple square root program also apply to the much more sophisticated programs that have been written in an effort to simulate human intelligence. Here we see a hierarchical outline of levels of function in a hypothetical intelligent computer. On the top, we see the level of intelligent programs – programs expressing operations that are presumably analogous to those carried out in the human mind. The next level involves very sophisticated mathematical operations in which these mental operations can be expressed. The next two levels refer to intermediate classes of operations, and the fifth level down the level of machine instructions corresponds to elementary instructions of the kind that we saw in the square root program. Each of these levels refers to abstract patterns which do not exist as real entities in the computer itself. In fact, we can see from this chart that even the machine instructions do not exist per se. They are only names for patterns of simpler machine operations.
For example, the adding of binary digits is only a name for a composite of logical operations of conjunction, disjunction, and negation. Here's the pattern of logical operations used to carry out binary addition in a typical computer. These operations, in turn, are also simply names for patterns of electrical operations that are being carried out by the actual physical circuitry of the computer. And if we look further, we can see that these operations resolve into patterns of electron flow through various kinds of circuit elements. This process of hierarchical breakdown might be carried further, but this should suffice. The basic point is that only the operations on the lowest level of the hierarchy can correspond directly to the actual events within the computer itself; all of the higher operations are only patterns and patterns of patterns of these actual events.
What does all of this have to do with consciousness? We have observed that conscious perception and measurable behavior fall into two different categories. Now, a behaviorist may wish to deny the existence of consciousness in this sense or to declare that statements about consciousness are meaningless. Nonetheless, we can directly perceive our own conscious perception and thus we can hardly doubt its existence. Furthermore, there is no need for us to label consciousness as something strictly subjective. Since other persons are similar to ourselves in origin, and not just in behavior, it is reasonable to suppose that they too are conscious. Thus consciousness is an objective feature of reality that lies in a different category than behavior.
[14:51]
If we then regard consciousness as objective, would it be possible for a machine to be conscious? Let us suppose for the sake of argument that it is possible and see where this assumption leads us. Suppose that we have a sentient computer and suppose that this computer operates according to the same principles that we have just discussed. In that case, the computer will have to carry out operations corresponding to the contents of consciousness: thoughts, feelings, perceptions, and so on. However, these operations will not exist as real entities in the computer itself. They will only be names for patterns, patterns made up of further names of patterns of names and so on. The actual physical events in the computer will be flows of electrons in the hardware of the machine.
In this diagram, the orange color refers to the actual reality of the computer, and the terms in the dotted rectangle above this refer to the various levels of abstract operations that have no direct existence. In the diagram, however, we have another colored section corresponding to the actual objective consciousness of our hypothetical sentient computer. This consciousness is also real, but what is its relation with the computer?
We see that the contents of consciousness do not correspond directly with the events occurring in the actual hardware of the machine; rather, they correspond to the highest levels in the hierarchy of abstractions that described the operations of the machine. How is it possible that something real, consciousness, can correspond with something that is not real? The answer is that we must adopt an approach to the mind-body problem, or in this case the mind-computer problem, that is called interactionism. We have rejected behaviorism since it denies the objective existence of consciousness which we are hypothesizing for this computer.
Another approach to the mind-body problem is to suppose that some kind of parallelism exists between consciousness and the physical states of the brain. That is, it is supposed that the contents of consciousness are somehow aspects or direct derivatives of the interactions of matter, and this is also called epiphenomenalism. We see here, however, that a parallelism between consciousness and the physical states of the computer is precisely what we don't have.
As we shall argue, our only alternative is some form of interactionism. We must suppose that consciousness corresponds to some kind of entity distinct from the computer itself which can be associated with higher-order patterns of operations in the computer. Now at this point, it might be argued that after all, we have no reason to suppose that a computer can actually be objectively conscious. Perhaps computers are limited simply to the production of sequences of behavior without any conscious awareness being associated with this. This may well be true. At this point, let us, therefore, switch to the consideration of human consciousness.
In the case of a human being, we know that consciousness is objectively existing. At least you know this to be true of your own consciousness by your direct experience. But in what way, we may ask, is this consciousness related to the physical events occurring in the brain? If we examine current scientific ideas about the functioning of the brain, we find that they are quite comparable with the principles of computer operation that we have just discussed. The physical phenomena directly occurring in the brain have nothing to do directly with thoughts, feelings, and perceptions.
For example, the brain is principally composed of billions of nerve cells called neurons. These cells can act as switching elements and they are linked together to form highly complex networks. We can regard single nerve impulses and their interactions at neural junctions as elementary operations, perhaps remotely comparable to the elementary machine instructions of computers.
Going lower in the hierarchy, we see that the elementary operations of neurons are themselves simply names for highly complex sequences of biochemical reactions. Here, for example, we see some of the biochemistry of a single synapse or nerve cell junction. Going higher, on the other hand, we can see that nerve cells tend to be organized into groups. The operations of these groups might be describable in terms of a higher-order symbolic language, similar to the higher-order languages of computers.
[20:08]
If we try to go still higher in the hierarchy of neuro-organization, we quickly reached the limits of current knowledge. However, most students of the brain presume the complexes of neurons will carry out certain higher-order operations. Patterns of these operations will correspond to still higher operations and finally, it is assumed that there exists an ultimate level of operations that correspond to thoughts, feelings, and ego – the awareness of self-identity. However, just as we saw in the case of the computer, these higher-order operations can have no existence as real entities within the brain. They are only names for abstract patterns and the actual physical events of the brain take place on the level of electrochemical reactions.
In this slide we see how one writer on artificial intelligence has presented this idea. The large 2+2=5 represents the conscious level of human thought, a level at which arithmetic, for example, can only be performed inefficiently. This 2+2=5 is composed of many small correct additions corresponding to the actual physical operations of the brain.
What then is the relation between the contents of consciousness and physical brain states? Just as we saw before, we cannot postulate any kind of direct parallelism or one-to-one correspondence between these two realities. As a consequence, we must suppose that consciousness corresponds to some kind of entity that is distinct from the brain, but that can associate with higher-order abstract patterns of brain states. Do we already know of an example of an association of this kind?
In fact, a familiar example is provided by a person reading a book. When a person reads, say a novel, he becomes conscious of the thoughts and feelings that are involved with the plot of the story, while the plot does not actually exist in the book itself. In the actual book, there is only ink on paper. And this arrangement of ink can be analyzed into a hierarchy of levels – letters, words, sentences, and so on – culminating in the plot of the story. Yet, only the ink and paper are actually present as we can see if we are confronted by a book in a foreign language that we cannot read.
We are proposing then that the objective consciousness of a human being corresponds to or is due to an entity that can read abstract patterns in the operation of the brain, just as a person can read a book. Now it might seem that we are solving the problem of consciousness by postulating a complete person within the larger person whose consciousness we were trying to understand. It might be maintained that this explains nothing and that we encounter an infinite regress when we try to explain that inner person's consciousness.
In answer to these objections, we reply that we are indeed postulating the existence of a complete sentient entity that is conscious and that is capable of interacting with the brain on a high level of complexity. However, we propose also that this entity cannot be broken down further by reductionistic analysis. What we are doing is introducing conscious life as a distinct irreducible principle that cannot be explained in terms of matter. Life and matter both exist as real energies or substances. They both obey their own laws and they are capable of interaction with one another. If we are to avoid an infinite regress in our effort to understand consciousness, then we are forced to introduce some irreducible life principle of this kind. Yet, what can we say about the objection that this is a useless step that explains nothing?
Our answer is that once we recognize the existence of conscious life as something real, the possibility is opened up that we may be able to directly study this real entity and learn more about it. In order to make progress in such study, it would be useful if we could refer to an established body of scientific knowledge concerning consciousness. Unfortunately, such a body of knowledge is lacking in the West. However, a highly developed science of consciousness is expounded in the Vedic literature of India, including the Bhagavad-gita.
[24:56]
Here, for example, is a verse from the Bhagavad-gita introducing conscious life as a distinct energy from matter. Life is described is prakritim param, or the superior energy, to distinguish it from the inferior energy, or insentient matter. From the Bhagavad-gita we find that this superior conscious energy is made up of innumerable individual conscious entities, which can be referred to as atmas, or quanta of consciousness. Each atma is an irreducible conscious personality, possessing inherent senses, intelligence, and powers of action. According to Bhagavad-gita, the atma can be neither created nor destroyed and thus it corresponds to the kind of fundamental particle that physicists have always sought as a fundamental constituent of matter.
Now one might ask, how can we hope to learn anything about this atma? Is it possible to demonstrate its existence by some kind of experiment? Could you see an atma for example, by using an electron microscope? In order to answer this, let us briefly consider what is involved in seeing such entities as atoms.
Here we see a famous picture which is said to depict the atoms in the tip of a platinum needle. How was this picture produced and could we produce a similar picture of the atma? The picture of atoms was produced by a device called a field ion microscope. In this device, helium ions are accelerated towards the platinum needle situated in an evacuated chamber. As the helium atoms strike the needle, they become ionized. Now the needle is given a high voltage with respect to a phosphorescent screen at the other end of the evacuated tube. As a result, helium ions tend to fly away from the needle and strike the screen creating bright spots. Since the electric field around the needle tends to be strongest where the atoms protrude most sharply from the surface of the needle, the ions tend to fly preferentially in these directions. Thus, the bright spots on the screen correspond to the distribution of atoms in the needle.
We can see then that the study of atoms using the field ion microscope requires an understanding of how atoms interact in an electric field. Similarly, in order to study the atma, we must understand how it interacts with other entities. According to the Bhagavad-gita, the atma does not interact directly with matter through the agency, say of an electromagnetic field, and thus we could not expect to observe an image of an atma in some kind of microscope based on known physical principles. Rather, the atma is understood in Bhagavad-gita, to interact directly with the Supreme Lord or Paramatma, who is all-pervading and who accompanies each individual atma within the heart.
The verse states in Sanskrit, isvara sarva-bhutanam hrd-dese 'rjuna tisthati bhramayan sarva-bhutani yantrarudhani mayaya. In translation, this means that “The Supreme Lord is situated in everyone's heart, O Arjuna, and is directing the wanderings of all living entities, who are seated as on a machine, made of the material energy.”