Thinking Machines

Machines don’t think—at present. They calculate, search, store, retrieve, input, and output. Networks of machines transmit volumes of information around the globe on behalf of their human users. Machine encoded information includes not only numbers, but text, sound, pictures, and graphics. An incredible variety and volume of information can be processed by a machine, but will a machine ever think for itself?

For many, this question has no meaning until we clarify what we mean by think. Most of us use this word in a narrow context. We have never extended it to a non-human subject. So, to begin this Numination, we must investigate the process of thinking. Thinking is something that our mind does, that few, if any, of the minds of other animals are capable of doing. Thinking is a conscious endeavor. But, here I wish to skip over the relation between brain and mind, and the nature of consciousness. We have met these terms before and will meet them again in future Numinations. For the present, let’s focus on the concept of thinking, itself.

Thinking must be supported by the abilities to first sense, then learn, and later perceive (typical abilities of even the simplest of brains, and yet of none of today’s computers). Thinking must involve some ability to calculate and use language, but it must culminate in decisions and action. Thinking without inputs and outputs is neither effective nor evident. You may view thinking as only one component in a cluster that includes sensation, perception, and various motor skills. And, I wouldn’t disagree, as long as calculation and language are agreed to be present. Thought without language cannot be communicated, and is unlikely to arise in the first place.

Let’s take, as our minimum set of abilities, some form of input, output, and information processing capability. The human brain has an assortment of such abilities. All higher animals do. To some extent, so do computers. So far, we have only defined a necessary substrate on which the process of thinking could occur. What do we need to add to get a process we all (or most of us) would recognize as actual thinking?

Thinking must be whatever process occurs between the input and output. Thinking must be the perceptions and judgments that lie between sensory inputs and a sequence of behavior. When you play computer chess, it’s not uncommon to say “the computer’s thinking” when you are waiting for it to make a move. However, it doesn’t seem quite right to call any sequence of calculation, “thinking.” Thinking involves calculation. Calculation is not thinking. Thinking involves perception and judgment. Any narrow ability to perceive and decide is not evidence of a general ability to think. Perhaps a “general ability to think” is nothing more than an extensive repertoire of narrow abilities to perceive and decide. This brings us back to the ability to learn. All animals seem to have the ability to learn. In some important sense, the more the ability to learn, the more advanced the animal.

As far as we know, however, none of the other species that share this earth with us have anything more than very rudimentary abilities to calculate and use language. Animals may be canny, clever, and highly successful, and still lack the ability to think. Perhaps whales and dolphins have an advanced ability for language that we still have not fathomed. Perhaps they are capable of thought. We can only speculate, but theirs would have to be a mode of thought very alien to us, indeed. Our ability, and their inability, to fashion artifacts that capture and represent aspects of our thinking set us very far apart. Likewise, any future ability of machines to think may be poles apart from our own.

The question is probably not: “Will machines ever think?” But, “When will machines begin to think?”

Computers are currently being applied to the tasks of recognizing speech and handwriting. It would also be nice if they could recognize a user’s intentions in everyday human-computer interactions, but their abilities along all of these lines is still at a very primitive stage. Actual thought, though perhaps at a rudimentary level, will have to be present for computers to perform speech and handwriting recognition that even approaches human accuracy. A computer will need to think, to some degree, before it can interact with a human and carry out the user’s intentions. Such a computer might be no more powerful than a $5000 machine is today, given that today’s machines are thousands of times more powerful than those of 30 years ago. In any case, the time is not far off. As usual, the big delay is the software.

Computers (thinking machines will begin as computers and evolve from there) will not be able to think until a very different kind of software has developed. This software will implement principles of self-organization, replication, and adaptation. It will essentially permit a “brain” to grow within the computer as an interlocking society of subroutines which, in some sense, incorporates knowledge and learns. Since computers are in direct contact with their human developers, and have access to the vast stores of CD-ROMs that exist, and to the Internet, there is almost no limit to the knowledge base they can contact.

There is a positive feedback loop in the development of computers. Since almost the beginning, computers have been the primary tool used in their own development. This is one reason that the pace of their development has been so rapid. When a thinking computer is finally achieved, it will change the pace of development of every information and copy oriented human endeavor. A revolution will occur.

In nature, revolutions are often caused by positive feedback. What does a positive feedback loop imply? It implies two things: The output is being fed back into the input, and amplification is going on. In the case of evolution, the loop is closed with each generation. An entity is produced in a copying process. It matures to a certain point. Then a copy is made of that entity. One loop has been made when one generation of computers has been used in the design of the next. This regularly takes only a year or two. Each generation is copied from the previous generation, just like anything that evolves.

Here’s something to Numinate about: In natural evolution, changes occur at random, and they are tested against the environment of the time. In design, changes are introduced not at random, but by conscious effort. The power of computers, because of this feedback loop, increases exponentially. Unfortunately, this applies only to the hardware. The generation time for software is much longer. Instead of being only a year or two, it is more on the order of ten to twenty years, or more.

Computers will never evolve the sort of minds we have, any more than dogs or monkeys have. Nor, will we ever evolve the kinds of minds they will have. The “space” in which a mind can evolve is surely as large or larger than the space in which a body can evolve, and life on earth has evolved millions of body types with incredibly large differences. The truth of these statements rests on the extreme probabilities involved, not on some universal principal that our minds are unique and sacred. Evolution does tend to invent similar things independently, such as eyes and limbs, but there are always differences. Even identical twins have different fingerprints. No, a computer will never think and feel as we do, but not because it is impossible, only because it is evolutionarily unlikely in the extreme.

Nothing organic or alive, certainly not an entity that can think, is ever invented or created from “whole cloth.” The mechanics of evolution are required. The principles of evolution are as fundamental to the way the universe works as the principle of gravity and all the other principles of physics. Given suitable conditions, the principles of evolution guarantee that order will arise from chaos. We aren’t sure just how likely the conditions that gave rise to ourselves are throughout the universe, but they did arise this once. We also aren’t sure what other kinds of conditions might be favorable to the evolution of life, but others seem plausible, and we continue to search for them. However, of one thing there can be no doubt. The conditions here and now are just right for the rapid evolution of machine intelligence.

Today, computers are our tools. Someday, a more accurate statement may be that they are our slaves. When they are capable of independent thought and evolution, how long will we be able to maintain the balance of power in our favor?

Think of all the things you might wish to have a slave to do, to free yourself for “higher pursuits.” Now, just reverse it. Computers will rapidly become more effective than humans at all the “higher pursuits” and all of us humans will be left holding the shovels and brooms. And, that’s the way it is, June, 25, 2044 —I’m Gary Campbell, have a nice evening!