INTRODUCTION
Copyright ©2010 by T. Pavlidis
1. Some Early History and its Modern Echoes
Computers are machines that perform numerical and logical operations using basic elements consisting of a switch in a circuit that can be turned on and off by the current in another circuit. In the old days (around World War II) such elements were electromechanical relays or vacuum tubes and computers tended to be quite bulky. In the late 1950's transistors were introduced and computer size went down, while memory and speed went up.Then came integrated circuits, devices that could have millions of transistors on a small piece of silicon and computers have been getting more powerful and cheaper every year.
The basic scientific principles of modern computers were established in the 1930's. A major contributor to this development was the British mathematician Alan Turing (1912-1954) [W] who went on to supervise the building of one of the first computers and using it in order to break the German codes during War World II. (The breaking of the code is a fascinating story in its own right.) Turing was a tragic figure and committed suicide in 1954 (at age 41). In 1966 the Association for Computing Machinery (ACM) established the Turing Award that is given annually for outstanding contributions to the field of computer technology. It is considered the equivalent of the Nobel prize for Computing.
| Turing was interested
in the comparison of mechanical to human thinking and he proposed
a test to determine whether a machine exhibits human level intelligence.
A person sits on a teletype and types questions that are sent to
someone in another room who then replies. The questioner tries to
determine whether the responder is a human or a computer. If it
is a computer and the questioner thinks the responder is a human,
then the computer passes the Turing test. There is an interesting
modern application of the Turing test called CAPTCH (see the block
on the right for the meaning of the acronym) [W]. |
|
 |
The picture on the left contains an example of CAPTCHA and it also shows the circumstances of its use. The BBC news web site allows users to send links to articles to their friends but BBC does not want computers to do that. So BBC displays some text (the numerals in red) and asks you to read it and type what it says. That particular text is displayed in a way that it highly unlikely that a machine could read it. (Machines can read text printed in one of the standard fonts.) That is the CAPTCHA. For people who visually impaired the BBC site offers an alternative CAPTCHA using sound. You activate that by the button marked "Listen". Similar arrangement are used by many web services, for example, to make sure that user accounts are set up only for people and not for computers. (Computer programs that visit web sites are called web-bots, a contraction of web robot.) For more on CAPTCHA see a Tutorial on CAPTCHA. |
Besides Turing, another early major contributor to computing had an interest in the comparison between human and machine thinking. Most modern computers are designed according to principles developed by the American-Hungarian mathematician John von Neumann (1903-1957) [W] who wrote a paper on programmable machines around 1945. He died relatively young because of cancer. According to the Wikipedia article [W]:
"While he was in the hospital he wrote a short monograph, The Computer and the Brain, observing that the basic computing hardware of the brain indicated a different methodology than the one used in developing the computer."
Von Neumann had a computer built at the Institute of Advanced Studies at Princeton and the technician that did the work was Leon Harmon (1922-1982) [W] who later joined Bell Labs and he often gave lectures on the contrast between human and mechanical thinking. (I was fortunate to meet Harmon when he gave a seminar at Berkeley when I was a graduate student there. We kept in touch over the years, although by the time I joined Bell Labs Harmon had left for Case Western University.)
2. Computer Programs and Software
Computers cannot do anything unless they are programmed to do a particular task. It would be more accurate to say "the program running on this computer does such and such" rather than "the computer does such and such" but we will stay with the latter expression as long as its meaning is clear.
A machine can be programmed by wiring its elements to perform particular functions and such machines are called special purpose computers. Modern machines contained some hardwired elements but allow the setting up of connections by a set of instructions stored in their memory. Such sets of instructions are called software because they can be replaced easily to make the computer hardware perform different functions. A particular set of instructions is called a program.
A simple example of a programmable machine is an elevator. When you press the button for, say, the fifth floor the circuit controlling the elevator is modified to make the elevator halt when it reaches the fifth floor. You program the elevator to stop at a particular floor.
In the early days of computers programs dealt directly with the basic elements of the machine, so programming was quite laborious. Today programs are written in high level languages that are translated automatically into the low level instructions. The key point to remember that the machine itself has no intelligence - the intelligence comes from the software, programs that are written by humans. (Even when special purpose hardware is used, it must also be designed by humans.) What is often refereed to as "machine (or artificial) intelligence" are programs that are supposed to replicate human thinking and cognition. It would be more accurate to say that we are discussing "Computer Programs Replicating Human Thinking and Cognition versus Real Humans" rather than "Computers versus Humans" but we opt for brevity.
3. We Do Not Know Really Know How Humans Think
A big obstacle in writing software that replicates human intelligence and cognition is that we know too little about how the human brain works. Leon Harmon (see Section 1 above) used to give the following analogy:
Suppose some visitors from another planet arrive on the Earth and want to find how our computers work. One of them puts a probe on the casing of a computer and records the electrical voltage sensed by the probe. Another opens up a computer, takes its circuit boards and perform a chemical analysis of them. Finally, a third finds a computer printout and tries to figure out how the machine works from what is printed there. None of the three would be able to figure how our computers work. But there are the methods we are using currently to study the brain. The first one corresponds to EEGs, the second to the study of brain cells, and the third to psychological studies.
What is critical in understanding human thinking is the organization of the nerve cells and we know too little about that. People who are focusing on some superficial similarities between the basic computer circuits and the basic elements of the human (and animal) brain miss the point for two reasons. One is that the similarities are indeed superficial [1] and the other that, even if the similarities were real, intelligence is the result of the organization of the elements and there is not the slightest evidence that brain cells are organized like computer switching elements. There are billions of them in the human brain and their connectivity is the result of millions of years of evolution. In order to build a computer able to simulate the human brain it is not enough to start with basic blocks that may be similar in the two systems. We have to figure out how they are interconnected. To put it in another way, we have to figure out how to write program that perform some of the operations that human brains do.
The fact that we have been successful in programming machines to solve mathematical problems tells us nothing about the prospect of replicating human intelligence. Mathematics is a human invention that came very late in our evolutionary history so our brains have not adapted to mathematical tasks. It is quite a different story with tasks such as recognizing faces and expressions.
4. It is All Mathematics
If a problem can be expressed in terms of Mathematics, then it may be possible to solve it by computer. (However, it is not guaranteed to be possible because there are mathematical problems beyond the ability of computers.) But if a problem cannot be expressed in mathematical terms, there is no hope for its solution by computer. Of course a program can use information from its human users to help it find a solution. This is something that Google's search engine does very well and we shall discuss it in the next section.
Efforts to replicate specific human abilities, such as reading text by computers, go under the rubric of Artificial Intelligence (AI). Topics include game playing by computers (not games played by humans using computers), speech recognition, text recognition, image understanding (figuring out what is in a scene), robotics, etc. There are broad schools of thought on this subject. One is that of General Artificial Intelligence that attempts to develop theories that model the human brain and then solve all the problems that we listed above. The other is that of Narrow Artificial Intelligence where each specific problem is dealt independently of the rest. There have been numerous successes of Narrow AI but none of General AI. General AI has the appeal of being able to solve a problem without really understanding it, we are getting something for nothing. As a result, there have been numerous claims over the years but they appear to be in the same class as claims for perpetual motion machines.
Researchers who use the Narrow AI approach tend to shy away from that term because it is misleading. As we shall see the computer solutions of these problems are often done in a different way than the way humans use to solve the same problems. Chess playing programs offer a striking example of such a difference. Sometimes the best approach is to redefine the problem. When retail stores started using computers there was a need for a device that read a tag of a piece of merchandise into a computer. However, reading letters and numbers was very hard for computers so the solution was to introduce tags using symbols that were easily read by computers. The result was barcodes.
First Posted: May 28, 2010 — Latest Update: May 28, 2010