The Mind of the Machine

A simple electronic calculator doesn’t have AI. But a machine that can learn from its mistakes, or that can show reasoning power, does. Between these extremes, there is no precise dividing line. As computers become more powerful, people tend to set higher standards for what they call AI. Things that were once thought of as AI are now ordinary. Things that seem fantastic now will someday be humdrum. There is a tongue-in-cheek axiom: We can call “computer intelligence” true AI only as long as it remains a little bit mysterious.

Robotics and AI

Robotics and AI complement each other. Scientists have dreamed for more than a century about building smart androids: robots that look like people, act like people, and can even reason like people. Androids exist, but they aren’t very smart. Powerful computers exist, but they lack mobility.
 
If a machine has the ability to move around under its own power, to lift things, and to move things, it seems reasonable that it should do so with some degree of intelligence if it is to accomplish anything worthwhile. Conversely, if a computer is to manipulate anything, it must be able to cause a machine to do physical work according to a precise program.

Expert Systems

The term expert systems refers to a method of reasoning in AI. Sometimes this scheme is called the rule-based system. Expert systems are used in the control of smart robots.
 
The heart of an expert system is a set of facts and rules. In the case of a robotic system, the facts consist of data about the robot’s work environment, such as a factory, an office, or a kitchen. The rules are statements of the logical form “If X, then Y,” similar to many of the statements in highlevel programming languages. An inference engine decides which logical rules should be applied in various situations and instructs the robot to carry out certain tasks. But the operation of the system can only be as sophisticated as the data supplied by human programmers.
 
Expert systems can be used in computers to help people do research, make decisions, and make forecasts. A good example is a program that assists a physician in making a diagnosis. The computer asks questions and arrives at a conclusion based on the answers given by the patient and doctor. One of the biggest advantages of expert systems is the fact that reprogramming is easy. As the environment changes, the robot can be taught new rules, and supplied with new facts.

How Smart a Machine?

Experts in the field of AI have been disappointed in the past few decades. Computers have been designed that can handle tasks no human could ever contend with, such as navigating a space probe. Machines have been built that can play board games well enough to compete with human masters. Modern machines can understand, as well as synthesize, much of any spoken language. But these abilities, by themselves, don’t count for much in the dreams of scientists who hope to create artificial life.
 
The human mind is incredibly complicated. A circuit that would have occupied, and used all the electricity in, a whole city in 1940 can now be housed in a box the size of a vitamin pill and run by a battery. Imagine this degree of miniaturization happening again, then again, and then again. Would that begin to approach the level of sophistication in your body’s nervous system?
 
Is the human brain nothing more than an amazingly complicated digital switching network? Or is there something more to the human mind? No electronic device yet constructed has come anywhere near human intelligence in every respect. Some experts think that a machine will someday be built that is smarter than its creators. Others insist that the very idea is ridiculous.
 
It is tempting to extrapolate: If technological trends of the past few decades continue indefinitely, will the only limit on machine intelligence be defined by human imagination?