Tutorial on Embodiment

2.2. Successes of Traditional AI


"The term embodied intelligence was introduced in the mid-1980s in the field of artificial intelligence as a reaction against the classical approach, which views intelligence as merely a matter of abstract symbol processing. What matters in the classical approach is the algorithm or the program - the software, if you like - and not the hardware (the body or brain) on which it runs. Abstract functioning that is independent of the specifics of a particular hardware is an extremely powerful idea and constitutes one of the main reasons why computing has conquered the world, so to speak: all that matters are the programs that run on your computer; the hardware is irrelevant. This line of thinking goes back to the famous Dartmouth conference, held in 1956 in the small town of Hanover, New Hampshire, when "artificial intelligence" was officially launched as a new research discipline. The American philosopher John Haugeland of the University of Chicago, author of the well-known book Artificial Intelligence: The Very Idea, an excellent philosophical treatise on traditional or classical artificial intelligence, coined the term GOFAI-"Good Old-Fashioned Artificial Intelligence"-to designate this approach (Haugeland, 1985). In the classical perspective of artificial intelligence the human being was placed at center stage, with human intelligence as the main focus. As a consequence, the favorite areas of investigation were natural language, knowledge representation and reasoning, proving mathematical theorems, playing formal games like checkers or chess, and expert problem solving. This last area became extremely popular in the 1980s. Expert systems, as these models were called, were intended to replace human experts, or at least take over parts of their tasks, in areas like medical and technical diagnosis, configuration of complex computer systems, commercial loan assessment, and portfolio management. These systems epitomize the classical approach of viewing humans as symbol processing systems, i.e., as systems that manipulate symbols as computer programs do. This so-called information-processing approach strongly influenced researchers not only in artificial intelligence but also in psychology and the cognitive neurosciences. And now it seems that scientists as well as people in general see human intelligence as information processing: "What else could it be?" is the standard defense of this view. Computer scientists and psychologists teamed up to develop information-processing models of human problem-solving behavior, in particular expert systems.

In the 1980s there was a lot of hype surrounding expert systems and many companies started to develop them - alas, many soon went bankrupt after this way of conceptualizing human expertise and human intelligence in general turned out to be flawed, as discussed in "Problems of Traditional AI"  (see also Clancey, 1997; Pfeifer and Scheier, 1999; and Winograd and Flores, 1986). By the mid-1980s, the classical approach had grown into a large discipline with many facets and with fuzzy boundaries, but despite some of its flaws, it can now claim many successes. Whenever you switch on your laptop computer you are starting up many algorithms that have their origin in artificial intelligence. If you use a search engine on the internet you are, for example, making use of clever machine-learning algorithms.

If you use a text-processing system, it in turn uses algorithms, which try to infer your intentions from the context of what you have done earlier, and will often volunteer advice. Natural-language interfaces, computer games, and controls for appliances, home electronics, elevators, cars, and trains abound with AI algorithms. More recently, data-mining systems have been developed that heavily rely on machine-learning techniques, and chess programs have been designed that can beat just about any human on Earth, which is a considerable achievement indeed! The development of these kinds of systems, although they have their origin in artificial intelligence, has now become indistinguishable from applied informatics in general: they have become an integral part of today's computer technology." (Pfeifer and Bongard 2007, p. 27, 29-30)



Pfeifer, R. & Bongard, J. C. (2007), How the body shapes the way we think: a new view of intelligence, MIT Press, Cambridge, MA.
Pfeifer, R. & Scheier, C. (1999), Understanding Intelligence, MIT Press.
Haugeland, J. (1985). Artificial intelligence: The very idea. Cambridge, MA: MIT Press.
Clancey, W. J. (1997). Situated cognition: On human knowledge and computer representations. New York: Cambridge University Press.