What Computers Can't Do
Hubert L. Dreyfus, Harper and Row, 1972

Reviewed by Ben Flanders, Computers and Society, Gordon College, 2002

Hubert Dreyfus provided a wonderful model of how NOT to argue in a debate. His book proclaimed to be a critue of the AI work over the decades 1950 to 1970 was often condescending and mocking to the leading scientist of AI. His arguement was presented as a philosophical writing and makes a valid arguement for the reason for impassibility and subsequent failures of AI. Dreyfus claims that AI scientist prefer optimism instead of seriously critiqueing their own field. Dreyfus also attributes the fact that AI scientist use improper assumptions and unfound theories as an explination of "failure" of AI work. His views are well founded but still very controversial.

     Hubert L. Dreyfus created a huge wake of controversy with his "critique" of the Artificial Intelligence field called What Computers Can't Do. This book has made an indelible mark on the field and the issues surrounding AI. However its reception was nothing short of cold. Dreyfus in fact gives the reader and any potential debator, a model of how NOT to critque a field of study and the field's major contributors. While backing up his views with rigorous agruements and well-reasoned philosophical study of human intelligence, he does all this with condescending and mocking tones. Our fair Computer Science professor here at Gordon likened mentioning Dreyfus' name in the AI community as mentioning the name Usama bin Laden within the New York community.

     But first let us take a look at Dreyfus' arguement and he reason for writing this critique. Hubert Dreyfus is obvioiulsy an extremely bright man. He did his undergraduate work at Harvard and at the time of writing What computers Can't do in 1972 he had been teaching at Harvard, MIT and Brandies. He consulted for RAND and worked in the Research Association in the Computer Science at Harvard Computation Labs through the National Science Foundation support. He was currently the Associate Professor of Philosophy at University Of California at Berkley. His critique came out of the observation of the trend with the field of AI and in seeing the "failures" of the field sought to shed light on the reasons for such failures. His audiance would probably not be the people he attacked within his book. Through use of explicit structures outlines and repition, Dreyfus was probably aiming to inform the educated reader that would not be in either AI or philosophy. Although it could be said that his work is a philosophical writing.


     Dreyfus begins with a survey of Western thought that was the motivation for the invested interest in finding intelligence within a digital discrete state computer. The basic assumption of the scientists within the 1950's and 1960's was that human intelligence and reason follow a set of rules or instructions. Once we have found these rules we could then code into a computer the necessary data that would express this intelligence. This idea first came to the Greek philosophers like Socrates and Plato who said that human thought could be "reduced to some calculation.XX" Socrates "proves" this in the Euthyphro by claiming if he didn't know an "effective procedureXX" for piety then how could someone ever be considered pious? In other words if we can't express piety then how do we really know what it is? Plato said "all knowledge msut be stateable in explicit definitions which anyone could apply.XX" However Aristotle said that these definitons must be void of intuition and judgement much like calculus. Leibniz sought such an actual universal language to express all human ideas. If this language could be found then any arguement could be simply settled with a pen and paper and a "Let us calculate, Sir" decree. Leibniz claimed to have invented an "elegant artifice" but did not deliver. Scientists such as Georg Boole and Charles Babbage claim closer to this reality by developing the binary algebra and designing the first computer by 1835. Obviously with the technology lacking the telephone switches that made the first computer impossible. Never the less with the introduction of transistors and the modern computer of today the question of AI has been brought to the forefront aided by logicians like Alan Turing.

     But even Turing himself was unclear of how it would ever be possible to have machines compete with humans. And thus begins the uncertain history of AI within the 50's and 60's. As mentioned earlier Dreyfus' impetus for his critique came out of a pattern of failure he recognized within the AI field. This pattern begins with a profound specific accomplishment in a machine showing intelligennt behavior in a controlled test, followed by overly optimistic projections of the future, a failure to be able to generalize the behavior, a long period of silence and then begins anew at another profound discovery. Dreyfus explains that this failure is due in part to the process AI use to conduct there research.

Inaccurate Self-Evaluation

     A recurring theme within Dreyfus book is the use of "optimism" as a negative conotation within the science field. Citing Noam Chaomsky as a leading and successful behavioral scientist, Chomsky says that "the experts have the responsibility of making clear the actual limits of their understanding and the results they have so far achieved.XX" However it is not hard to see how the early experts of AI provided the public with far too exaggerated claims of the current status and potential of AI research. For instance, Life magazine introduced Shakey to the world as a robot that could learn and think, plus it had an ego. "Several distinguished computer scientists are qouted as predicting that in from three to fifteen years 'we will have a machine with the general intelligence of an average human being...and in a few months [thereafter] it will be at genius level...'" We know today this is not true. But Dreyfus also had computer scientists within his own day making such exaggerated claims and not admitting defeat. It is scientific practise that if you hypothesis does not work out you have to change it. Dreyfus saw the AI maintaing there similar, disproven views without any sign of change. He thought this was one reason for their failure and inability to make progress within the field.

Improper Assumptions

     Dreyfus outlines four basic assumptions that the AI scientist use to conduct their research. If these assumptions are not well founded then research under these would lead to inaccurate results. The first assumption of AI scientists back in 1970 was a "biological assumption" that the human brain was a complex neural network in which the neurons were on/off switches that mimic the gates used in computers. No empirical evidence supports this assumption however. The brain viewed as a complex "symbol-manipulator" just doesn't agree with any research.

     The second assumption was the "psychological assumption" that regardless of the construct of the brain, it operated on bits of information according to formal rules. Dreyfus argues this is inaccurate. There are too many exceptions to the rules that we lay down for human "information processing." Humans have the ability to distinguish between relavent and irrelavent facts. Man and his composotion is physical and chemical, but this doesn't suggest that the process we use and recieving and process stimuli is similarly mathmematical in nature.

     The "epistemological (the study of thinking and knowledge) assumption" assumes that knowledge itself can be formulized into Boolean functions that in turn could be fed into a computer. Intelligent behavior may still be formulized even though processing may not be, and thus reproduced in a machine. This is "easily" refuted by observing that planets don't "know" how to follow planetary motion, they just do. Similarly while riding a bicycle now calculates their center of balance, we simply adjust and shift our weight. So even though we can observe mathemtical properties that doesn't mean agents involved follow them in order to execute behavior.

     The "ontological (the study of being) assumption" believes that all data can be fed into a computer and be discrete, explicit, deterministic and most importatnly independant all the time. If not then how are we to hope that any rulelike procedure can be applied to an unordered, interrelated set of data. However Dreyfus argues that understanding what an object is means also understanding how it relates to other objects and to human beings. Dreyfus argues that "context-free" data is useless and thsu refutes the intological assumption.

Unfounded Theory

     Regardless of how plausible the Platonic view of human behavior might be, it has been a rejected notion within the philosophy circles. Man as an object that can be constructed and observed within a set of rules was rejected by philosophers such as Heidegger and Wittgenstein. Again in quoting Chomsky, Dreyfus argues that "one should maintain enough perspective to be able to detect the arrival of that day when the research that can be conducted with these tools is no longer important." Dreyfus argues that AI refused to do this and not shift its views over the years to compensate for subsequent failures.


     The history of AI is in fact "marred" by the outrageous claims of scientists over the years. This doesn't necessarily negate the study but it does bring into question the methods and explination of failures. I think Dreyfus major aurguement is for the AI field to become more "scientific" and continue to make progress. But the question of what kind of intelligence are we looking for is the most important question. The search for human intelligence in machines may be fruitless although the field has brought about some important methods and discoveries. Dreyfus in attempt to bring up these issues has in fact alienated himself from being a contributer in the field. He has set the AI community in the defensive by lamgbasting the founders. Even though we are now seeing his views begin to be addressed as essential to the growth of the field, his book is not engaging at all and does not foster or promote change. For all his critisisms of AI community not being scientific enough, I think Mr. Dreyfus would have been shocked to find out that his arguement although profound and philosophical wasn't very scientific either. However I think by know he's heard that message.