What Computers Still Can't Do

“What Computers Still Can't Do” written by the philosopher Hubert L Dreyfus (1) exposes the shortcoming of Artificial Intelligence (AI), like the claim that in order to act intelligently  people must have a model of the world in their mind. Which is known as mental representation of the world. According to Herbert Simon this model is a system of symbols which serve  as representations of reality, and since computers manipulate  symbols they may act intelligently like people. Dreyfus  maintains that disembodied machines cannot mimic human intelligence and hitherto he was right. The  embodiment concept is discussed elsewhere.

One may argue that today’s computers, are still too slow for such a task. What about computer systems, like the WEB? Might the WEB act intelligently? “Not at all!” says Dreyfus, since the WEB is disembodied.  True,  it has some embodiment, like SKYPE or VOIP yet this is not enough. The WEB is only a tool and not an intelligent machine.

What Dreyfus fails to realize is that the WEB is a living system embodied by its users. So why  can’t it be intelligent?  Elsewhere I mentioned other living systems, like the stock exchange which cannot be labeled as intelligent. What kind of embodiment is required to make an intelligent system?

At the time when AI (artificial intelligence) was named by John McCarthy,  Wittgenstein's philosophical investigations came out against mental representations.  Heidegger had already done so in 1927 with “Being in Time”.  Yet the  AI researchers still ignore it. They are followed by neuroscientists who place mental representations into the brain. Other regard the brain as a computer. A neural net of representations. They ought to wake up from their Cartesian slumber and turn to phenomenology which is the motto of the present trail.

An intelligent machine can be created only if adopting the viewpoint of phenomenology, which criticizes the notion of mental representation. Yet phenomenology does not point the way to the creation of an intelligent machine. It only explains why Cartesian reductionism is wrong. Nevertheless it is a basis from which AI ought to continue. To me phenomenology is a way to tackle complexity. Embodiment is an attribute of a complex system. It can be realized even in a relatively simple two CA system. (v also)


1. Hubert L. Dreyfus What Computers Still Can't Do: A Critique of Artificial Reason
ISBN 0-262-54067-3

Back to complexity index