Dialectical Argumentation As A Heuristic For Courtroom Decision Making

In his 1994 article 'On the Artificiali ty of Artificial Intell igence', Crombag claims that artificial intell igence (of the symbolic kind) is of little psychological relevance. According to Crombag, our understanding of the mind (and of the legal mind in particular) is hardly enhanced by artificial intell igence research. The main reason for his opinion is that there is no indication that the brain is like a digital computer in structure or in behavior. I agree with Crombag that digital computers and brains are very different in structure. However, no one not even the early far too optimistic advocates of good old fashioned artificial intell igence of the 60s has ever had a different opinion. I disagree with Crombag that digital computers and brains are different in behavior. Of course a desktop computer (often used for text and message processing only) is very different in behavior from a human brain, but that is irrelevant. In an important sense, digital computers and brains can be similar in behavior. The key point is that digital computers can be programmed, and that there is virtually no limit to the kinds of programs that can run on digital computers. From the perspective of information processing, digital computers are good approximations of implemented universal Turing machines, which essentially means that they can be programmed to mimic any kind of information processing. And this really means any kind. Not only logical reasoning and playing chess, but also interpreting visual data and finding analogies are kinds of information processing. Even legal reasoning can be fruitfully viewed as a kind of information processing. Thagard's (1996) introduction to cognitive science gives many examples of what he calls the computational-representational view of the mind. This does not mean that if a digital computer mimics certain brain behavior, it mimics this behavior at all l evels of inspection. Of course not: a digital computer is not the same as a human brain. If one probes deep enough, the differences become obvious. It remains a miraculous fact that at a sufficiently high level digital computers can mimic any kind of information processing behavior. Even though, internally, the digital computer Deep Blue and Kasparov's brain found their chess moves in very different ways, at the level of playing grandmaster chess Deep Blue's and Kasparov's behavior were much alike. The only limitations on the information processing capabiliti es of digital computers are computational speed and memory, but these bounds are still pushed back at a high pace. Crombag (1994) paraphrases the well-known connectionist challenge to artificial intell igence of the symbolic kind, advocating neural network models. The challenge comes down to the demand of designing a model with a slightly deeper level of detail than in symbolic artificial intell igence. Instead of only looking at the whole brain as an information processor, the role of individual brain cells and their interconnections is taken into account. Interestingly, a more recent challenge to traditional artificial intelligence does not consider the level of individual brains as too high (as in the connectionist challenge), but as too low. The higher level of interaction of brains with their environment and with other brains is given due attention. The best models of brain behavior must of course take all intertwining levels of looking at the brain's behavior into account: the low level of brain cells, the intermediate level of brains, and the high level of communicating brains interacting with their environment. The psychological relevance of artificial intell igence is however not to be found in the claim that digital computers can behave like brains. The importance of the rise of artificial intell igence is that it has added a powerful tool to the research methodologies of psychology: models of brain behavior can now be successfully implemented on digital computers. This new methodology has at least three advantages. First, the goal of implementation of a brain behavior model on a digital computer sets very high standards to the design of the model itself. It has to