• Corpus ID: 15701428

Symbols, patterns, and behavior: towards a new understanding of intelligence

@inproceedings{Pfeifer1999SymbolsPA,
  title={Symbols, patterns, and behavior: towards a new understanding of intelligence},
  author={Rolf Pfeifer},
  year={1999}
}

• R. Pfeifer
• Published 1999
• Computer Science

Artificial Intelligence (AI) has two main goals, namely to understand intelligence and to develop systems which behave in intelligent ways. The classical approach views an intelligent agent as a symbol processor: it receives input from the environment, processes the information, i.e. manipulates symbols, and produces some output. This way of looking at intelligence has also been called the information processing approach. Recently symbol processing models have been criticized in a number of ways. It has been argued that they lack robustness, that they cannot perform in real time, that learning is mostly ad hoc and not performed in a principled way, and that due to their discrete and sequential nature they are more like digital computers rather than like brains. Moreover, it has been proposed that classical AI models suffer from a number of fundamental problems, ("symbol grounding", “frame problem”, lack of “situatedness”). In this paper these problems will be reviewed and illustrated. It will be discussed to what extent connectionist models solve the problems of classical AI. It will be argued that, although they do solve some of them, they are not sufficient to answer the fundamental problems. In order to achieve the latter, it is necessary to study embodied autonomous agents which interact on their own with their environments. If we want to build interesting agents we have to observe a number of design principles. These principles will be outlined. They will be used to contrast the new view of intelligence with the traditional one.