The limits and capabilities of machines-a review

Understanding of the limit of machine capability, which is central to the 'deeper' question of the feasibility of artificial intelligence, is examined. The limitations and capabilities of machines are reviewed from the standpoints of computational, complexity, physical, conceptual, and conjectural limits. The old dilemmas of statistics, logic, pattern recognition, and natural language understanding are considered in this context and related to both the need for a world model and to some of the important results from algorithmic information theory. The more informal arguments against machine intelligence are also summarized, and the major problem of bridging the gap between the 'old' concept of machine and the hypothesis of artificial intelligence is again highlighted. It is concluded that although insight into such limits is still meager, there is the prospect that some powerful results will emerge from the research on complexity and (quantum) physical limits.<
>