1. Difficulty in capturing "deep" knowledge of the problem domain. MYCIN, for example, lacks any real knowledge of human physiology. It does not know what blood does or the function of the spinal cord. Folklore has it that once, when selecting a drug for treatment of meningitis, MYCIN asked whether the patient was pregnant, even though it had been told that the patient was male. Whether this actually occurred or not, it does illustrate the potential narrowness of expert systems.•

2. Lack of robustness and flexibility. If humans are presented with a problem instance that they cannot solve immediately, they can generally return to an examination of first principles and come up with some strategy for attacking the problem. Expert systems generally lack this ability.•

3. Inability to provide deep explanations. Because expert systems lack deep knowledge of their problem domains, their explanations are generally restricted to a description of the steps they took in finding a solution. They cannot tell "why" a certain approach was taken.•

4. Difficulties in verification. Though the correctness of any large computer system is difficult to prove, expert systems are particularly difficult to verify. This is a serious problem, as expert systems technology is being applied to critical applications such as air traffic control, nuclear reactor operations, and weapons systems.•

5. Little learning from experience. Current expert systems are handcrafted; once the system is completed, its performance will not improve without further attention from its programmers. This leads to severe doubts about the intelligence of such systems.•

1. The use of computers to do symbolic reasoning, pattern recognition, learning, or some other form of inference.•

2. A focus on problems that do not respond to algorithmic solutions. This underlies the reliance on heuristic search as an AI problem-solving technique.•

3. A concern with problem solving using inexact, missing, or poorly defined information and the use of representational formalisms that enable the programmer to compensate for these problems.