There has been some good and informative responses here but the point of Lisp and Prolog has either been missed, marginalized, or not emphasized enough.
Lisp and then later Prolog emerged in an era when the main AI research revolved around symbolic processing. A simple example of symbolic processing is how we humans do algebra, calculus, or integrals by hand. We symbolically manipulate the variables and constants to derive equivalent relationships. Lisp and Prolog were designed for this purpose.
Symbolic manipulation is not trivially implemented in C++ or Java for they were not designed with this purpose in mind. However C++, Java or similar languages may be buzzword languages in AI nowadays because there now exists several variations of AI research that do not deal with symbolic processing.
One form of AI deals with using statistical methods as the basis of knowledge and this requires using much leaner languages to reduce computation time. Also many so called AI systems are nothing more than specialized systems to serve a particular niche purpose. Of course these systems may be best programmed in a non-Lisp/Prolog language, and rely less on 'reasoning' or common-sense knowledge acquisition and more on processing data from inputs.
Even Watson (which is programmed in Java, C++, and a little Prolog) is arguably a highly specialized system. It appears Watson was designed to acquire a vast amount of facts whereby it then sorts through these facts using sophisticated search algorithms (not sure though and IBM would probably resent me for saying that). The future AI implementations will likely combine AI paradigms and implement various languages for each specialized part. Even Lisp and Prolog may one day make a comeback.
