Artificial intelligence is an enormous area with several subdisciplines. The special library you'll want will most absolutely be a function of what you're trying to do.
If you are trying to do any kind of machine learning then you'll apparently want to find a library including state-of-the-art classifiers, like neural networks, SVMs, boosted decision trees, k-means classifiers, etc.
If you are trying to develop robot control systems then you'll apparently want to look into particle filters, Kalman filters, probabilistic road maps, and many more.
If you are trying to do computer vision or speech recognition then you might need a way to signal processing libraries in addition to the aforementioned machine learning tools.
If you are trying to have the computer play a game then you might want libraries for minimax, UCT, or Monte Carlo simulation.
In precise, there are a large number of libraries. The first step should be recognizing what type of problem you want to solve.