This chapter demonstrated how to build a natural language interface for two different domains: breeding rules of the Pokémon game, and course advising for college students.

We developed a pipeline with three components: first, we determined what the user's question was about, known as its "intent," using the Rasa library. Then we computed the answer to the question by referring to domain-specific logic implemented in Prolog. Finally, we generated a natural language response using the data found by the logic backend.

As far as the user is concerned, they provided an English-language question and got an English-language response. It would be straightforward to handle voice input and speech output by using Google's Speech-to-Text and Text-to-Speech APIs. These two services would be added to the beginning and end of the pipeline, respectively, without requiring any changes to the existing three-stage pipeline. Finally, we also addressed a few issues related to evaluation to ensure the natural language interface works well for users. Since these kinds of interfaces engage users with regular language rather than code or an application-specific graphical interface, there are issues and insights unique to these kinds of interfaces that we would not normally find in common user tracking and feedback techniques.

In the next and final chapter, we'll discuss how AI is perceived by the public and how it suffers from hype cycles, that is, dramatic shifts in AI's popularity. We'll also offer advice for businesses who wish to make productive use of AI technology without becoming victims of the hype cycle. We conclude by looking at the near-term future of AI.