Let's say we want to develop a chatbot whose objective is to be able to talk to people about cars (i.e., "What do you think about BMWs?", "What is the fastest sedan in the market?", etc.). I am narrowing the domain on purpose.

In order to build such a bot, some approaches come to mind:

1. Hardcoded answers: use NLP to understand the meaning of the user input, go through defined rules (if-else basically) to choose the appropriate answer from a set of hardcoded, pre-defined ones.

2. Machine learning on existing conversations about cars: if you have access to lots of conversations about cars, you could use machine learning to train the bot to understand the meaning of the user input and to formulate an answer based on the data.

3. Machine learning on generic information about cars: same idea as above, but using generic infomation about cars (i.e., wikipedia, magazines) to formulate answers. NLP would need to be used both to understand the user input and to formular syntatically and semantically correct answers.

Questions: Are these approaches indeed viable? Are there others? Are there examples of bots using 2 or 3 out there?