At this point, it is an ugly marriage of a bunch of hand coded pieces of AI that basically type out answers to questions, and some kind of back-end processing that helps them process the information so that it can be presented to the end user. In order to get these things to work, the designers of conversational software need to have a good understanding of all of the different programming languages as well as the business model they are trying to use. However, this is only the beginning. For the moment, it is an ugly marriage of a bunch of complex pieces of AI that kind of work together, and a lot of hand coding by human designers.
Over time, as these programs get better, we will probably see more automated systems used in many different ways. Right now though, they are limited to answering basic questions. As we move into the future, these conversational software will start answering more complex questions. This may be where their real value comes in, because if they can't answer a question, the users might not continue to use the bot. In other words, this could be the beginning of a wide-scale internet application that is highly useful and sets the bar for future artificially intelligent bots.
Currently, the most successful artificially intelligent bot still comes from Google's initiative, called "Google Brain". This bot can easily beat the best human expert on chess, in one instance. Currently, there are commercial chat bot software packages that incorporate some form of deep learning technology. However, if these software developers can figure out how to make these tools work together, they may be able to create systems that are better than even the current best human programmers.
Currently there is no clear direction in which conversational software developers are going to take. They may want to follow the path taken by AlphaGo, who has recently defeated the grandmaster of chess, Gary Kasparov. That move appears to be to look at different bot techniques, and then combine them into a system that appears to have a stronger ability to learn from its past experiences. Alternatively, they may choose to build conversational software that looks like a computer program. However, the appearance of a computer program does not mean it can carry on conversations with people. These have to be done in a way where the interaction is natural, as is the case when you type or talk on your keyboard.
For example, when I am typing on my keyboard, I am actually interacting with the bot, and not the computer program I'm running. Therefore, the conversational ai software needs to pick up on the natural language I am using, and then translate that to my own language. Deep learning will eventually need to be combined with the translation of speech to natural languages, which will then allow the conversational software to hold true conversations. Another area of research that is being worked on is how conversational software can hold true conversations with voice recognition software. If we were able to give machines the ability to replace a whole range of human speech patterns, such as accents and inflection, then maybe they would be more effective at understanding human communication.
Another area where conversational software can be used is to write a blog post. When I am working on a blog post, I am using the graphical user interface tools, and not the conversational software. That is because the graphical user interface is very easy to use and familiar, and people like having their every need met when they are working on a project. However, if I were to use a conversational user interface, I might find it a little more difficult to write a blog post. Perhaps in the future we will see more conversational user interface tools being used on the website and applications.
This post: https://simple.wikipedia.org/wiki/Artificial_intelligence will help you understand the topic even better.