Annual forecasts for 2018 predict a significant increase in the use of chatbots and voice assistants in all industries. A combination of chatbots and machine learning can significantly increase the added value, as the chatbot continuously learns from the dialogs and thus provides a better service.
One reason why chatbots up to now have sometimes been perceived as rather poor is that, compared to a classical web form, they often provide no added value. Added value is mainly generated when the chatbot not only follows pre-programmed rules, but also responds individually and flexibly to the current dialog with the user. The use of machine learning, and thus a step towards artificial intelligence, can make this dynamic possible. But how simple is this in reality? Which factors are key?
Looking at the functional areas of a chatbot, the user interface with Natural Language Processing (NLP), Natural Language Understanding (NLU) and Natural Language Generation (NLG) benefits with great certainty from elements of artificial intelligence. This makes it possible to understand the user's input more quickly and reliably, to derive the user's intentions and to process and display the results in the user's language. The cloud-based services of various providers, like Microsoft, Amazon and Google, which are emerging or already exist, provide a high level of functionality which utilizes machine learning in the background and which can be integrated into a chatbot with little effort. Therefore, it is hardly worth developing these elements for a new chatbot.
The greatest potential of artificial intelligence for a chatbot lies in the process expertise that is reflected in the dialog. The better the chatbot is able to conduct a dialog, the more efficient the chatbot's use becomes and the more value it generates for the user. Since a chatbot has to connect the needs of the user with many different data sources and systems depending on the situation as well as with a process, a machine learning model has to be customized and integrated into the chatbot. The model interacts strongly with the chatbot's personality (persona) and the design of the dialog. Therefore, it takes not only the UX design competence but also the technical understanding of the systems to be connected and of the possibilities and limitations of a machine learning model in order to realize the greatest possible benefit.
Last but not least, it should be noted that a learning system can also learn incorrect things, e.g. due to an attacker or unresolved problems in the process. One example was the Microsoft Twitter chatbot "Tay", which had to be switched off again after a short time because it had "learned" too many unfiltered, sometimes racist comments from other users and incorporated them into its communication. It is therefore not enough to just design the fair-weather case, but any number of various possible development scenarios must be considered, since a learning system will continue to develop independently to a certain extent. The quality of the chatbot is strongly related to the data sources and learning inputs used. The monitoring and change processes of this data must constantly ensure that the chatbot delivers adequate results and meets user expectations. In this case, a chatbot benefits from artificial intelligence if all the influencing factors are carefully analyzed and taken into account.