AI, ML, NLP and Chatbots - an Expert Interview

Artificial Intelligence (AI) is a cutting-edge topic that occupies and sometimes challenges many large companies. The goal of Artificial Intelligence is to transfer the functioning of the human brain to computers in order to optimize processes. In an interview with Senior Software Engineer Christopher Kintzel from aiaibot, we dive deeper into the topic and answer basic questions about AI and Chatbots.

AI, ML and NLP! Can you explain in a few simple words what these are?

The term Artificial Intelligence (AI) is very broad and is defined in different ways. However, to put it simply, it describes an intelligence that is not applied by humans but by machines. At aiaibot, we deal with a specific subarea of AI in the context of Conversational Interfaces: Natural Language Processing (NLP), i.e. the processing of natural language. Different methods can be used for this purpose. One of these methods is Machine Learning (ML). In Machine Learning, rules are not defined and programmed by a human, but are learned automatically by the machine with the help of training examples.

To what extent does AI come into play in the context of Chatbots?

In the context of Chatbots, different methods from the field of AI can be used. A typical application is intent recognition. Texts are analyzed and categorized for this purpose. Subsequently, the dialog can be controlled on the basis of the recognized category or the recognized intent.

Another possibility is the automatic recognition of so-called entities. Structured information from the input text is recognized and read out. These can be times or addresses, but also customer-specific entities such as product groups. Here, too, the dialog can be controlled accordingly on the basis of the entities. For example, if the date and time are already recognized in the user's initial text for a request to make an appointment, the Chatbot does not have to query them again in the further course of the dialog.

A third method from the field of AI is Question Answering. Here, existing content of a company is indexed and a corresponding algorithm is trained so that questions about the content can be answered automatically.

Which of these options can be used with aiaibot?

The AI module can already be used to train classifiers that can be applied not only for email classification but also for intent recognition. The Machine Learning algorithms mentioned at the beginning are used here. But also the two other application areas (Entity Recognition, Question Answering) are already under development at aiaibot and will soon be available for our customers.

What is the difficulty of training algorithms?

Normally, the use of such Machine Learning algorithms and the training of classifiers require very specific know-how and expert knowledge. One of the goals in developing aiaibot is therefore to simplify this process and make it accessible to more people. Based on the feedback from our customers, I believe we have achieved this goal.

The biggest challenge we see in the implementation of customer projects is the ability to provide high-quality training data in sufficient quantity.

The biggest challenge we see in the implementation of customer projects is the ability to provide high-quality training data in sufficient quantity.This has a crucial influence on the results that our customers can achieve. In order to improve our offering even further, and to enable customers without their own training data to use it, we also want to offer the option of using ready-trained classifiers with predefined categories in the future.

There are two training methods available on the aiaibot platform - what are the differences?

When training a classifier for the first time, we recommend choosing the «fast» training method, as the training process here only takes a few minutes, and the classifier can thus be used by the customer immediately. However, we see better results in terms of higher accuracy in the correct classification of texts for most customers with the «accurate» method - but this can take many hours depending on the number of training examples. Nevertheless, every customer should test both options in any case, as the results can vary depending on the customer-specific use case.

How did aiaibot implement these solutions?

The algorithms we use are based on technologies from the open source sector and current research. These are continuously monitored by us, so that we can also integrate new developments into our platform. This way, our customers can automatically benefit from the latest research results from universities and large tech companies!

Would you like to learn more about AI? Read more about aiaibot's AI module here.

Thank you very much for the interview, dear Christopher.

Christopher Kintzel is a Senior Software Engineer at aiaibot, where he is responsible for AI as Technical Lead. He has extensive know-how in the areas of Machine Learning and Natural Language Processing. His core competencies are in the area of Chatbots & Conversational AI.