Rule-based or AI-based: The technologies behind chatbots

The interaction between man and machine has not always been as straightforward as it is today. It wasn’t until the development of the graphical user interface (GUI) some 40 years ago that the general public was offered a simple way to communicate with computers. Of course, quite a lot has happened over the last four decades. A great many tools and processes have proven their worth, while others have become obsolete and been replaced by more innovative and intuitive solutions. Developments in the field of speech recognition, voice control, and artificial intelligence (AI) have made and continue to make it more convenient for users to communicate with end devices. These advances, coupled with the resulting proliferation of voice and chat wizards, have also led to standardized technological approaches that make it simpler for companies to use conversational applications – and easier for developers to create them.

When it comes to building conversational apps today, AI and rule-based approaches are among the most commonly used.

Rule-based approaches

Prior to the advances in artificial intelligence and machine learning (ML), conversational applications were usually created using rule-based approaches. Even today, this remains the fastest and easiest way to model a basic demo of a voice or chat wizard.

Rule-based frameworks are essentially capable of responding to questions for which there are predetermined answers. They provide a framework to host the logic for word processing, guiding users through a developer-defined decision making path (decision tree).

The core logic behind interpreting incoming requests and delivering appropriate responses usually consists of innumerable implemented rules. If the messages – in other words, the requests – correspond to a certain pattern, script-based responses are returned. The applications are designed to be capable of taking a given action in response to each identifiable pattern or condition.

Since rule-based frameworks do not offer AI capabilities for parsing (syntax analysis) or classifying incoming requests, the developer has to code all the necessary message processing and interaction logic manually. Even simple applications typically require hundreds of rules to map the different dialog states in a typical conversation. On top of this, conflicts and redundancies between the rules create a level of complexity that is virtually unmanageable. Consequently, more sophisticated, AI-based approaches tend to be the method of choice. 

AI-based approaches

The main difference between AI-based and rule-based approaches is that AI-based applications use technologies such as machine learning and deep learning to understand and correctly respond to current, past, and future conversations. Based on success or failure rates from past interactions, they can modify their answers to questions or change their response.

With the help of comprehensive training data, machine learning constructs a general model that allows unknown inputs to be automatically assigned to a specific output, with no need for a manual program. To be able to recognize and analyze data and the relationship between data properly, machine learning encompasses several methods that operate based on experience with the elements of their environment. One of the best known methods is the model of artificial neural networks (deep learning), which simulate the structure of the human brain.

In practice, applications that combine both rule-based and AI-based technological approaches with a range of different speech recognition methods are particularly well established. Depending on the situation and context, they use defined patterns and rules as well as AI methods. To be able to learn the linguistic nuances, similarities, and meanings of different sentences, they also make use of various speech recognition methods such as NLP (natural language processing) and NLU (natural language understanding).

Several of our customers have already optimized their existing chatbot solutions using Mindbreeze InSpire. Are you interested in a demo that is precisely tailored to your requirements? We’re looking forward to hearing from you.