The Thick Blue Line Between Chatbots

Since the recent upsurge of chatbots, a lot of confusion has surfaced about chatbots in general. Similar to the emergence of laser technology when no one was quite sure whether we were entering the star wars era with military satellites shooting each other using laser guns, some people mistakenly expect that the chatbot technology can deliver results close to the fantasies portrayed in Hollywood movies. To make things more explicable, we first need this basic distinction: Transactional chatbots versus expert chatbots.

TRANSACTIONAL Chatbot is a simple interactive system delivered in a dialogue style interface where the main objective can be as simple as organizing calendars, booking reservations, helping check-out process, or presenting retail options.

EXPERT Chatbot is a sophisticated conversational system that actually engages in a dialogue with the user about the expert knowledge it embodies. The objectives are more ambitious such as help desk, tech support, employee assistant, or online advisory systems.

The Difference
The blue line in between these two categories of chatbots is actually not thin, if not very thick. The thickness comes from the scientific and technological challenges involved comparing one to another.

Transactional chatbots operate in a single dimension: Natural Language Processing (NLP). Using NLP, whatever the user says (or types) is to be detected in order to perform some form of transactional response. If the chatbot is designed to operate for a specific task, the sphere of language detection is nicely confined within a narrow band. As a result, some basic NLP technology would be enough to power these chatbots as long as the users’ dialogue do not diverge away from the chatbots main objective. Many messenger chatbots fall into this category, and some has real commercial future.

The situation with expert chatbots quickly gets much more complicated because they must operate at least in 2 dimensions. The second dimension, machine learning (ML), is needed to absorb content (expert knowledge) into the system. As shown in the diagram above, ML will also help the NLP capability (green line) with increased content and training. The problem here is that the conventional ML algorithms (like neural networks) are complicated, rely on data volume and quality, and are black-box approach. They are mainly an “engineering” approach to knowledge processing devoid of innovations in computational linguistics, semantics, and even in psychology. New approaches to ML are needed to fit this particular operation before expert chatbots can flourish.

Yet There is the 3rd Dimension
One of the largely misunderstood aspects of chatbots is the missing 3rd dimension: dialogue behavior. This 3rd dimension gives “life” to a chatbot which would otherwise act lifeless without any active speech or goal-oriented behavior.


To give you an example, if the purpose of the chatbot is to be a sales person, then it should have a salesmanship behavior. Behavior modeling is another untapped territory that involves disciplines as far away from engineering as it can be: psychology. Game-theoretic, behavioral algorithms can actually provide a framework for NLP and ML processes to fill in the dialogue engine with knowledge and language detection.

We, at exClone, are pioneering this combination using an umbrella term “digital cloning technology” that can be accessed via our Creators’ Platform coming out soon. The good news is that chatbot developers will not need any coding, or even experience in any of these disciplines. Although “advanced” options will be provided for those technical geniuses, the platform will aim enterprise clients who want to solve the problems of communication quality, scalability and cost across enterprise operations.

The Last Category: Purposeless Chatbots
I would originally call this group of chatbots as “useless chatbots” because they are. Unfortunately, there is an alarming rate of them being advertised (surprisingly) by the misguided PR departments of tech giants like IBM, Microsoft, Apple, and Google. Let me give you two examples, then you can fill the rest.

What is Siri? Apple adopted Siri, which could be a good move, but then deployed it without any clear “purpose” by hijacking an important button functionality. As a result, 9 out of 10 things you ask is not fulfilled by Siri. Siri does not ask you any question, it does not guide you to any specific problem solving. Why not deploy Siri with an actual purpose, even if the purpose is limited to, say organizing your calendar? A clear lack of vision (after Jobs) manifests itself by not even being a platform (unlike Amazon’s Echo), so that purposeful chatbots could have been launched from a Siri platform.

Another example is Microsoft’s Tay, now its derivative Zo. Using a pure engineering approach of machine learning, they are training the system using tweets to create a chatbot that has no clear purpose. What problem does it solve? If it is a social experiment, Zo will become an average teenager, with knowledge of teenager tweets, without any distinct behavior, and no contribution to society beyond chit chat engagement. Such projects will eventually come to a realization that knowledge representation, natural language, and human-like dialogue cannot be handled solely by an engineering approach, and such purposeless, “do-it-all” attempts will fall short.

Purposeless chatbots are destined to suffer from the lack of usefulness, and they will fail to fulfill a promise that has been advertised beyond reality. Although we are accustomed to see these companies shooting themselves in the foot, it is somewhat interesting to see that while they are promoting “learning” technology, they are failing to learn themselves not to hype the tech markets for short-term sensationalism.