The Point: Why This Article Matters

  • Where we are. Most chatbots today can answer simple questions accurately about the weather, what’s on TV or return an account balance, but more complex questions elude them — with ChatGPT now the big exception.
  • Where we're going. Over the years, chatbots have evolved and advanced to become more humanlike and sophisticated, with some capable of performing tasks such as writing computer code and composing music — with no end in sight.

Even though the conversational chatbot ChatGPT is all the rage right now, many people will probably be surprised to learn that chatbots have actually been around since the 1960s when ELIZA was unveiled by Joseph Weizenbaum at MIT. Chatbots have come a long way since then.

Perhaps the most recognizable chatbot is Apple’s Siri, but chatbots can be found almost anywhere that people and machines interact today. 

“Chatbot itself is a bit of a catchall, because it's a really handy term because there's two components of any chatbot, there's the language processing and action taking, or there's the chat bit, and there's the bot bit,” said Will McKeon-White, an infrastructure and operations analyst at Forrester.

While there are accuracy, veracity and copyright issues with the latest generation of generative AI chatbots like ChatGPT, ChatSonic, Jasper Chat and others, they are garnering a lot of attention because these chatbots answer questions in much the same way a human would — only much, much faster. They can also write simple computer code, create art, compose music, and generate marketing content and essays that rival what humans can do.

“One of the biggest barriers for this technology has been in respect to earning the trust of the user,” said Natarajan Venkatesan, director of intelligent automation for NTT DATA. “Earning trust comes with two other important factors, security and privacy. Striking the right balance between security, privacy and trust has been critical, and ChatGPT has broken this barrier.”

Related Article: 6 Ways ChatGPT May Change Digital Customer Experience

Understanding the Different Types of Chatbots

While the underlying technologies that make chatbots function — decision trees, natural language processing (NLP), natural language generation (NLG), text and voice user interfaces (UIs) and the like — are all similar, the way these technologies are combined and deployed as well as the use-case for which they are being used all factor into each chatbot’s identity, said Rosaria Silipo, head of data science evangelism at KNIME, a data science platform provider.

“The variety of chatbots around at the moment differs depending on the domain, i.e., the use case, the more or less advanced technology in the background, and the UI, for example voice based, text based or even button based,” she said. “But, frankly, I think they are all a variation of the basic chatbot structure.”

In fact, after searching online and asking three experts, the best and most succinct response to this question came from ChatGPT itself. (To ensure accuracy, ChatGPT’s answers were cross-referenced with blogs from Spiceworks, the startup MessengerPeople, the chatbot company HelpCrunch and others.)

According to ChatGPT, chatbots can be classified in various ways based on their capabilities, applications, and underlying technologies. Some of the more common types of chatbots include:

Learning Opportunities

  • Rule-based chatbots that rely on pre-programmed rules, keywords and patterns to understand and respond to user inputs.
  • Self-learning chatbots that use AI to learn from user inputs, new data sets to improve understanding and responses over time.
  • Hybrid chatbots that use a combination of rule-based and self-learning techniques to create a natural user experience.
  • Task-specific chatbots that are designed for specific tasks such as customer service, ecommerce, healthcare or education.
  • Avatar and emotion-aware chatbots such as Replika that interact with users through avatars, which can be in the form of a human, an animal or a cartoon character. These chatbots recognize and respond to human emotions, such as happiness, anger or sadness.
  • Personal assistant chatbots such as Amazon's Alexa and Google's Assistant help people with everyday tasks and can be used to control home automation systems with voice commands.

Related Article: ChatGPT: What You Need to Know

The Advancements in Chatbot Technology

Under the hood, chatbots fall into two main categories: syntactic-based or semantic-based, said Venkatesan. Syntactic chatbots are single-turn in that they do not understand context or have conversational recall. Because of these shortcomings, interacting with them is often frustrating. Semantic-based chatbots, like ChatGPT and Siri, are multiturn (i.e., they are conversational), context-aware, so they can better mimic humanlike interactions. 

“The commonality across these technologies is NLP, NLU [natural language understanding] and AI/ML, but the way these are used and the algorithms are different,” he said. “ChatGPT for instance uses the generative pretrained transformer algorithm, while other technologies use other algorithms like recurrent neural network or long short-term memory. By creating these layers of neural network, chatbots use deep learning to imitate humanlike responses.”

Chatbots Will Continue to Improve

While ChatGPT is impressive, it is the exception. Most chatbot interactions today can be characterized as less than satisfactory. Most chatbots today can answer simple questions accurately about the weather, what’s on TV or return an account balance, but more complex questions elude them.

ChatGPT and other generative AI, large-language model (LLM) chatbots are seen as game changers because their two core functions — understanding a request and acting on it — are exponentially better than what is currently in widespread use.

“ChatGPT and other similar technologies like Amelia and Kore.AI, when used within an enterprise," said Venkatesan, "fundamentally change the game by bringing in the ability to be more accurate and complete, solving a problem end to end without human involvement."