The Evolution of Conversational Chatbot Systems

chatbot history

Conversational chatbots like ELIZA and PARRY in the 1960s and 1970s were early, crude examples that were basic in their implementation and effectiveness. Over the decades, the technology has increased considerably, resulting in examples like Alexa, Siri, and XiaoIce. We’ll take a closer look at the latter below, but let’s first go over the details of chatbots capable of realistic communication with human users.

Types of chatbots

Here’s a quick rundown of the various types of chatbots based on their intended purpose.

  • Chatbots for Entertainment – Early chatbots like ELIZA, PARRY, and others were conversational chatbots primarily meant for entertainment purposes. These early examples of chatbots were intended to give practical examples of the technology available at the time and had no real-world uses most of the time.  
  • Task-completion Conversational Systems – These chatbots are concerned with more than simply chatting with users. While able to understand different natural language inputs, they were meant to solve specific tasks like fetching the weather. Typically, they’re only able to handle a limited number of input types.  
  • Intelligent Personal Assistants – Beyond simple task-completion systems that only handle a handful of queries, intelligent personal assistants like Alexa and Siri are able to understand even more requests. These chatbots are typically tied to a single user and don’t need to handle interacting with multiple users.
  • Modern Social Chatbots – Modern chatbots created to interact on social media harken back to the days of ELIZA and ALICE, but they’re even more advanced than their early counterparts. Additionally, they’re typically released into a public space on an existing social media network.

Design principles of social chatbots

As chatbots have evolved over the years, most of the design principles have remained the same, but they have improved in numerous ways.


This refers to emotions and intellect. Basically, chatbot creators want to ensure their creation is smart (IQ) but at the same time can connect with users on an emotional level (EQ). Here’s a few of the ways this is achieved.

  1. Empathy – Understanding individual users is key for a chatbot to know how to respond in a way that invokes an emotional connection with the user.
  2. Personal Responses – A chatbot needs to recognize queries from individuals and be ready to respond in a way to maximize empathy with the user.
  3. Bot Personality – In addition to changing it’s tone or approach when queried by different personalities, a chatbot should maintain its own personal identity when replying.

Over the years, chatbots have become better at finding a good balance between EQ and IQ, allowing them to become even more “human” in the process.

Social chatbot metrics

Measuring results of task-oriented chatbots is relatively trivial, but when it comes to metrics for conversational chatbot systems, it’s not as easy. A common metric to use for these types of chatbots is known as conversation-turns per session (CPS). This refers to the amount of replies to question and how long the chatbot is able to continue a conversation.

Visual awareness

Another area where chatbots are becoming more intelligent is visual awareness. This is especially important for social media chatbots as most social networks heavily rely on art, photos, and visual elements. Typically, this is achieved with machine learning. A chatbot is trained to recognize objects within a photo so that it can formulate an appropriate response.

Chatbot case study: xiaoIce

Xiaolce “Little Bing” is a good example of a modern social chatbot. Released in China back in 2014, it’s become a hit with Chinese citizens over the years. Designed using the principles outlined above, XiaoIce is based on a 19-year-old female persona and has hundreds of skills. Worldwide, over 100 million people interact with the chatbot which has racked up over 30 billion conversation turns.

Originally deployed in China, XiaoIce was introduced in Japan in 2015, the United States in 2016, and India in 2017 by utilizing the scalable architecture of the original chatbot. Using visual awareness, the chatbot is able to “see” and comment on photos with an ever increasing amount of realism. Beyond interacting with people, XiaoIce has written and published hundreds of articles for QianJiang Evening News.

While XiaoIce is far more advanced than ELIZA and early conversational chatbot systems, the technology still isn’t able to fully emulate communication between two humans. That day may soon arrive, however, which is why it’s important to establish guidelines for the ethical use of chatbots. As the evolution of social media chatbots continues, philosophical questions may come to outweigh the technical problems.