Did you already have a productive conversation with a chatbot? This blog post gives a short introduction to how bots have developed from funny games to serious business.
Chatbots are starting to make an impact on our world. Recently the story of Microsoft’s Xiaoice, a Chinese language chatbot, surfaced. Reports are claiming that average users interact with the system on a daily basis, and even attach emotionally to the software based conversational partner. While the purpose of Xiaoice seems to be research and entertainment, the majority of chatbots projects now have commercial goals of information delivery and persuasion. But how did this all start?
The first chatbot software dates back in the 60s. A computer program called ELIZA, written by Joseph Weizenbaum, was based merely on pattern recognition and preprogrammed templates of conversation by rephrasing. People were emotionally attached to the “personality” created by the software, even though Joseph Weizenbaum had built the software to express the superficiality of human machine conversation.
Then came text based adventures, surprisingly funny to play. Suddenly, a complete story could be played interactively via a chatbot interface. My preferred one was The Hitchhicker’s Guide to the Galaxy, reflecting the hilarious humor of Douglas Adams.
Let’s take a closer look at the advances in chatbot technology. The early versions of chatbots were completely rule based, a line of conversation that was not programmed, did not yield any productive output. This resulted in a slowdown of developments and deployment of chatbots. Programming each possibility of a conversation was too time consuming. However, in the recent years, statistical models were generated from vast amounts of conversational data. Those models could be used to overcome the problem of rule implementation. A chatbot using such statistical models would be able to respond to unknown inputs in a more flexible way, but opening the issue that such conversations could go completely wrong when the conversational data in the model was biased. Recently this risk materialized badly in the case of a chatbot becoming racist.
The root cause of both issues is that the bot does not understand what it is saying and has said. In the same way the bot has only a very limited model of the state in which the human user is. But this is about to change. Latest progress in cognitive computing, as advertised here, is going in the right direction. I came across an impressive AI based chatbot with the name Luna here. However, I found no way to look behind the video if the conversation is genuine or fake. Do you have more information about Luna? Comment below! Another impressive example of a cognitive AI is the performance of the Watson system in the quiz show Jeopardy. Cognitive chatbots will allow much more natural conversations with humans, deepening the ability to persuade humans, but also increasing the value they create for humans.
Meet a Bot
Here in Vienna we have a fine community working with chatbots, a good entry point is the Botshub Vienna Meetup. And there is a good opportunity to meet a real chatbot at our next TEDxVienna Event Adventures. During the event, the Austrian Patent Office will present a chatbot capable of answering questions about trademarks and IP protection. As for us, TEDxVienna has a very own Idea Bot, eager to show you any TEDxTalks of your interest!
I am looking forward to pose some tricky questions about trademarks and cost for international protection at the event. Also I am curious if the bot remembers the context of the conversation 🙂
Cover image by Pixabay