Mitsuku Chatbot Wins Most Humanlike AI In Loebner Prize For A Second Time.

Adam Milton-Barker | Oct 13, 2016 | Artificial Intelligence | 15095

 

In all the hustle and bustle going on in the world of ChatBots at the moment, one crucial event that is very relevant to the future of AI, has seemed to go relatively unnoticed. If you are interested in ChatBots/AI you may of heard about the Loebner Prize, created by Hugh Loebner, which is a form of the Turing Test that Alan Turing first set up in the 1950's. The Loebner Prize is an annual event where AI specialists from around the world join to play off their bots against a panel of judges, in a battle for the most humanlike AI and a goal of tricking the judges into thinking they are talking to a real person. 

The event has previously been hosted all over the world, but since 2014 the event has been hosted in Bletchley Park in the UK, due to the connection between Alan Turing and Bletchley during the second World War. The event has seen a host of Artificially Intelligent entities claim most humanlike AI, but no one has ever successfully managed to fool the human judges into believing that they were communicating with a human and not a machine. 

When the Loebner Prize first began it consisted of a 5 minute conversation on a specific subject such as Shakespeare or hockey. As someone came close to winning, the test changed to become a chat bout any subject. Again as someone nearly won, the test evolved again and is now a 25 minute chat where anything goes! The closest anyone got to fooling a judge was when a human contestant started pretending to be a bot by saying things like "bleep" and "bloop". The judge thought he was so stupid, he must be the chatbot ;)

Something that seems very strange to me, is that although these events are covered live by Sky, in the bot community you hear very little about them, or the work of the chatbot pioneers that forged the foundations for the current bot craze over the last many years. 

Whilst having a conversation on the Bots Facebook group on day, I met a guy called Steve Worswick, creator of an AI called Mitsuku. Our common interest in AIML, the Artificial Intelligence Markup Language, got us talking. Up until that point, I kind of thought that AIML had died off in favour of machine learning, I currently only know one or two existing AIML developers, but was very happy to meet a developer that was still using AIML in their AI projects. At one point most of the AI systems I developed were powered by my own custom version of the AIML markup language and engine, which I named TAIML, but on reviving my projects I had moved away to other solutions. 

As we got to know each other more via Facebook it began to be apparant that our backgrounds and interests were fairly similar, both us were originally from the music industry, myself as a DJ and Steve as a producer,  both of us got into development through that industry and eventually phazed out the music in favour of developing, and both of us had a passion about AI and also the AIML markup language. 

In Steve's case, his journey into the world of chatbots began around 10 years ago when he stumbled upon a website called Pandorabots where you can create your own chatbots for free. His first bot was a teddy bear chatbot which he created in 2004 and hosted on his site. Before long, the bot was attracting more attention on his site than his music, taking that as a hint Steve decided to give up the producing and focus on chatbots, leading him to the creation of Mitsuku. 

Mitsuku is one of the most popular chatbots running on Pandorabots: the world’s leading conversational artificial intelligence chatbot platform. Pandorabots offers a free web service for building chatbots, and has powered over three billion conversations between bots and end-users to date. Mitsuku alone fields hundreds of thousands of queries daily from users all over the world, and -- in contrast to offerings like Apple’s Siri -- developers can use the AI as a service.

Back in 2013, Steve entered Mitsuku into the Loebner Prize and went on to win the prize for the most humanlike AI conversational program. Incidentally, that year someone actually unofficially entered Siri into the test, Mitsuku won hands down with Siri placing 14th! More info here. Mitsuku was also entered into the competition in 2014 and 2015, placing 4th in the 2014 contest and coming second by 0.5 points in 2015. Bringing us back up to present day, Steve once again entered Mitsuku into the Loebner Prize this year and the bot absolutely smashed it! In the qualifying round Mitsuku scored 90%, with  2nd place trailing behind at 78%, and in the final, where the best score is 1 (lower is better), Mitsuku scored 1.25 while 2nd place scored 2.25! I have a little bit of a claim to fame with Mitsuku, I was very happy to find out that Steve actually listens to my DJ mixes on Soundcloud whilst coding Mitsuku, and did so whilst preparing her for the Loebenr Prize this year, I like to think I played my own little part in the success of Mitsuku in this years competition, but obviously without a doubt, the success of Mitsuku is down to the hard work and dedication of an amazing programmer and his passion to get that one step closer to the reality of true AI.  

So for those that did not know too much about the Loebner Prize, this article will of given you a brief introduction to the event and what it is about, as well as giving some well deserved recognition to a great developer and friend, Steve Worswick. 

This is my second article that I will be contributing to Chatbots Magazine, and I will make a point of continuing to share information about the history of ChatBots, I hope my articles are of interest to the new comers and open their eyes about the history of the community.