Mitsuku recently appeared in her Loebner Prize form at a live event hosted by Radiolab, a popular station based in New York City, in their recent podcast about chatbots. She appeared as part of a Turing Test in which the audience tried to tell the difference between a human and a computer program but could she fool people into thinking she was a real person?
After a discussion about the Loebner Prize by Brian Christian (a human confederate in the Loebner Prize) with presenters Jad Abumrad and Robert Krulwich, a volunteer from the audience was asked to take part in a Turing Test. This involved simultaneously talking to both a human (Jad) and a computer program (Mitsuku) before trying to decide which was which. To avoid giving away any hints, the human and the machine were known as Strawberry and Blueberry.
After talking to both parties for around 3 minutes, the majority of the audience decided that Strawberry was Jad. However, the volunteer decided incorrectly that Mitsuku was the human and Jad was the computer! After learning the results, she announced, “I’ve definitely never had that much chemistry with something that WAS human”. Around 13% of the audience were also fooled into thinking Mitsuku was a real person.
This brought the experiment to the second stage. Each member of the studio audience was given a cell phone number which would connect them to either a human volunteer or Mitsuku. After 5 minutes, they had to decide what they were speaking to.
Once the results were announced, it was revealed that about 45% of people talking to Mitsuku believed she was a human. This smashes the traditional 30% target of the Turing Test!
The success of these two experiments is testament to the power of the AIML chatbot language and the Pandorabots platform. I just hope some of the studio audience are recruited as Loebner Prize judges this year, as Mitsuku defends her back to back wins!
However, these experiments make us think about the Turing Test and is it really necessary for a computer to pretend to be human? I’m sure back in Alan Turing’s day when he thought of the test, the pinnacle of intelligence was surely the human brain. As technology has developed over the last 60 years, computers now have the internet at their disposal and have amounts of disk storage and processing power that Alan Turing could only dream about.
One of my biggest frustrations about entering contest such as the Loebner Prize is having to “dumb down” my entry so it passes for a human. This involves putting deliberate spelling mistakes and changing many answers to be less accurate. For example:
Human: How high is Mount Everest?
Bot: 8,848 m (29,029 ft)
Human: No idea but I know it’s the tallest mountain
While the bot’s answer is more intelligent and useful, it’s certainly not humanlike which makes me question why we need machines to pretend to be human any more. Only the arrogance of humans would allow us to believe we are the most intelligent things in existence. Let’s use AI and chatbots as a useful tool rather than trying to deceive people. After all, we wouldn’t be happy if any other equipment we used was deliberately designed tro not function with its true potential.
Catch the finished Radiolab podcast in which they also discuss other aspects of AI here: https://www.wnycstudios.org/story/more-or-less-human
To bring the best bots to your business, check out www.pandorabots.com or contact us at firstname.lastname@example.org for more details.