As young students studying the basics of cognitive science, we are often introduced to artificial intelligence paradigms and developments. At the time of my schooling simpler chatbot AI algorithms such as the one used for ELIZA were available. Though AI is so much more than a talking avatar, in fact most of the machines we use today have some AI features; the AI-humanoids were the ones that attracted my attention the most.
From R2D2 in Star Wars and Gerty in Moon to David in Prometheus, a cognition which we program, has always fascinated us one way or another and is an important feature in any scifi story, fantasy saga or visualisation of the future.
When first doing the Turing test in 2004, the “who was non-human” was clearly obvious. The language of the chatbot was simple and the algorithm wasn’t as sophisticated, hence it took one philosophical question and you’d know if it was a computer or a human making the remarks. My first try was playing around with ELIZA. Today, software such as the Cleverbot score a 59.3% human on the ratings, compared to 63% scored by actual humans. That is enormously good as results above 50% rating is considered a success.
Though some clues emerge, there is always a sense of doubt.
"I often ask myself if it is because the humanoids are becoming more elegantly programmed to learn smarter or if humans are just simply getting, let’s just say, less elegantly programmed when we utter ourselves on the internet."
Most of us have seen the popular video of two chatbots talking to each other on youtube.
What struck me was how two algorithms produced a more intelligent conversation than most humans I’ve heard converse out in public. What a leap from poor old ELIZA. If you haven’t talked to her, please do. You will find statements such as “Tell me more” and “Why do you think there is moon tonight” and let’s not forget the classic “We were talking about you, not me” every time you try to make it personal. So, in sciphi lab spirit, the first experiment to reach this level of conversation was performed (with some modification to original procedure).
A simple chatbot called Evie was set up to talk to another Evie. Two questions were chosen "what is time" and "what is intelligence" asked in two separate sessions. One Evie's answer was typed in to the other ones chat window, eliminating human interaction out of the equation of “intelligence”. We could just let them say hello to each other, but that has already been explored.
They were allowed to chat for 5 minutes and no editing is done to the video to disclose the entire procedure, hence the slow pace.
The algoritms were good. They learn from us, and with one million conversations logged, their verbal range is wider than many other chatbots. Until they started counting....(then they loose much credibility). In conclusion, I don't think we need to worry about Terminators roaming the planet keeping us as slaves quite yet. But if they do, perhaps the conversations will be interesting. //AOT
AND OTHER CHATBOTS ON THE RISE....