How It Works
The system depicted was created by combining three components: a chatbot, a text-to-speech synthesizer, and an avatar renderer. Chatbots are machines designed to emulate the conversational abilities of humans, conversing with a human user and generally attempting to convince the user into thinking that the machine is human. In such a scenario, if a sufficiently adept human on one end is fooled into thinking the machine is another human, the machine would be credited as passing the famous Turing Test for intelligence. Over 60 years after its proposal by Alan Turing, there are arguably still no machines capable of passing this test. The chatbot we initially used was Eliza, a prominent early milestone from AI's infancy in the 1960's. This tended to produce fairly boring conversations, so we switched to a much smarter, constantly learning chatbot: Cleverbot. Publicly available on Cleverbot.com, this state of the art chat engine was created by AI researcher Rollo Carpenter, who can be contacted via his company, Existor. Cleverbot will continue to learn, and Existor are soon to add new capabilities aiming at a Turing Test pass sooner than you might expect. The second piece of the system is the text-to-speech synthesizer, which takes the text generated by the chatbot and creates a spoken, audio version. There are many services able to accomplish this; we chose Acapela because it was easy to use and sounded decent.
The final piece is the avatar renderer, which synthesizes an animated character whose gestures and lips are synced to the sound stream. For this we used Living Actor Presenter.
We tied these three components together in Python, producing a single machine (one of the two screens) that can converse with a user. We then plugged the output of one machine into the input of a second, and the output of the second back into the first, producing endless comic robotic entertainment.