It's hard to write about virtual humans without talking about the Turing Test, as we do in Chapter 2. Whilst it is no longer seen as a valid test of "intelligence" it is still a reasonable test of the natural language and conversational ability of chatbot programmes.
Here are a few relevant links which we'll keep adding to:
- Turing's original paper about "The Imitation Game"
- Our 2009 paper critiquing the Turing Test in terms of not providing a level playing field (typically the bot is the only one taking part who doesn't know it's a test), and some suggested variations in the form of covert and group tests.
- A 2014 paper describing a covert Turing Test inspired by our paper, which got a 78% deception rate
- Our 2016 paper describing a group covert Turing Test which managed a 100% deception rate
- A good summary book on the Turing Test
- A Turing Test 50 Years On paper
- An interesting paper on chatbots "taking the 5th" in the Turing Test - a great strategy
- A review of the Loebner Prize, and annual staging of something close to the Turing Test, and another one here.
It should also be noted that the term "Turing-capable" implies a system which could pass the Turing Test (i.e. fool a human), and that the term can be applied to other aspects of a virtual human (e.g. looks, emotional response, movement) quite apart from the natural language. In fact the "Gold" version of the Loebner prize looks at a Skype call type situation rather than a text-chat one.