The Loebner Prize gives an award every year to what they think is the best chat bot. This is actually a really hard problem, and there are some assumptions bundled into it as to what an ideal AI chatbot would be.
Here’s the winner from 2013: Mitsuku (ignore the creepy manga girl drawing – yeesh)
If you play around with it for even a little, you can tell you’re not talking to a person.
What I wonder though is why that’s the goal. I mean, I get it. People are fascinated with creating a general purpose conversational AI. Clearly. There’s one in my book. But let’s say that someone builds that.
What good is it? Wow, now I can have the same inane chats with a computer that I could have with a real person. Yay.
The value in conversational AI is purely as an interface to other things. Maybe it’s a help database. Maybe it’s an imperturbable, wildly scale-able triage system for 911 or other emergency calls. Maybe it’s the interface to magical gun.
But once you start thinking about the needs of those individual systems, it turns out that you don’t necessarily need a fully capable AI that can hold generalized conversations. The linguistic and subject domains of each of these problems is much, much smaller than the general case.
My guess is that we’ll get virtual customer service reps that are mistakable for humans long before we get a generalized system, due to the magnitude of the conversational domains. Of course, if someone does build a true generalized system, it could be leveraged for everything. I just don’t think that’s going to happen first. And, if one (or more) groups build ones that solve for the monetizable use cases, it’s going to put a serious damped on further research into the general case.Buy Lincoln, Fox and the Bad Dog on Amazon.com right now, or get the first half for free right here if you're still on the fence (.epub download to read in iBooks, Google Play Books, etc.)