Re:Show HN: I'm a 14 year-old dev who has programmed an arti...
by mst on Thursday April 16, 2015 @11:43AM
(#9388726)
Isn't machine learning a first step towards "how to train itself" ?
Honestly, I would say "chatbot" is a class of primitive AI - I'd also class Eliza as an (extremely) primitive AI.
I feel like you're making a distinction without a difference because you have some arbitrary barrier in mind for "sufficiently non-primitive to have earned the term AI" and are then applying that.
Would "a chatbot with primitive AI capabilities" seem like a reasonable compromise definition?
Re 1. Yes, it does attempt to pass the Turing test.
Re 3. Yes, that includes interacting with a human in a similar way that a human would with another human, that's part of the Turing test.
My own experience was: it asked me how I was, I told it I had a hangover, it thought that was great and that I was 100% happy. Which wasn't a long test. Obviously your own experience might have varied.
Re: 3. Isn't machine learning just machine learning and NLP just NLP?
I personally don't feel like my app suddenly becomes 'AI' if I import nltk and start tokenising some corpus to train it with, or load opencv and tell my drone to avoid the big red thing. If the app would work out how to /train itself/ I would consider it an AI.
Eg, the drone having the objective to follow me at a safe distance for as long as possible, then working out that the big red thing is dangerous and avoid it.
Or the chat both to, say, have an objective of beating the human at the turing test for a certain amount of time, then working out how it gives itself away, then avoiding doing that.
I acknowledge that's just a feeling though, and many definitions online would qualify 'thing that loads library and needs human training' as AI.
However I can't help but think that's going to promote the Reading University 'captain cyborg' school of cranks.