Add a Comment
Email This
Statistics
Another related concept I forgot to mention earlier was the Turing Test. It's basically a test for a machine to demonstrate intelligence. From what I heard from the podcast I originally posted, some chatterbots are coming pretty close.
Well, the Turing Test is an interesting thought experiment, but rather interesting and unexpected things happen if you try it out in practice. There've been ongoing competitions where they really do set up the scenario Turing outlined - though it's usually set up on a "restricted domain".
Basically, the conversations are required to center on some particular topic area, chosen by the test subject. That allows the programs to be rather hyper-specialized, like having lots of information about, say, Shakespeare's plays. Or, in an example more relevant to the the topic, it could pretend to be a young woman expecting to be chatted up on the 'net.
The "test" then typically has a couple of dozen terminals. Some of them are connected to programs, some are connected to real people, who've chosen their own domain. A group of testers then gets a chance to interact with each of the terminals, and then votes whether they think it's a person or a program on the other end.
One thing that I think is nearly as interesting as the fact that programs are getting better and better at passing for humans is that there are almost always human subjects who the testers think are programs. So, there's a secondary "test" that happens right along with the Turing test. While the straightforward test is seeing whether programs can pass for humans, the testers are being tested for their ability to recognize other humans in the first place - and it seems they're not really very good at it.
On researcher specifically designed a bot that hung out in online chat rooms and pretended to be a somewhat flirtatious young woman. He would then examine the chat logs to see how well she was doing at fooling people. One young man spent something like thirty hours over the course of several weeks trying to get her to go out with him. So, does that say that the program passes the Turing test, or that her suitor failed it?
RSS
RE: Humans driven out of the service industry
I mentioned some places where it is already happening, but what I meant were some more non-obvious places where Chatterbot tech could increase the penetration--things that require better interaction than you get in the run-of-the-mill ATM machine or self-checkout aisle. Your synthesizer example was also a good one. I remember a service a few years ago, and I don't know if it is still extant or not, but you called a number and asked a question in natural language and voice recognition software and something like chatterbot tech would formulate an appropriate response and speak back the answer to you. Think more sophisticated than the automated message about your balance from your credit card company. Ask "How do I get to the airport" and that sort of thing. A lot of customer service and tech support stuff could be handled this way using basically the FAQ with a little vocal query capabilities. The realism is the 'bot's ability to understand what the person really wants and deliver the appropriate information in the correct context. Another aspect of realism is giving the person on the other end the perception that they are really talking to a person--a lot of these people simply want to vent and get some sympathy.
Another related concept I forgot to mention earlier was the Turing Test. It's basically a test for a machine to demonstrate intelligence. From what I heard from the podcast I originally posted, some chatterbots are coming pretty close.
View Full Discussion