Wednesday, September 24th, 2003
I’m looking into adding dotcomments for this blog. Let’s see if it works.
![]() |
blog |
I’m looking into adding dotcomments for this blog. Let’s see if it works.
today’s thoughts are of self-maintaining websites. these would be things like messageboards or online communities that effectively run themselves without any real mediation or oversight. imagine that you code a websites in which no content is really ever static. one example would have been / was the now defunct indiecred.com site which I had put up. althought i wouldn’t say that that was really self-sufficient since no one ever used it…
and of course we’re not saying that we wouldn’t be putting any time into the site. let’s say that the maximum requirement in human time would be about 5 hours a month for it to still be considered self-sufficient.
It is an exciting idea. I wonder if any books or sites have been dedicated to it…or if i have the time to think about them.
this weekend i’m going away, but we’re going to practice again on monday. should be good.
so tonight was our third practice. we’ve got a new song. it’s pretty good. i think it will work very nicely once we get the verses all mapped out and keyboard parts written. i’m excited about us recording a demo version of it quickly so we can get it out onto the internet.
So we got together last night. We jammed for a while. Nothing like what I wanted yet. It’s going to take some time. We’re trying a new band model where I just sing and don’t play guitar. It was our first time hanging out with our new drummer Steve.
Went to a lecture on CAPTCHAs. Very interesting. Seems to me there ought to be some sort of “language-based” CAPTCHA that exploits computers not being able to parse language. If we could find a way of having them none the less generate a question which puts a human’s understanding of grammar to the test, then we should be set. Obviously we can’t have them generate grammatical questions (because if they could do that, then they could solve the puzzle, too), but maybe by having some sort of one-question standard Turing test in which a human is asked to choose which of ten sentences was written by a human, and which by a bot, the test could be doable.
One idea would be to submit a human sentence to 10 or so “conversation bots,” then ask the human to decide between another human sentence and the output of the bots as to “which was written by a human.”
Let’s give this a shot.
I’m going to start with, “The dog crawled up my leg and nested in my neck.”
Here are the bots responses:
ALICE: “Tell me more.”
LinguBot(TM): “I hope you will tell me how you are doing today. ”
Ella: “Is that a fact.”
John Lennon Artificial Intelligence Bot: “I hear you.”
Now these response aren’t too good. I bet if you had to choose between:
“I loved eating cotton candy.”
“Tell me more.”
“I hope you will tell me how you are doing today.”
“Is that a fact.”
“I hear you.”
You would choose the first sentence as being the most human. So? Could a computer be designed to make that distinction. Not yet I don’t think. There is one thing, of course. The cotton candy reference is far more specific than anything in a Bot’s output. A script could just take the probability of a specific word being used and use that to make the selection. This is probably the sort of CAPTCHA that would require you to perform the test three times. Other wise there is a chance that in one question a Bot could fool you.
REFERENCES
An idiolect (sometimes misspelled: ideolect) is a variety of a language unique to an individual. It is manifested by patterns of word selection and grammar, or words, phrases or idioms that are unique to that individual. (That is, every individual has an idiolect; the grouping of words and phrases is unique, rather than individuals using specific words that nobody else uses.) An idiolect can easily evolve into an ecolect - a dialect variant specific to a household.