Comments on: The Latest Research From the Michael Scott Academy of Computer Science

By: Andrew Roth

Andrew Roth — Tue, 10 May 2011 21:30:40 +0000

Thy rod and thy staff, they comfort me, as the actress said to the bishop.

By: Tim C

Tim C — Tue, 03 May 2011 19:11:50 +0000

@Eric “Never”

BS. I was just in bathroom at work…one of the soap things is broken. I was in front of sink with working soap thing. Someone else says “excuse me” and I let him use that sink, saying “Oh you don’t want to use the broken one? Ridiculous!” He replied, “I just don’t feel like pumping it and pumping it only to have a dribble come out.” (I am not making this up.) The reply was far too easy, but as Carlin said, “F you, I think it’s hilarious.”

By: Rune

Rune — Mon, 02 May 2011 17:02:42 +0000

@jppatter

So the UK version of “that’s what she said” is “as the actress said to the bishop”??? The British can be so weird.

Well, only if they are the same age as Alfred Hitchcock. The quote says more about the writer than it does of Brits.

By: jb

jb — Sun, 01 May 2011 18:56:33 +0000

“I hope no tax dollars are going into this…”

That’s what she said!

By: Robert

Robert — Sun, 01 May 2011 17:12:14 +0000

I hope no tax dollars are going into this…

By: Bernard

Bernard — Sun, 01 May 2011 10:14:36 +0000

-r

By: Bernard

Bernard — Sun, 01 May 2011 10:14:21 +0000

It’s an irony that as clever programmers and advancers in artificial intelligence move computers ever closer to passing the Turing test there’s a concurrent process by which reality tv and text messaging language move the average human ever closer to failing it.

By: Eric

Eric — Sun, 01 May 2011 07:17:44 +0000

‘The pair describe the “TWSS problem” as recognising when it is funny to follow a sentence with “that’s what she said”’

The answer is always ‘never.’

By: bbartlog

bbartlog — Sun, 01 May 2011 00:10:28 +0000

@Alex: I say 70% sucks because presumably 50% can be achieved by random guessing (if the test involves half ‘funny’ and half ‘non-funny’ sentences). So 70% is about 2/5 of the way to perfect. Also they describe their algorithm (weighting words for sexiness) and guess what, it doesn’t sound like the computer is parsing anything, it’s just scoring words by points.

‘I think a human wouldn’t do much better’

Some human must be able to do a hell of a lot better, else how would they score the test? For a human this isn’t rocket science, though as I noted above there would surely be disagreements as the task is slightly ambiguous.

‘Just about the only type left that computers still aren’t better or just as good at.’

They still are bad at language translation, most kinds of poker, Go (last I checked), speech recognition and some other things. Certainly they’re getting there! But the thing is, this piece of work is not actually an advance in AI. Word sexiness scores? Not even going to parse or use bigrams or attempt to extract meaning? It’s only good as a press release.

By: parse

parse — Sun, 01 May 2011 00:05:46 +0000

The system turned out to be around 70% accurate, but the pair say this is deceptively low because much of the training data did not consist of TWSS jokes, and with a more even data set it could achieve 99.5% precision.

What criteria did they use to determine whether the computer was correct? As bbartlog suggested, I don’t think there’s an objective criteria that determines whether a particular line is made humorous by adding a TWSS. When people are in the right mood, adding TWSS after anything at all can provoke laughter.

By: jppatter

jppatter — Sat, 30 Apr 2011 23:29:24 +0000

So the UK version of “that’s what she said” is “as the actress said to the bishop”??? The British can be so weird.

By: Elliot

Elliot — Sat, 30 Apr 2011 21:31:27 +0000

Curse you, kriznol!

By: Elliot

Elliot — Sat, 30 Apr 2011 21:29:06 +0000

“This is huge.”

TWSS

By: Alex

Alex — Sat, 30 Apr 2011 19:33:09 +0000

“70% success is not a sign of much AI going on, either. It’s an interesting project but the results are underwhelming.”

Why do you say that? I think a human wouldn’t do much better, and this sort of thinking is just about the only type left that computers still aren’t better or just as good at.

By: B

Sat, 30 Apr 2011 18:13:39 +0000

Well, I think this is pretty impressive, but I can barely program my thermostat…

By: Joe

Joe — Sat, 30 Apr 2011 17:35:44 +0000

Talking about jokes: LA's finest in action! Hey wait, that is real? I am sure Dunphy and Patterico can explain it all.

By: bbartlog

bbartlog — Sat, 30 Apr 2011 16:58:25 +0000

‘…with a more even data set it could achieve 99.5% precision.’
Making a claim like this without actually having done it is total horseshit. You probably couldn’t even get two people to achieve 99.5% agreement on whether a given sentence would or would not be an actual joke when ‘…that’s what she said’ is appended.
70% success is not a sign of much AI going on, either. It’s an interesting project but the results are underwhelming.

By: qwints

qwints — Sat, 30 Apr 2011 16:41:58 +0000

As amazing as this research is, I can’t help being paranoid about its potential for repression. Imagine a system analyzing sentences for a ‘disloyalness’ function allowing the government to actually spy on every e-mail or phone call.

By: Nipplemancer

Nipplemancer — Sat, 30 Apr 2011 16:06:44 +0000

With the automation of dick jokes I will become obsolete. :(

By: James J.B.

James J.B. — Sat, 30 Apr 2011 15:58:21 +0000

Finally a scientific initiative that I can get behind