Re: Language in the meat machine

Freeman Craig Presson (dhr@iname.com)
Sat, 17 Jul 1999 10:24:02 -0500

Really, this sounds like a brute-force attempt to do something that has been part of Lenat's Cyc for a long time. Check out the paper on contexts at http://www.cyc.com/public.html.

Great pithy quote from the Cyc site:

"When someone searches for 'Bolivia' on the Web, Cyc knows not to offer a follow-up question like 'Where can I get free Bolivia online?'"

It's funny how the most interesting thing about IEA (the grading program) is the schemes to beat it (OK, that's not really true, it's interesting to see how much a brute-force attempt can do, but it's not the AI breakthrough that that august journal, USA Today, thinks it is).

If the student knows there is little chance a human will check the essays, then he could try submitting one of the papers from the database. Of course, if a perfect match might be suspicious, then a concatenation of parts of two or three papers would be a better travesty.

Now off to finish reading the Cyc papers ...

On 16 Jul 99, at 14:11, Darin Sunley wrote:

>
>
> hal@finney.org wrote:
>
> > This has an obvious flaw. The professor is checking for style, syntax,
> > and logic. If students know that this program is going to check their
> > papers, all they need to do is to write something that produces the
> > right words in roughly the right positions, without regard to sentence
> > structure.
>
> Apparently they've tried this. They said most "travesties" failed because
> the mix of "apparently unrelated" but actually important words
> (prepositions, linkers, grammatical elements) was off. The programmer
> speculated that to generate a good set of keywords, you seem to have to
> write a full proper essay.
>
> Darin Sunley
> rsunley@escape.ca
>
>