Re: sign up security:

From: "Marti McCuller" <marti@agassa.com>
Sent: Monday, January 28, 2002 4:58 PM
Subject: Re: sign up security:


> So, thinking about the users who might be left out (if a .wav file is
> provided), I come up with two types, a deaf/blind using a Braille device.
> And someone using a text only 'terminal' connection that does not display
> images and does not have a sound card.  Does that cover it or can
> anyone think of others that might not be able to 'pass' the security test?

> What sort of suggestions does anyone have for a method that would
> include these two groups?

You have to look at the set of hard computing problems that humans find
easy, and the cost of solving them vs the value of having lots of robot
accounts.

Unfortunately the gap is narrowing and the areas where humans beat machines.
We still beat them at perception, but cannot really use that in this
scenario because we cannot rely on any given form of perception existing in
the subject.

Another area that is expensive to implement in a machine is world knowledge
and inference.  The problem here is that it is a hard problem for a computer
to be the interviewer as well as for a computer to be the interviewee.  This
is what makes the 'phone call' a compelling solution.  The test is
administered by a human, but because the human is costly to run, it is only
used in the minority of cases who cannot respond to the .png (say) or .wav
formats.  The test is valid but again we have an issue with the medium
because the phone requires hearing and speaking.  I suppose in that case an
email exchange probably would be the most accessible means of administering
the interview.

The interview method of course requires a human operator for the website's
end.  At this point I have no suggestions for an automated method.

Received on Monday, 28 January 2002 12:40:07 UTC