1996-05-22 - Re: Senator, your public key please?

Header Data

From: “E. ALLEN SMITH” <EALLENSMITH@ocelot.Rutgers.EDU>
To: tcmay@got.net
Message Hash: 0287f69679c20f4970fbe182821316fc3fe7e87fd096ff7cb082c261d4827b5c
Message ID: <01I4ZLQFIGAM8Y5IL9@mbcl.rutgers.edu>
Reply To: N/A
UTC Datetime: 1996-05-22 11:22:28 UTC
Raw Date: Wed, 22 May 1996 19:22:28 +0800

Raw message

From: "E. ALLEN SMITH" <EALLENSMITH@ocelot.Rutgers.EDU>
Date: Wed, 22 May 1996 19:22:28 +0800
To: tcmay@got.net
Subject: Re: Senator, your public key please?
Message-ID: <01I4ZLQFIGAM8Y5IL9@mbcl.rutgers.edu>
MIME-Version: 1.0
Content-Type: text/plain


From:	IN%"tcmay@got.net" 21-MAY-1996 04:37:46.63

>(ObCaveat: I personally think a free society cannot/should not outlaw
>discrimination in any form, save that by government.)

	Agreed.

>By the way, this issue has some echoes of another technogical issue: the
>use of neural nets for loan approval software. Turns out that when a bunch
>of things are entered into a NN loan package, including the all-important
>default rate, the applicant's age, income, race, sex, education, employment
					    ^^^^
>history, credit history, etc., that NN loan packages "end up" rejecting
>many black applicants, more so than white or Asian applicants. (The NN
>"concluded" that blacks were higher risks for default than whites/Asians.)

>Even if no human being ever entered his or her biases and prejudices, the
>NN spit out this result.

>I recall there being talk about requiring "equality of outcomes," and that
>such NNs might have to have deliberately-biased inputs fed in, but I don't
>know what ever happened to this issue.

	The obvious solution in the above case is not to feed in the race
information for governmental decision-making. I recall a similar issue when
NN's were used for college admissions... while they didn't feed in the
applicant's race, they did feed in the applicant's name, and the length of it
turned out to have some correlations. (Why they were feeding in the name in
the first place is another question entirely, although NN's and GA's tend to
work the best when you don't understand the system at hand - which also tends
to mean that you shouldn't try to interpret what information is necessary.)

>In any case, I think this sort of issue, and the semi-related issue of
>"discrimination via key signatures," to be likely important issues in the
>courts in the coming years.

	It's related to that of the outlawing of IQ testing because of (I'd
say for environmental reasons) blacks having lower IQ scores on average. The
civil rights crowd are arguing for the PC egalitarian viewpoint that everyone
_should_ be the exact same in capabilities (for some absurd view of
"fairness"), and so want people treated the exact same despite obvious
differences and the validity of such tests.
	-Allen





Thread