Re: Not reputation again! (Was: The Nature of the Cypherpunks List)

At 6:45 AM 9/25/96, Greg Burk wrote:

>I know, I said I was through arguing "reputation theory" and I really
>thought I was, but this latest impelled me to say more.
>
>tcmay@got.net (Timothy C. May) writes:
>> --Tim May, whose reputation is, like the list, whatever it is, for whatever
>> reasons, and who thus needs no defenders
>
>Indeed, your reputation needs no defenders among those of us who know
>you directly (if electronically)
>
>But here's something for you to think over: If there are such things as
>negative reputations, why hasn't your reputation zoomed to godlike
>status upon the ... let us say unreserved... condemnations of the
>Poster With Nothing Better To Do, if s/he has a negative reputation (I
>presume you would say so)

First off, I certainly don't know what the precise "calculus of
reputations" looks like. That is, how it adds, subtracts, how the reps of
others factor in, etc.

I think studying how it might work--and how it compares to other fields,
such as movie reviews, music reviews, book reviews, etc.--might be an
excellent Masters-level thesis for someone in sociology or even CS (if done
with the proper slant).

However, I would never think the calculus is something so simplistic as: "I
rate Alice's reputation as very negative. Alice just said some bad things
about Bob. Therefore, Bob's reputation will go up a great deal."

More likely, something like: "Alice just said a bunch of things bad about
Bob. Cool. Maybe this Bob person deserves a look."

In other words, smaller steps, with diminishing returns. And with few
sudden movements, except by direct judgment.

>Or you could contend that there's no such thing as collective
>reputation, but I think there are two major flaws: His/her 1-to-1
>"reputations" clearly add up to a collective consensus among us annoyed
>cpunk readers. Even if you contend that reputations in general do not
>behave collectively, it is no defense in this case at least. And I think
>if "reputation theory" predicts no collective behavior, it must be
>pretty weak.

I believe that, to first order, reputation is a tensor. Given N people,
imagine this matrix:

            Alice  Bob   Charles  Daphne  Earl  Fiona  Gloria  Harold

Alice       .99    -.21    .75     .94    .94    .83    -.03    .22

Bob         .72     .96   -.02     .85    .71   -.60     .10    .32

Charles     .82     .02    .97     .90   -.50    .42    -.10    .70

Daphne      .45     .87    .23     .92    .74    .87     .11    .23

Earl        .89     .54    .34     .90    .95    .23     .23    .46

Fiona       .87     .50    .32     .68   -.34    .97     .78   -.15

Gloria      .59     .78   -.23     .15    .29   -.30    -.80    .51

Harold      .65     .03    .34     .78    .51   -.76    -.51    .97


This can be read thusly: Alice has a .99 rating of her own self, a negative
.21 rating of Bob (or Bob's opinions, or his posts, or whatever is being
rated), and so on. Bob has a .72 rating of Alice, a .96 rating of himself,
etc. "Alice(Bob) = -.21"

Now in this example I made up, some various observations can be made.
Nearly everyone rates Harold pretty low, except for Harold. Thus, Harold's
own opinions of others, if expressed, probably won't change too many other
opinions.

Nearly everyone rates Daphne very highly, and her opinions are read
carefully. However, Gloria does not rate Daphne highly...but then Gloria
rates her own stuff a negative .80, so Gloria has some psychological issues
to deal with and others typically rate her pretty low.

And so on.

In the real world, I think we can see how such a matrix could be
constructed, based on either direct inputs (votes) from people, or based on
their apparently positive or negative comments, etc. (For example, it is
fairly obvious that I might give Hal Finney a rating of .90, and Vulis a
-.90. And so on, for others.)

Now is there a "collective reputation"? There are various additive
properties, with easily understandable meanings. (If Harold is in a lot of
kill files, this says a lot, for example.)

Anyway, as I said before, the "calculus of reputations" is not worked out,
so far as I know. Some weeks back I suggested that the mathematics of
belief as developed in "Dempster-Shafer belief theory" has some nice
properties that make it seem a promising area to look into.


....
>Because s/he can *never spend reputation down to zero*.

So you say. I see no reason reputations cannot be negative, in the sense
that not only do I take the opinions of such a person very seriously, I
tend in fact to believe the opposite opinion is more likely. This is a
"negative " reputation. Thus, a reputation can be "spent down" to zero, and
below.

>Frankly, I think you should just admit that your reputation theory is
>flawed and rethink it. I would be interested in hearing it, but what you
>have now is IMHO badly flawed.


--Tim May

We got computers, we're tapping phone lines, I know that that ain't allowed.
---------:---------:---------:---------:---------:---------:---------:----
Timothy C. May              | Crypto Anarchy: encryption, digital money,
tcmay@got.net  408-728-0152 | anonymous networks, digital pseudonyms, zero
W.A.S.T.E.: Corralitos, CA  | knowledge, reputations, information markets,
Higher Power: 2^1,257,787-1 | black markets, collapse of governments.
"National borders aren't even speed bumps on the information superhighway."

-----BEGIN PGP SIGNED MESSAGE-----

It's silly to conflate ratings of people with ratings of
people's opinions with ratings of people's ratings of etc.


Some of the people I work with have awful opinions on politics,
other people, music, movies, books, and food, but great opinions
on computer hardware, algorithms and programming languages.


But their ratings as people are generally very high.


(Insert inaccurate but amusing joke to the effect that if 
someone has good taste in restaurants they are likely to have 
poor taste in video cards, or some such.)


Check out Firefly (www.ffly.com).  (Cpunk comment:  they only
track your ratings by nym, no True Name required.)


Firefly has the "if I agreed with Bob a lot in the past on this
subject, I'll probably agree with him again" heuristic.  It
appears to be a pretty ugly heuristic when applied to me.  Ffly
keeps trying to suggest music which is in the same _genre_ as 
music I have previously recommended, but what I am interested is
music that is similarly _good_.  My tastes don't correspond well
to genres.


But this is a mere anecdotal impression.  Give it a spin.  


I wrote in the suggestion box that they add meta-ratings
(especially ratings on the little "movie reviews" submitted by
other FFly users).  Some guy wrote back (same day!) that they
were planning on it.


Regards,

Bryce




-----BEGIN PGP SIGNATURE-----
Version: 2.6.2i
Comment: Auto-signed under Unix with 'BAP' Easy-PGP v1.1b2

iQB1AwUBMkkjQkjbHy8sKZitAQF5wAL/WgD5iPOR7RtCc5SNgLMGqKUJqHLiS7bM
XItX4p8Z1Uri5DZgAUoOCDZ7iEoP3rleDBThGq3rFG8y6I9tbFueC00TRwoft5rb
fR7pqeaDj2AzcSt4rYCi67ucsW4wQi1C
=rL1w
-----END PGP SIGNATURE-----

On Wed, 25 Sep 1996 bryce@digicash.com wrote:

> 
> -----BEGIN PGP SIGNED MESSAGE-----
> 
> It's silly to conflate ratings of people with ratings of
> people's opinions with ratings of people's ratings of etc.
> 
> 
> Some of the people I work with have awful opinions on politics,
> other people, music, movies, books, and food, but great opinions
> on computer hardware, algorithms and programming languages.
> 
> 
> But their ratings as people are generally very high.
> 
> 
> (Insert inaccurate but amusing joke to the effect that if 
> someone has good taste in restaurants they are likely to have 
> poor taste in video cards, or some such.)
> 
> 
> Check out Firefly (www.ffly.com).  (Cpunk comment:  they only
> track your ratings by nym, no True Name required.)
> 
> 
> Firefly has the "if I agreed with Bob a lot in the past on this
> subject, I'll probably agree with him again" heuristic.  It
> appears to be a pretty ugly heuristic when applied to me.  Ffly
> keeps trying to suggest music which is in the same _genre_ as 
> music I have previously recommended, but what I am interested is
> music that is similarly _good_.  My tastes don't correspond well
> to genres.
> 

This is incorrect.

Firefly actually looks at what you have recommended and then finds others
who have recommended or favor the same songs/albums/groups and selects new
groups albums or songs based on the number of people who agree with your
recommendations who suggest them.

i.e. if you give the eagles a 5 out of 5, and 20 others also gave the
eagles high ratings, the system weeds out what other groups those 20 gave
high ratings to and suggests them to you.  As you confirm or deny your
interest in those groups, the system narrows it down even more, using
those choices to find others who seem to have similar tastes to you and
suggesting albums and groups not yet in your file which have been
recommended by those with tastes like yours.

This trancends genres.  It's quite a nice system.

It also makes an important point about reputations.  Using a catagory
reputation scale is, as the author above describes, not very efficient.

> 
> But this is a mere anecdotal impression.  Give it a spin.  
> 
> 
> I wrote in the suggestion box that they add meta-ratings
> (especially ratings on the little "movie reviews" submitted by
> other FFly users).  Some guy wrote back (same day!) that they
> were planning on it.
> 
> 
> Regards,
> 
> Bryce
> 
> 

> 
> -----BEGIN PGP SIGNATURE-----
> Version: 2.6.2i
> Comment: Auto-signed under Unix with 'BAP' Easy-PGP v1.1b2
> 
> iQB1AwUBMkkjQkjbHy8sKZitAQF5wAL/WgD5iPOR7RtCc5SNgLMGqKUJqHLiS7bM
> XItX4p8Z1Uri5DZgAUoOCDZ7iEoP3rleDBThGq3rFG8y6I9tbFueC00TRwoft5rb
> fR7pqeaDj2AzcSt4rYCi67ucsW4wQi1C
> =rL1w
> -----END PGP SIGNATURE-----
> 

--
I hate lightning - finger for public key - Vote Monarchist
unicorn@schloss.li