Why PICS is the wrong approach
Vladimir Z. Nuri
Why TCM is wrong about PICS being wrong:
>PICS is the wrong approach becuase it oversimplifies the ratings of
>content, because it places the ratings made by the author in the payload
>itself, and because third-party ratings systems are cut out of the loop
>(effectively).
bzzzzzzt. please read about it. there are multiple protocols. some
of them allow third-party rating services. some of them support
ratings within pages. the standard is neutral.
>One computerish way to think of this is that the "binding" is too early. At
>the time of distribution, say, I mark my work something with some PICS
>label, based upon my best understanding of the PICS labels, ratings,
>agencies, and laws. But once set, the "binding" has been made. Later
>reviews or reviews by other entities cannot affect the binding, at least
>not for this distributed instance.
you have a good point, but PICS is about letting the net decide. it
supports both self-rated and third-party ratings. we will see whether
one eclipses the other in the long term. personally I suspect both
will coexist.
>And of course it is quite likely that things important to others in their
>ratings are not as important to me. I might even ignore certain points, not
>even seeing the need to point out things in the work. This is inevitable,
>as there is no uniform view of truth, no uniform set of values and
>priorities, and no hope there ever can be such a monistic view.
this is a ridiculous misunderstanding of the rating system concept.
the PICS standard expressly supports diversity by letting a thousand
rating services bloom, to borrow a phrase from your own book. some
rating services may claim to be canonical, but you don't have to
believe them. there will be competition of rating services for a long
time into the future. this has already happened with all the filtering
software out there.
also consider the new Firefly system that doesn't
actually have fixed ratings on objects, but in which ratings are
determined dynamically based on your own personal ratings of pages.
Consider
>the recent example of AOL's lists of banned words, even words in "harmless
>situations" (e.g, the example someone cited of "tits" being banned, despite
>being the name of a bird...would an animal-lovers Web page or posting with
>"Tits and Asses!!!" prominently in the title be PICS labelled as obscene?
>Some would surely think so.).
this would be an example of the most rudimentary and simplistic filtering
or rating service, which of course the market would generally ignore
in favor of more sophisticated alternative schemes.
>A much better solution is to let the unique ID block of an article--the
>Usenet article ID, or some hash of the headers, whatever--be a pointer that
>other ratings servies could then use to provide for their customers or
>clients as a filtering mechanism. This would allow as many ratings services
>to exist as clients would be willing to support.
that's exactly what PICS is about when you read about it more deeply.
>More importantly, the "payload" does not carry some particular set of
>fairly-arbitrary PICS evluations. Binding by the censors instead of by the
>originator, which is as it should be.
PICS supports both, as it was expressly designed to.
what Timmy is repeatedly failing to comprehend despite much
evidence staring him in the face is that ratings services are
going to be a very significant new information industry, if they
haven't already become one. there are now many different filtering
packages out there and the market is large for them, as has
been proven by *existing* sales. this industry will grow. yahoo
and many other indexing services are in fact implicitly
rating systems, because they utilize editorial discrimination in
deciding who to include and who to exclude. they just don't say,
"this is rated yahoo approved" overtly.
(timmy is also upset that a massive new industry is growing without
his personal approval or anticipation. I will amuse myself by
counting the days until he does a flip in position and begins
to advocate rating system's efficacy while pretending his position
was never otherwise)
let a thousand rating systems bloom. PICS is about finding good
content as much as rejecting uninteresting content.
Brian A. LaMacchia
At 01:11 PM 12/11/96 -0800, you wrote:
>
>PICS is the wrong approach becuase it oversimplifies the ratings of
>content, because it places the ratings made by the author in the payload
>itself, and because third-party ratings systems are cut out of the loop
>(effectively).
Um, first, there's nothing in the PICS spec that requires ratings to be
embedded with the content. PICS labels can be distributed one of three ways:
1) embedded in the content, assuming there's a method of embedding defined
for the particular type of content. (For HTML, you can put it in a META tag.)
2) sent "along-with" the content as part of the transmission protocol.
(For HTTP, there's a standard by which PICS labels can be sent as RFC-822
headers in the HTTP reply, but no one is using that to the best of my
knowledge.)
3) distributed from the third-party label bureau.
Method of distribution is independent of author of the label, too. It's
perfectly reasonable for the author to distribute labels for his content
via a third-part label bureau, or for an author to embed a label from the
GoodMouseClicking page rating service within his document.
By the way, PICS labels nominally apply to a document named by a particular
URL. You can elide the URL, I think, if the label is sent along with the
content or embedded within the content, but when you ask a label bureau for
labels you request them by URL.
Second, I don't see how PICS oversimplifies content rating. If anything, I
would expect complaints that PICS complicates things too much because there
can be an infinite number of rating systems and a human is forced into the
loop to read and evaluate the meaning of any particular rating system. It
is true that PICS currently only permits rating values to be numbers, but
enough PICS users need non-numeric values that I expect this to be changed
in January at the PICS WG meeting.
>One computerish way to think of this is that the "binding" is too early. At
>the time of distribution, say, I mark my work something with some PICS
>label, based upon my best understanding of the PICS labels, ratings,
>agencies, and laws. But once set, the "binding" has been made. Later
>reviews or reviews by other entities cannot affect the binding, at least
>not for this distributed instance.
It's true that reviewer B cannot affect reviewer A's labels, but B can make
statements about the quality of A's labels, and I can choose (in a more
general trust management environment) to accept labels only from label
authors who are vouched for by some particular vouching service. So
GoodMouseClicking might say I'm a reliable rater of content in "Bal's
crypto-relevant rating service" but a lousy judge of pages for "Joe's cool
jazz pages rating service."
>More importantly, the "payload" does not carry some particular set of
>fairly-arbitrary PICS evluations. Binding by the censors instead of by the
>originator, which is as it should be.
The Feds obviously believe in "encouraging" self-rating as a means of
defending yourself when they haul you into court, but in general I think
people will tend to defer trust to particular third-party ratings services
that they choose over an author's self-labels. After all, if I'm looking
for movie reviews the last thing I read is the self-promotion put out by
the distributor; I look for a third-party I know who tends to agree with my
tastes. Same thing with product reviews (e.g. Consumer Reports) or book
reviews.
--bal
Dale Thorn
Vladimir Z. Nuri wrote:
> Why TCM is wrong about PICS being wrong:
> >content, because it places the ratings made by the author in the payload
> >itself, and because third-party ratings systems are cut out of the loop
> >(effectively).
> bzzzzzzt. please read about it. there are multiple protocols. some
> of them allow third-party rating services. some of them support
> ratings within pages. the standard is neutral.
> also consider the new Firefly system that doesn't
> actually have fixed ratings on objects, but in which ratings are
> determined dynamically based on your own personal ratings of pages.
If Firefly is an example of what PICS is or could become, the hell with
PICS. Firefly encourages and rewards group behavior and suppresses
individuality. Firefly would reward the discussion of the latest album
by a Columbia or Capitol artist, and discourage discussion of material
from independent (real independent) labels. I know because I've been
there and spent quite a bit of time trying to get a rating.
[remainder snipped]
Vladimir Z. Nuri
>If Firefly is an example of what PICS is or could become, the hell with
>PICS. Firefly encourages and rewards group behavior and suppresses
>individuality. Firefly would reward the discussion of the latest album
>by a Columbia or Capitol artist, and discourage discussion of material
>from independent (real independent) labels. I know because I've been
>there and spent quite a bit of time trying to get a rating.
>
this is absolutely ridiculous. the rating system is designed to
be incredibly individualized. it uses sophisticated statistical
techniques to find the correlations in your UNIQUE ratings given
over a set of items with other people's ratings, and weighs
future ratings based on these correlations. it may be even
dealing with anti-correlations. in fact what you have, in effect,
is a system with *no* hardwired ratings. the ratings space is
different for *every*single*person* who uses the service.
"you know"? get a clue, please. these systems are not at all like
the record rating systems you get in stores. I sympathize with
your plight however and agree that the record labeling system
is a lame way to go about it. all the more reason to support things
like firefly and grouplens (another interesting system that may
have inspired firefly, do a yahoo search to find it). superior
rating systems will begin to flourish instead of inferior ones...
Joseph M. Reagle Jr.
At 07:58 PM 12/11/96 -0500, you wrote:
> PICS is the wrong approach becuase it oversimplifies the ratings of
> content, because it places the ratings made by the author in the payload
> itself, and because third-party ratings systems are cut out of the loop
> (effectively).
Perhaps I don't understand what you are saying. I just want to ensure that
you understand that the PICS labels can be distributed in multiple ways.
(document, server, label bureau.) I suspect you do, and what you are
objecting to is that documnet-embedded labels will have a greater weight
than those distributed by third parties:
> agencies, and laws. But once set, the "binding" has been made. Later
> reviews or reviews by other entities cannot affect the binding, at least
> not for this distributed instance.
And consequently authors have a greater responsibility/liability than you
would like:
> More importantly, the "payload" does not carry some particular set of
> fairly-arbitrary PICS evluations. Binding by the censors instead of by the
> originator, which is as it should be.
In which case, I disagree. I think accurate, consistent, "objective" (I
know this is an argument on the other thread, I think one can get
relatively "objective ratings" see my RSAC case study for a break down on
the qualities of rating systems on my ecommerce page (home page below))
well branded and reputable agents will have a greater weight, and will have
a market motivation for accuracy exceeding regulatory pressure. (Plus,
there is nothing preventing thresh-hold tolerances for use with multiple
ratings.)
_______________________
Regards, Restlessness and discontent are the first necessities
of progress. -Thomas A. Edison
Joseph Reagle http://rpcp.mit.edu/~reagle/home.html
rea...@mit.edu E0 D5 B2 05 B6 12 DA 65 BE 4D E3 C1 6A 66 25 4E
E. Allen Smith
From: IN%"rea...@rpcp.mit.edu" "Joseph M. Reagle Jr." 12-DEC-1996 23:12:04.05
> > More importantly, the "payload" does not carry some particular set of
> > fairly-arbitrary PICS evluations. Binding by the censors instead of by the
> > originator, which is as it should be.
>In which case, I disagree. I think accurate, consistent, "objective" (I
>know this is an argument on the other thread, I think one can get
>relatively "objective ratings" see my RSAC case study for a break down on
>the qualities of rating systems on my ecommerce page (home page below))
>well branded and reputable agents will have a greater weight, and will have
>a market motivation for accuracy exceeding regulatory pressure. (Plus,
>there is nothing preventing thresh-hold tolerances for use with multiple
>ratings.)
Umm... I pointed out a while back the considerable problems with
the RSAC attempt at objective ratings. See
http://infinity.nus.sg/cypherpunks/dir.archive-96.05.09-96.05.15/0092.html
for a review of my objections. The system in question is obviously
much more subjective than, say, one that had:
Does this page contain any female frontal nudity?
Does this page contain any male frontal nudity?
Does this page contain any female rear nudity?
Does this page contain any male rear nudity?
and so on. The parts on violence are particularly subjective.
-Allen
Joseph M. Reagle Jr.
At 11:51 PM 12/12/96 EDT, E. Allen Smith wrote:
> Umm... I pointed out a while back the considerable problems with
>the RSAC attempt at objective ratings. See
I agree that it is not a purely descriptive system, however it is much
moreso than others. I thought the following was a useful breakdown for my
own purposes:
Caveat on Vocabulary
Before proceeding with the analysis we must first discuss some of the terms
used in the analysis. The usage of terms "objective" and "judgmental" can
be rather contentious. To address this, we disassociate any of these terms
with any pejorative meanings ( opinionated gut feelings about Web content
can be very useful ) and posit that there are three variables with which
content labeling systems can be considered:
descriptive/judgmental - does the label describe the content, or provide an
opinion about the "appropriateness" of the content.
deterministic/non-deterministic - is the previous process a deterministic
process, or is it "gut" based, and
voluntary, mandatory, or third party - does the author label his works
voluntarily, is he required to label his works by some other agency, or can
other services label his content.
No rating system we discuss is purely descriptive or deterministic. Rather,
each system varies with respect to where it falls between extremes.
E. Allen Smith
From: IN%"rea...@rpcp.mit.edu" "Joseph M. Reagle Jr." 13-DEC-1996 00:19:47.19
To: IN%"EALLE...@mbcl.rutgers.edu" "E. Allen Smith"
CC: IN%"cyphe...@toad.com"
Subj: RE: Why PICS is the wrong approach
Received: from RPCP.MIT.EDU by mbcl.rutgers.edu (PMDF #12194) id
<01ICXWWYI...@mbcl.rutgers.edu>; Fri, 13 Dec 1996 00:19 EDT
Received: from dialup-273.lcs.mit.edu by rpcp.mit.edu with SMTP id AA05830;
Fri, 13 Dec 1996 00:20:21 -0500
Date: Fri, 13 Dec 1996 00:17:54 -0500
From: "Joseph M. Reagle Jr." <rea...@rpcp.mit.edu>
Subject: RE: Why PICS is the wrong approach
To: "E. Allen Smith" <EALLE...@mbcl.rutgers.edu>
Cc: cyphe...@toad.com
Message-id: <3.0.32.19961213...@rpcp.mit.edu>
X-Envelope-to: EALLENSMITH
Content-type: text/plain; charset="us-ascii"
X-Sender: rea...@rpcp.mit.edu
X-Mailer: Windows Eudora Pro Version 3.0 (32)
Mime-Version: 1.0
At 11:51 PM 12/12/96 EDT, E. Allen Smith wrote:
> Umm... I pointed out a while back the considerable problems with
>the RSAC attempt at objective ratings. See
> I agree that it is not a purely descriptive system, however it is much
>moreso than others. I thought the following was a useful breakdown for my
>own purposes:
The below (from your essay) is a reasonable way to look at it. I
would tend to compare a system to obviously possible systems as well as to
simply what else is out there, however.
>descriptive/judgmental - does the label describe the content, or provide an
>opinion about the "appropriateness" of the content.
In this regard, I would call the system in question about midway
between a truly descriptive system and such obviously judgemental systems
as SafeSurf. This partially judgemental nature is probably unavoidable in
systems in which one uses a non-binary rating scheme; if the presence of
something inevietably means that the system gives a "high" rating, then
the system is judging that the something is more important than other
factors. In this case, the judgement is pretty obviously that they deem the
something in question (such as "hate speech" that calls for "harm" to some
class - although I doubt they'd include pro-Affirmative-Action speech in
that...) to be worse than the "lower" rated actions.
>deterministic/non-deterministic - is the previous process a deterministic
>process, or is it "gut" based, and
I would agree that RSAC's system is pretty deterministic; the
choices of what is labelled are rather arbitrary, but that falls under
the description vs judgement category above.
>voluntary, mandatory, or third party - does the author label his works
>voluntarily, is he required to label his works by some other agency, or can
>other services label his content.
As currently set up, RSAC is either voluntary or mandatory - a
government could require that it be used as a mandatory system, directly
or indirectly (e.g., under threat of lawsuits for "corrupting minors").
>No rating system we discuss is purely descriptive or deterministic. Rather,
>each system varies with respect to where it falls between extremes.
Agreed.
-Allen
Dale Thorn
Vladimir Z. Nuri wrote:
> >If Firefly is an example of what PICS is or could become, the hell with
> >PICS. Firefly encourages and rewards group behavior and suppresses
> >individuality. Firefly would reward the discussion of the latest album
> >by a Columbia or Capitol artist, and discourage discussion of material
> >from independent (real independent) labels. I know because I've been
> >there and spent quite a bit of time trying to get a rating.
> this is absolutely ridiculous. the rating system is designed to
> be incredibly individualized. it uses sophisticated statistical
> techniques to find the correlations in your UNIQUE ratings given
> over a set of items with other people's ratings, and weighs
> future ratings based on these correlations. it may be even
> dealing with anti-correlations. in fact what you have, in effect,
> is a system with *no* hardwired ratings. the ratings space is
> different for *every*single*person* who uses the service.
As I said, I spent considerable time there, and typed in considerable
data. I used the help facilities, and I corresponded with the persons
who do support. Firefly was not able to make a single correlation or
suggestion for me, which BTW, Alexander Chislenko acknowledged as a
weakness of the (current/old) Firefly system. Perhaps you should ask him
what the problem is, if you have a curiosity about it.
[snip]
Phillip M. Hallam-Baker
Dale Thorn <dth...@gte.net> wrote in article <58q40v$k...@life.ai.mit.edu>...
> Vladimir Z. Nuri wrote:
> If Firefly is an example of what PICS is or could become, the hell with
> PICS. Firefly encourages and rewards group behavior and suppresses
> individuality. Firefly would reward the discussion of the latest album
> by a Columbia or Capitol artist, and discourage discussion of material
> from independent (real independent) labels. I know because I've been
> there and spent quite a bit of time trying to get a rating.
I think there are two issues here, Firefly and PICS. Confusing one
with the other is a bad thing. PICS is simply one way of applying labels
to content. Its sole reason for being was to head off the CDA. Now
that the judicial route has been taken I see little likelyhood that PICS
can succeed since either the supreme court will uphold the CDA
and we have the Singapore scenario or the CDA gets booted out
and the matter is over.
If the US congress had wanted to do any good instead of making
itself look good then the voluntary approach of PICS with the
multiple rating schemes was the one most likely to work. I don't
think that it would stop 16 year olds from seeing pornography
but since children of that age are in most jurdisdictions permitted
by law to engage in sex on their own account it seems a bit bizare
to prohibit them from seeing pictures of sexual acts.
The problem with Firefly is that its a good(ish) idea baddly
implemented. The much vaunted "agent" technology uses only a
very primitive nearest neighbour type match. There is no attempt
to draw structural inferences from the material, such as abstractions.
For example if I enjoy Dire Straights and Peter Gabriel then a
shop assistant would peg me as a late 70s rock fan and point
me towards the Fleetwood Mac and such. Firefly has 50% of
the structure needed to produce interesting matches but
lacks the ability to make inferences. At least if it is the same
technology as Ringo, the previous generation.
This shortcomming of Ringo means that it is very slow work
training it. To get a useful measure it needs hundreds of data
points. When I visited I was faced with page after page of US
90s chart bands which I've not heard of and have no interest
in.
Phill