robison@eosp1.UUCP (Tobias D. Robison) (12/26/84)
This memo discusses the cost of moderating most news, and a
possible way (requiring "ai" software?) to avoid human moderation.
Assuming that sites receive news from satellite, and feed
news articles generated from their site via the existing news
network, there will be a great savings in use of that network if
we establish a system of "roots" feeding from small capilaries
up to trunk feeder lines to the site(s) that feed the satellite.
If it is necessary to moderate everything that goes via satellite,
there will be considerable extra costs:
(1) Just to overread everything. Even if people donate their
services, that's a lot of donated services.
(2) Moderation will slow the timeliness of responses, damping many
useful discussions.
(3) The access of moderators to the net has cost implications:
- If moderators are spread all over, then news must be fed to
them, rather than to the trunk satellite feeder sites. That
type of feeding will be more complicated and costly.
- If the moderators are physically spread out, but make long distance
phone calls to moderate news at the satellite feeder sites,
these phone calls will be expensive.
- If moderators are required to live at the trunk feeder sites,
so they can pre-check all news via local connections, the result
could be elitist control of the news.
It is possible that federal laws will make moderation necessary.
But I think that most of it could be done via software! Consider:
(0) We would continue to distinguish between groups that need a
moderator jsut to keep the net from sending illegal stuff, and groups
that are moderated in order to achieve filtered/summarized discussion.
(1) Human modertors will not be perfect; they will occasionally let
something slip through that shouldn't. Software need not be "perfect"
to replace or assist them, just very good.
(2) Software can scan for swearwords, suggestive language, expletives,
phone numbers and credit card numbers faster than humans.
(3) Software might be able to detect cases where hundreds of people
send similar short messages (such as "yes, my byte magazine was
delivered late too").
A reasonable procedure would be for human moderators to read anything
caught by software checkers, and to let the rest go through.
Obviously, while installing a procedure like this, Everything
would be overread for a while.
Analyzing news to detect inappropriate material is an interesting
challenge for the ai community, but I think not that difficult
to do fairly well.
- Toby Robison (not Robinson!)
{allegra, decvax!ittvax, fisher, princeton}!eosp1!robisonsteiny@scc.UUCP (Don Steiny) (12/28/84)
** Maybe we could get the National Security Agency's list of key words that they look for when monitoring phone conversations. Anything that was flagged could be checked to keep the net out of trouble. -- scc!steiny Don Steiny - Personetics @ (408) 425-0382 109 Torrey Pine Terr. Santa Cruz, Calif. 95060 ihnp4!pesnta -\ fortune!idsvax -> scc!steiny ucbvax!twg -/
lauren@vortex.UUCP (Lauren Weinstein) (12/28/84)
I don't consider software to be adequate or desirable for netnews screening. From the standpoint of avoiding the transmission of libelous, copyrighted, or otherwise unsuitable materials, no software could be designed that would handle such tasks except in the most obvious of cases. Such software could also be easily circumvented through techniques that should be obvious to all of us. From a legal standpoint, even if human moderators occasionally let things slip through, we would at least have shown we made a good faith effort to do things rights if we had people watching over the material. If we had some silly software doing it, any court would laugh itself sick over the premise that THAT, given the state of the art, represented any real sort of screening. Apart from screening for unsuitable materials, it is my hope that the groups sent by satellite will eventually represent a better quality of material. And just like the editor of Time Magazine doesn't publish every piece of material that crosses his desk or that people send in, this service doesn't need to either. In fact, nobody would read Time if he did. This service is not to REPLACE Usenet, but rather is to provide an alternative for people who do not have the time, inclination, or money to handle the ever increasing volume of calls (which will get far worse as the net grows) with a smaller and smaller percentage of messages representing useful information to them. People who want to carry on their rapid fire discussions in such groups as net.religion and net.singles can go ahead -- but there are quite a few people who could live quite nicely without those groups (and some other groups like it) and would really like to spend their time reading material with a higher percentage of usefulness. The idea is to give these people a choice -- the full, growing dialup network for those who want it (sort of analogous to standing at a sewer outfall), and something a little more controlled and filtered for people who can't afford the time or money to wade through all that. One point is certainly true -- careful consideration must be given to the flow paths toward stargate to avoid undesirable delays. However, my own concept is that most of these materials would be MAILED directly to the moderator, not passed slowly through the netnews links. The current experimental model does not represent the long term picture that would be necessary to make things really work. Also, it would seem reasonable that, ultimately, moderators/screeners/editors would be compensated in some way for their time. I don't think a nationwide news broadcasting service can operate totally on volunteer labor forever! Remember, what you see right now is an experiment, not the shape of any possible future production system. --Lauren--
tim@cmu-cs-k.ARPA (Tim Maroney) (12/29/84)
What is this nonsense about screening out "swear words" from satellite news? I doubt that the law requires this, considering that uncensored movies are transmitted via satellite all the time. Let's not introduce such juvenile foolishness into the news system unless the law mandates it. There is no reason to think that an article containing words some consider "obscene" could not be well worth reading. -=- Tim Maroney, Carnegie-Mellon University Computation Center ARPA: Tim.Maroney@CMU-CS-K uucp: seismo!cmu-cs-k!tim CompuServe: 74176,1360 audio: shout "Hey, Tim!" "Remember all ye that existence is pure joy; that all the sorrows are but as shadows; they pass & are done; but there is that which remains." Liber AL, II:9.
robison@eosp1.UUCP (Tobias D. Robison) (01/04/85)
(10-line quote at end)
The suggestion to screen net software for obscene words
comes from me, and is part of a larger, more
interesting problem that may be unsolvable at
the present time. I still think it is worth research.
Behind my argument lie these assumptions:
(1) In the future, moderation to avoid legal
liability is inevitable.
(2) Moderation will slow the flow of news
and should be avoided wherever possible.
From a new perspective:
Imagine that you are about to submit an article to the
future net. You may write about anything you please,
but you know that any article that might conceivably
be libellous or illegal will be scanned by a human
moderator. Your artcile will be screened by a
computer program to determine whether moderation
is necessary. For the sake of this discussion
I assume that a moderator never edits your text,
but simply determines whether it is legally safe to
broadcast it. You can write about anything you like,
but you have two choices:
(1) Write an article that certainly deserves to
pass the computer screening. It will be posted
to the net relatively quickly.
(2) Write an article including anything you like.
You will acceprt the delay required for human
over-reading.
In the specific case of Tim Maroney's concern, you
may include obscene language if you feel this
is appropriate, but of course your note will be
screened by a moderator.
The PROBLEM is to write software that can distinguish
between the two types of articles as accurately as
a human reader. Bear in mind that a human
reader will not be perfect either.
The program that does the screening should be very
conservative in what it will pass.
Most of its algorithm should be public knowledge.
The algorithm will simply establish a style
that is acceptable for quick-distribution-notes.
Now while someone (I hope) thinks about the AI
implicatiions of this screening algorithm,
I invite net.games.pbm subscribers to propose
pathological cases that will fail; that is,
how easy would it be to write a nasty, scurrilous
note that would sneak past the software screen?
If such notes are very hard to write, the
existence of software screening in the future
can greatly reduce our reliance on
human moderation.
- Toby Robison (not Robinson!)
{allegra, decvax!ittvax, fisher, princeton}!eosp1!robison
In article <20980040@cmu-cs-k.ARPA> tim@cmu-cs-k.ARPA
(Tim Maroney) writes:
>What is this nonsense about screening out "swear words"
>from satellite news?
>I doubt that the law requires this, considering that
>uncensored movies are
>transmitted via satellite all the time.
>Let's not introduce such juvenile
>foolishness into the news system unless the law mandates it.tim@cmu-cs-k.ARPA (Tim Maroney) (01/05/85)
Toby Robison is interesting as usual, but I feel the point of my concern has not been addressed directly. Is there some legal requirement that satellite-broadcast USENET messages not contain words which some people call "obscene"? If not, then I strongly suggest that such words not be used as a criterion for rejection of an article by satellite article screeners. This objection is not made on personal grounds -- anyone who follows my messages knows I very rarely use such words myself (since there are usually more expressive ways to communicate). The objection is that ANY unneccesary censorship is to be avoided at all costs, and this should be considered a general ethical principle. -=- Tim Maroney, Carnegie-Mellon University Computation Center ARPA: Tim.Maroney@CMU-CS-K uucp: seismo!cmu-cs-k!tim CompuServe: 74176,1360 audio: shout "Hey, Tim!" "Remember all ye that existence is pure joy; that all the sorrows are but as shadows; they pass & are done; but there is that which remains." Liber AL, II:9.
mark@cbosgd.UUCP (Mark Horton) (01/06/85)
In article <20980049@cmu-cs-k.ARPA> tim@cmu-cs-k.ARPA (Tim Maroney) writes: >Toby Robison is interesting as usual, but I feel the point of my concern has >not been addressed directly. Is there some legal requirement that >satellite-broadcast USENET messages not contain words which some people call >"obscene"? If not, then I strongly suggest that such words not be used as a >criterion for rejection of an article by satellite article screeners. Obscene words are not the issue. The problem is another type of message: the one that encourages and assists illegal behavior. Such as A working telephone credit card number is xxx-xxx-xxxx. Have fun! or I have proof that <insert name of person> has embezzled large sums of money from <insert name of company>. In cases like this, someone gets hurt. That someone could be looking for someone to sue, and the company with the transmission facility is an obvious target. The recent bboard case, where the computer on which the bboard resided was confiscated, sets a precedent. We have to take whatever measures we reasonably can to prevent such things from happening. I understand that the company in question has specifically insisted that everything they broadcast be screened. I can't imagine how an AI program could be expected to detect something like this. Besides, if such a program were put into place, it would have bugs that would quickly become well known, and it would become easy to fool it. Mark Horton
kay@flame.UUCP (Kay Dekker) (01/07/85)
[[][]] >......... You may write about anything you please, >but you know that any article that might conceivably >be libellous or illegal will be scanned by a human >moderator. Your artcile will be screened by a >computer program to determine whether moderation >is necessary. For the sake of this discussion >I assume that a moderator never edits your text, >but simply determines whether it is legally safe to >broadcast it. Excuse me, but I think there may be a problem here. Both obscenity and libellousness are rather difficult to screen for. 1) According to English law, 'obscene' is defined as 'having a tendency to deprave and corrupt'. This is extremely knotty: the 'Lady Chatterley' and 'OZ' cases illustrate this. 2) There are cases where seemingly-libellous material may in fact not be so. For example, of the publication is 'in the public interest', or is 'fair comment'. I cannot see software (or even moderators) being able to screen articles for 'obscenity' or 'libellousness': it has taken juries many days to argue over these points. Furthermore, I gather that the laws which govern permissible public utterances vary wildly between countries. The screening rules must then have knowledge of the different regulations that apply over the various countries into which net-contents enter. For example, in England, we have a law which makes illegal 'Blasphemous Libel'. Prosecutions for this offence are extremely rare: it was last trundled out in 197[67] by our protector of public propriety, Mrs. Mary Whitehouse. She was offended by a poem by James Kirkup, "The love that dares to speak its name", which appeared in the British gay newspaper, "Gay News". The prosecution was successful, and the editor and the paper were fined heavily and the editor given a suspended prison sentence. How many other archaic laws and regulations would this screening software have to know about? Kay. -- "But what we need to know is, do people want nasally-insertable computers?" ... mcvax!ukc!flame!kay
rrizzo@bbncca.ARPA (Ron Rizzo) (01/10/85)
Screened by a computer program to decide if a human moderator (=censor) is required? Either the prudes are lightyears ahead of us in AI, else they'll simply force us to use euphemisms to frustrate searches for "keywords" (locutions like "the love that dare not speak its name" for homosexuality); or will they dump The Quean's Vernacular into their database, accelerating the creation of new slanguage. The only way to defeat such a counterreaction is to proscribe entire classes of nouns, verbs, etc. (OED goes into the database). This is precisely what happens in Orwell's 1984. Nicefeels doublegood, Ron Rizzo (This ISN'T my real name!)
pgp@hou2h.UUCP (P.PALMER) (01/11/85)
I think this discussion should be moved IMMEDIATELY to something like net.security. The whole idea of "moderating", which is a gross euphemism for censoring, is obnoxious (and unlikely to be accepted) anyway.
smh@mit-eddie.UUCP (Steven M. Haflich) (01/12/85)
Have I missed something? The ongoing discussion on detecting `libelous' postings addresses only certain kinds of libel -- scurrilous or obscene descriptions of persons with defamatory intent -- but misses entirely kinds of libel rather more likely in this environment. Suppose I were to write: In his recent posting, Toby Robinson (not Robison!) wrote: I feel the future of AI programming lies in assembly language, since only by using assembly language can the careful programmer attain those important last few percent of available machine performance, so important to successful AI applications. I would not work for any company that insisted on my writing code in inefficient languages like Lisp or Prolog. I cannot agree with Toby on this point. ... Note that my `posting' is about a valid technical subject and is written in neutral terms of the technical field. Unless the fictional Robinson had actually made such a statement, such a posting would (I believe) be libel. With flagrant disregard for the truth, it clearly damages Robinson's reputation and presumably could also damage his employment opportunities. It is *not* necessary for me to claim someone practices nonconsentual sex with laser printers in order to libel him. He would have legal recourse against me and my employer. It might be possible, I suppose, for the automatic censor to verify quoted inclusions against the article database, but what about: At the recent SIGAI meeting in Nepal Toby Robinson (not Robison!) told me he felt the future of AI ... ... ... I cannot agree with Toby on this point. ... There is no way for a machine to verify this one. If the automatic censor must kick out any quoted or paraphrased citation for review by a human, almost *everything* will have to be reviewed! So why bother? Steve Haflich, MIT
sdyer@bbncca.ARPA (Steve Dyer) (01/13/85)
Please remove the newsgroup reference to net.motss on any subsequent
discussion of this topic. It's hard to imagine a less appropriate
newsgroup.
--
/Steve Dyer
{decvax,linus,ima,ihnp4}!bbncca!sdyer
sdyer@bbncca.ARPA