Fatal
Filtering: Software Deterrents to Expression on the Web
by
John E. Bowes
School of Communications
Box 353740
University of Washington
Seattle, WA 98195-3740
Draft Copy
Prepared for the
Communication Technology and Policy Division
AEJMC Midyear Convention,
Denver, CO
25-26 February 2000.
I |
n
the several years since the US Supreme Court denial of the Communications
Decency Act of 1996,[1] public and
governmental attention has shifted to imposing Internet content controls at the
consumer side of transmission – in preference to the First Amendment problems
arising from controls upon content producers.
Since the refutation of this law and its successor, the Children’s
On-Line Protection Act,[2]
other changes in industry economics and policy have raised the stakes to find
agile, constitutionally defensible ways of protecting the public from unseemly
Internet sites. Most efforts have settled on a new class of software, filter
agents, which interdict “on the fly” unwanted content and keep it from
being viewed.
Broadly,
the problem examined in this paper is the accuracy of these agents in (a)
filtering only objectionable content; (b) having clear policy statements about
the action of filters – how they work and the list of blocking terms or
websites used; and (c) imposing editorial and citizen review to mitigate
“blind” censoring by automatic filtering based on key words. The net impact of inaccuracy in these
programs is to exclude from public view certain minority groups (gays and
lesbians), charitable or informational groups (breast/testicular cancer
assistance), and free web site customers (anyone unable to afford commercial,
institutional or their own web servers). Further, filter agents may serve to
restrict access to comment on sensitive issues and groups, regardless of
source. Opinions on pornography, pedophilia, AIDS, safe sex, birth control,
Nazis, anarchy, prayer in schools – to name a few – have been banned by clumsy
association with sites offering obscenity, hate, sexually explicit materials
and extremist religious or political views.
In
reaction, a number of groups, ranging from the American Civil Liberties Union
to the American Library Association and the Gays and Lesbians Allied Against
Defamation have mounted a vigorous campaign to fight mandatory filtration
use in libraries and schools, and to force manufacturers of this software to
improve accuracy and flexibility of their products. With equal vigor, social
conservatives and their organizations like the Family Research Council have
fought back, advocating not only home use of filter agents, but its mandatory
use in libraries, schools and “upstream” in the delivery servers of internet
service providers (ISPs). To both poles
in this debate, the stakes are high: the access to minority opinion and
sensitive topics by one side and protection of children and social order by the
other.
This
study examines the approximately 50 filter agents presently available on the US
market for the clarity of their mechanisms to consumers, the care exercised in
making revisions (as when wrongfully blocked sites are made known), and the
visibility of filtering to the end-user who should know that filtering or monitoring
is in place. Based on these findings, we make recommendations for improving the
evaluation of filters and features that minimize blocking of socially useful
minority content. We examine technical
information and public product descriptions for clues to its operation – the
sort of action careful consumers should take before purchasing this software or
supporting its use in community institutions. Later, we discuss the
methodologies needed for systematic, valid comparison of agents in actual
field-test situations.
The
Recent Evolution of Filter Agents: In their
1998 report, Censorship in a Box,[3]
The ACLU suggests that use of filtering software has experienced “explosive
growth” since the defeat of the Communication Decency Act of 1996 in Reno v.
ACLU. The report notes that in 1997, some $14 million in filtering software
was sold with projections that “blocking” software products would grow to $75
million in 3 years – by the year 2000. The American Library Association reports
an Internet access rate of 60% by 1999 in US libraries, up from 28% in
1996. The universal access mandates of
the Telecommunications Act of 1996 help assure that this figure will climb to
near totality over the next few years. The Constitutional problems apparent in
“supply side” control of Internet content as seen with the Children’s Online
Protection Act (COPA) leave filtering as the main “protective” product
available to schools, libraries and parents. Faced with escalating statistics
on “cyberslacking”[4] – workplace
misuse of corporate Internet connections – businesses have joined as profitable
partners to worried parents.
Growth
and Use: Paralleling the growth of
filtering products is its increasing availability on the servers of Internet
access providers.[5] This
“upstream” filtering requires no user implementation; offending websites never
appear past the filter placed on the network server. One filter program, CyberPatrol,
has served America Online, Compuserve, Prodigy, AT&T, Bell-AtlanticNet, and
Scholastic Net, among others – over 24 million subscribers. A half decade ago, the variety of Internet
providers might have offered more choices from no to heavy censoring. They were
local and relatively small, reflecting regional norms. But since the mid 1990s, there has been a
rapid consolidation of ISPs. Local “mom & pop” providers are increasingly
bought-up or competitively flanked by national scale organizations.[6] Some 160 Internet service providers were
consolidated in 1998 with another 70 meeting a similar fate in the first three
months of 1999.[7] America
Online has attained the scale of a large international telecommunications
organization with over 19 million subscribers. Portal page providers like Yahoo,
have become billion dollar organizations with customers across a half dozen
nations.
As
these newly large corporations face anxious politicians and a tangle of
differing local and national criteria for acceptable content, they veer towards
the safest and most conservative standards. They have a lot to lose by
protracted legal problems with government and policy problems with frightened,
angry parents. In one extreme case, a complaint of Internet pornography
transmission laid by the Bavarian state prosecutor in 1995, Compuserve
was obliged to disconnected customers across Europe from over 200 Usenet news
groups. Felix Somm, a Compuserve manager, was sentenced to a heavy fine
and a two-year suspended prison term because he had "abused" the
Internet and “allowed” child pornography and Nazi literature – both of which
are illegal in Germany – to be available to German CompuServe users.[8] It made little sense to the court that Somm
had no real ability to monitor or control the flood of international items
coursing through the Internet.
It
is cost-efficient for AOL, Compuserve and others to hand over
“vigilance” by buying a popular commercial filter and giving their subscribers
the choice to turn it on. The problem is that software selection may be geared
to minimal price, low supervisory overhead and a common denominator bland inoffensiveness. In short, as size, consolidation and
corporate standardization of software take place, filtering has grown in
impact. In contrast to the landscape of thousands of small Internet providers
several years ago, each with a particular local outlook and varied policies,
consolidation narrows the choices.
Far
from an abstract problem, these trends could mean a progressive denial of
access to socially useful information considered “controversial” or pressing
the margins of “risk control” for litigation-shy Internet providers. Over-cautious filter software rarely excites
hostility from mainstream consumers and wary government officials. The losses of useful content are mainly
unseen unless specifically tested for (see below), while “bad” sites slipping
through filter agents can excite considerable protest. Corporate risk
management is likely to win over more abstract notions of “fairness” to
marginal interests whose omission can be difficult to detect.
Filtering Legislation as a First Amendment Work-around: Since the failure of the
Communications Decency Act, there is ample evidence that governments
increasingly see filtering as their quickest fix for calming those worried
about Internet content. The negative outcomes given outright censorship in ACLU
v. Reno and later in a second round with the Child Online Protection Act
have increased reliance on filters, most recently seen in The Internet
School Filtering Act (S. 1619) of 1999 proposed by Sen. John McCain. If Federal efforts weren’t enough, the ACLU
reports in 1998 that 10 Internet censorship bills were proposed in state
legislatures, 5 of them specifically requiring filter or blocking software in
schools and libraries.[9] By recommending filtering software, the main
gate-keeping task is transferred from provider and regulator to parents,
schools and libraries. More importantly, in the hands of vulnerable local
institutions, filters become an easy pressure point for special interests,
local and state politicians and any parent at odds with the Internet.
Lower
courts are beginning to take a critical view of filtering by civic
institutions. The recent Mainstream Loudoun v. Loudoun County Library decision
showed that filtering all of a library’s machines was equivalent to
“removing books from the shelves.”[10] The decision in late 1998 found a number of
problems with indiscriminant use of filtering software: adults were denied
access to constitutionally protected material because it was unfit for
children; standards were not spelled out, and there was a lack of safeguards or
process for judicial review of policy.
Despite this decision, it has not slowed legislative efforts to mandate
filtering software. As Karen Schneider, organizer of the pioneering Internet
Filter Assessment Project described, over 15% of libraries now filter.[11]
Apart
from national-level freedom of speech interest groups, international
associations responsible for global Internet policy have entered this
controversy. Prominent among these is The
Global Internet Liberty Campaign, an international union of human rights
organizations favoring unfettered access to electronic information. Their
member statement to the Internet Content Summit at Munich this September (1999)
is worth repeating:
“The creation of an international rating and filtering system for Internet content has been proposed as an alternative to national legislation regulating online speech. Contrary to their original intent, such systems may actually facilitate governmental restrictions on Internet expression. Additionally, rating and filtering schemes may prevent individuals from discussing controversial or unpopular topics, impose burdensome compliance costs on speakers, distort the fundamental cultural diversity of the Internet, enable invisible "upstream" filtering, and eventually create a homogenized Internet dominated by large commercial interests. In order to avoid the undesirable effects of legal and technical solutions that seek to block the free flow of information, alternative educational approaches should be emphasized as less restrictive means of ensuring beneficial uses of the Internet.”[12]
It is important to note that fears of filtering
extend well beyond exclusion of minorities. With wide, uncritical adoption of
filtering, there is fear that Internet will become bland, branded and
standardized.
Faulty Filters: A major complaint by critics
is that filters have unintended consequences. While concerned minorities have
prevailed on software manufacturers to fix glaring problems, a variety of
forces can make the best-intentioned efforts go wrong. Karen Schneider
describes in grim detail just how bad this software could be for gay and
lesbian-oriented information. As she aptly concludes, “ . . . It isn’t
realistic to expect that a $40 piece of software could effectively manage
something as huge as the Internet – and it’s equally unrealistic to expect a
piece of software to instantly customize itself to the values of each person
who uses it.”[13] The
evidence of her conclusions abound.
In
Passing Porn, Banning the Bible, The Censorware Project tested one
popular blocking product, N2H2’s Bess. As the title suggests, this
popular filter software had many unintended consequences, despite the firm’s
assurances to Congress that human editors reviewed all banned sites. Web sites advocating celibacy, having
information on cats, about ghosts, Serbia, or sites critical of Internet
censorship were all caught in Bess’ filters. All free web page sites
were banned as a matter of policy. These servers, where individuals may post
their own web sites without cost, are among the largest sources of Internet
content. On the other hand, many obviously pornographic sites slipped by Bess,
such as “stripshowlive.com,” “legal-sex.com” and “digitaldesires.com.”[14] It should satisfy no one that safe and
risqué topics seem abused by filtering in apparently equal measure.
These
errors clearly fault the idea that software presently available can adroitly
filter the massive 20-plus terabytes of data that make-up the Internet
database. A 1996 study estimated the Internet at 1.5 terabytes (then) and
calculated that better than a third of the web pages changed per month.[15]
One current estimate is that new
pages now are added to the web at a rate exceeding 25 pages per second. Filters
are swiftly blind-sided by changes in key word usage, semantics and the onrush
of new web pages. They demand continual maintenance by editors who should
quickly see the categorical differences between, say, Asian wildlife and
“hotasianfoxes.com.”
Will
institutions pay the costs of continual maintenance? And will parents who have
installed blocking software on home machines do likewise? Can any software
organization maintain massive review and vigilance for the sake of a $40
product? Data can’t be found on these questions, but we suspect for many –
individuals and organizations – the task may fall into the class of “to do”
computer chores that never quite get done.
It’s just too time-consuming and troublesome. Editors must also
anticipate inventive countermeasures by those most likely to be screened out,
such as loading sites with tame keywords deliberately to frustrate filters. A
site like wwwmen.com, despite the innocuous name, is about gay
pornography, not about men’s health or fashion.
Most
filters and blocking software are proprietary or secret in terms of their
precise functioning and so too their keyword lists of censored terms. One must test filters with benchmark
programs to learn, for example, that Searchopolis, the engine underlying
Bess, returns the same website list on “testicular cancer” as it does on
“cancer.”[16] Only through experience do we discover that
“testicular” is banned, having no influence on the search. While some filter
software firms are responsive to citizen criticism and have advisory boards to
adjudicate tough categorical issues (such as OK-for-kids gay websites), very
few reveal the details of their products’ operations.
A
growing number of studies have tested filters, but few with anything
approaching a comprehensive benchmark characteristic of the vast range of
material making-up the Internet. It is estimated that the best commercial
search engines rarely cover more than about 20% of the web’s contents. The analytic overhead of not only
representative surveying of web content, but of gauging accurately its use, is
the focus of intense commercial competition and debate. In many respects, the
magnitude of this task is like that of a national census – expensive to do,
open to continual political controversy and likely to be biased, however keenly
safeguards are employed.
Open Platform and Search Engine Accuracy: Because software engineering is among the most competitive of
enterprises, few inner workings of filters are made public. To find products to
examine, we accessed several current “master” lists of products: Anne Bubnic’s
list of Parent Control Resources,[17]
GetWiseNet’s “Tool’s for Families,[18]”
and AT&T’s key sources for parents.[19] Promotional materials and technical data
from the makers’ websites were examined for key characteristics of how the
filter agent worked and the extent of human intervention to make sure socially
useful sites were not unintentionally excluded. We were especially concerned
with several hallmarks of filter accuracy:
1.
Method and Visibility of Control: How a filter agent works is important to its
flexibility and subtlety with new content and content that may be about such
banned topics as pornography, but not pornography itself. Proxy servers, for example, commonly pool
and route all machines in a library or school to one high capacity Internet
connection. These connections can be programmed to block unwanted sites from
being available anywhere on networked machines. In short, the Internet is filtered for all local users by an
“upstream” computer or “firewall” that checks each request for web content.
Other systems make no demands on local hardware or software. Operating as
Internet service providers, H2H2’s Bess and other child-safe sites
“clean-up” and select all material to be made available at the “supply” side of
the network content stream.
Blocking uses key words and web addresses to
specifically screen certain topics and sites from view on the client’s
computer. But not all blockers work
alike: an important distinction is between blockers that use solely keywords and
those whose action is supplemented by actual web addresses. The latter requires
editorial oversight to screen individual web pages, but it is considered more
accurate, screening out fewer innocent sites.
Some keyword blockers also try to detect “context” so as to distinguish
between images of women’s “breasts” and information on “breast cancer.” These
may use complex schemes to detect themes and the social utility of the
information. Key word blocking, in short, tries to anticipate undesirable web
pages, whether old or brand new, and eliminate them on the fly. Address
blockers, contrary, require the earlier preparation of either a “not” list of
forbidden addresses or an “allow” list of only those sites that may be used.
These may run into the thousands of entries.
Taking an entirely different approach, stealth
monitors do no filtering at all. Instead they take snapshots of the screens
each viewer accesses during a period of use (these may be every screen or a
sample). Such systems rely on control occurring after inappropriate
material has been viewed. They may be used to deter “cyberslacking” by
employees or may become the basis of a disciplinary discussion with children
who access inappropriate sites. What is or is not acceptable is determined by
reviewing the record of each user. Most
such monitors are invisible to users, leaving warnings to those supervising use
of the web.
The key point is that hidden means constitute the
basis of 60.5% of the software available, counting stealth monitoring together
with proxy server and content access controls that can operate without warning
messages. Some 54% of software packages
mention “secrecy” or “invisibility” in their product promotions.
About 27% (n=13) of products reviewed use several
methods of filtering at once. Most commonly, content access barriers –
passwords and specialized websites – pre-filter content. Otherwise, stealth
monitoring is included for parents and employers who wish to track use by
children and employees in a “cleaned-up” environment.
Table 1: Showing secondary
filtering features by primary filtering method. Agents using a single method have been combined.
|
|
Single
Method |
Content
Access. |
Stealth
Monitor |
Total |
Other Methods |
24 |
0 |
0 |
24 |
| |
100.0% |
0 |
0 |
100.0% |
| Blocking |
4 |
6 |
5 |
15 |
| |
28.6% |
40.0% |
33.3% |
100.0% |
| Proxy
Server |
7 |
1 |
1 |
9 |
| |
77.8% |
11.1% |
11.1% |
100.0% |
| Total |
35 |
7 |
6 |
48 |
|
|
75.5% |
12.2% |
12.2% |
100.0% |
There are other more blatant but less used means of
filtering the Internet. Hardware devices really are a simply lock and key that
prevents the computer’s modem from reaching the telephone line. Ratings attempt
to assign numbers to web pages based on their violence, profanity, explicit
depictions of sex and other characteristics. But this requires a widely used,
agreed-upon definitional scheme and clear rules for judging severity. Movie, cable television and video games have
struggled with this idea for years with little consistent success.
2.
Is the methodology of the filter agent discussed? These descriptions range
from a single sentence promising that threatening content will be banished to
complex discussions of how keyword filters operate in all their complexity.
Most descriptions fall in-between these extremes with very little disclosure of
details. These products compete in a busy marketplace and thus try to find a
middle ground between overloading parents with technical details and touting
their sophistication. Too much detail also gives competitors and critics
insights into techniques that may be copied – or hard to defend.
In some 62% of filter agent descriptions, how the
filter worked received more than single sentence description, sometimes
continuing on for several pages. A few
agents did not need description due to their simplicity. Hardware locks simply
put one’s modem under lock and key. Some descriptions were so perfunctory or
alluded to engines leased from other providers that we classed these
descriptions as questionable.
We also need to note in passing the dependency of
filter agents upon words and text. Images can’t be recognized by automated
software available to consumers. How, for example, would $40 software detect
images of Greek sculpture apart from modern-day pornography based on form
alone? The important point is to reckon just what software does internally with
its keywords and the level of human oversight into this process.
3.
Availability of block word list: Most filtering – either upstream with an Internet
provider or operating on a user’s machine – employs a list of keywords to block
offensive content. These may be common words like “sex” or “bomb,” slang like
“hash” or “pot,” hate words like “nigger” or “faggot,” or technical terms
offensive to some parents like “testicular” or “fetus.”[20]
Each page summoned from the web is tested against the list for the presence of
these words. The problem is that such terms have both innocent and offensive
uses, context depending. Filter software differs in the agility of these words
to be detected in their offensive contexts, yet allow the innocent use to pass
through to the user. One defense is to
allow the keyword or blocking word list to be viewed by purchasers of the
filter agent. Few (8.33%) allow this, preferring to keep them as competitive
secrets or – recognizing largeness of these lists – not to provide them for
reasons of economy. The difficulty is
that individuals must learn proscribed words by testing filters with pages
known to contain these terms. Some filter providers (CyberPatrol) offer
web test pages where terms can be tried against simulated filtering action. In
the figure above, certain filter agents, particularly those offering only
monitoring of users do not actually filter anything. They merely report what
people have been viewing. As such, they really are not applicable (N/A) to this
count.
4.
Customizing Agents: A majority of agents allowed parents and others to add blocking terms
(words, phrases) and URLs (website addresses) to the blocking list. In a few instances, factory blocked URLs and
terms could be “excepted” from blocking, allowing a more complete customizing
control. It is important to note that some packages use an “allow” list where
all but permitted terms and sites are rejected in contrast to the more common
practice of a “not” list where forbidden terms and sites are listed. Obviously,
the first option yields a highly constrained world of possibilities, but with a
lower risk apparently of offensive materials slipping by the filter. Filters
made for very young children tend to employ this severe method, the idea being
that diversity is secondary to protection.
The use of “not’ lists has a claimed 90 to 95%
screening success, while “allow” lists claim much higher. The cost is in the
innocent and presumably helpful sites that vanish because they are not on an
“allow” list. Particularly devastating are changes that take place in both
language and websites with time. “Not” lists will catch such shifts much more
often those terms not imagined for inclusion on an “allow” system.
5.
Regular Factory Updates: Because of rapid social changes, language,
semantics, slang and politically incorrect speech can also change swiftly. In an earlier time, the term “bareback”
referred to horse riding without saddles. In today’s slang, it also means
unprotected sex between gay males – an implication that is only a year or two
old. If parents lack the time to search out changes in offensive content and
the new websites supporting it, then the manufacturer may do this as a free or
paid service. Some filter agents can be regularly updated automatically with a fresh
list on a monthly basis. Usually a small subscription fee is charged. Regular
updating vital to the effectiveness of filter agents, given the fast content
turnover and the rapidly evolving pop culture terminology pervasive on the web.
About half of the filter providers offer updates,
either as full revisions of their software or as add-in lists that update the
original with the program. Clearly this is a task that could be done by
concerned parents and institutions, but its likely few have the persistence and
expertise to manually update this kind of utility software month-after-month.
Fully automatic software that fetches updates on its own is rare for home
users. Only staffed “upstream” filtering by Internet providers afford this care
on a continuing basis. The difficulty is that very few providers of upstream
filtering said much about how often and by what procedure they keep their
filter lists up-to-date. Those providing solely monitoring (indicated as N/A)
do not filter and thus have no need of lists or updates.
6.
Review Boards: A few providers of filter agents maintain advisory boards of parents,
experts and, occasionally, minority groups who advise on new words and sites to
consider for filtering. More importantly, perhaps, they may report unintentional
filtering or provide a minority viewpoint on “acceptable content.” Gay and
lesbian groups, for example, have been vocal critics of filter agents because
socially useful content about this community is often filtered as indecent or
at least upsetting to socially conservative families. In this item, we counted only 12.5% of providers as clearly
having some kind of editorial board to review key word and web site addresses
used for blocking. A much higher level – 37.5% -- was indeterminate where it was
difficult to tell if regular staff simply reviewed lists periodically or that
outside consultants were specially deputized to review unfair exclusions and
bad key word choices. In normal practice, automatic web crawlers or “spiders”
are commonly used to compile blocking lists as an alternative to costly human
oversight. With only few firms offering review panels, continuing errors are
likely.
7.
Multiple Users: Half of the products reviewed could be adjusted for multiple users.
This is important in allowing appropriate material to be selectively allowed
according to a user’s age or other criteria. It permits adult users to have
little filtering and very young children to have higher levels of “safety.”
Users each have unique log-in names and passwords with preset levels of
filtering. On-line Internet providers, proxy systems and software resident on
individual PCs all can have this capability.
This
review must be tempered by the fact that relatively few of these products have
high use. Just one product, CyberPatrol, has (arguably) more than half
the sales of filters for individual machines.[21] Reliable sales data for proxy and upstream
filters aren’t available, but these agents are likely to prosper as states and
school districts mandate centrally controllable filters for all machines in the
community. Some firms have taken extraordinary steps to refine their lists and
address complaints of wrongful blocking. Yet, these powerful and often clumsy
tools are sold from discount store shelves with little description and
incentive for the consumer to question the actions they silently make, once
installed. Happily, several of the most popular products make strong efforts to
update, review and give thorough explanations of how their products work.
Monitoring and Privacy: Monitoring
and use of filter agents to control content is part of a larger national issue
of covert surveillance and privacy on the Web.
In data gathered from the Georgetown Internet Privacy Policy Survey
(GIPPS), of 7,500 busiest servers on the World Wide Web, ninety-three percent
of the sites in this survey collect personal information from consumers; some
66% post at least one disclosure about their information practices.[22] The Online Privacy Alliance, a coalition of
industry groups, found that 99% of sites in their study collect personal
information.[23] A 1999 report by the Forrester Group claims
that 90% of commercial websites are deficient in protecting the privacy of
their users.[24] Beyond commercial interests, government has
persisted since the Reagan years to increase the surveillance powers of the
Federal government though electronic means to counter “threats to the
infrastructure” or solve problems in pacing the electronic security
counter-measures of organized crime.[25] Businesses have a variety of means to
monitor use. In one recent survey, 74% of businesses queried had some kind of
oversight on their web users.[26] It is perhaps difficult to make a moral
point to children that the rights of others – including privacy – need
protection when the parent is aggressively using surveillance as well.
What
then are the hallmarks of good privacy on the Internet generally and for
monitoring software in particular? What should consumers be asking from
websites, Internet service providers and makers of filtering agents? While it
is difficult to rate any one item as more important than others, several points
are clear:[27]
- Obviousness or Visibility: With filtering, sites
blocked should be identified, rather than being removed invisibly with
“upstream” filters that surreptitiously block forbidden requests and
troublesome keywords. Filters
should be identifiable when they are operating on a given machine. Software capable of various filtering
levels should be easily read as to the severity of filtering being
imposed. Monitoring software, as an adjunct to filters either as a
separate package (Cybersnoop) or a capability included with
a filter (CyberSitter 99), is almost always clandestine in
operation, since its purpose is a kind of parental entrapment for wayward
youth or employees. It is
inherently undermining of trust and presumes wrongful behavior.
- Notice: The website describes openly its information
acquisition and use from consumers. Similarly, the editing action of
filter software can be disclosed in detail without endangering the
intellectual property rights of the code’s authors. The extensiveness of
“not” or “allow” lists and the level of human editorial oversight are both
important consumer information. Ideally,
this is done in a clear and obvious way on the filter product package or
portal website’s “top” or home page, rather than as a small typeface legal
notice buried elsewhere.
- Duration: With blocking sites or software, the
frequency and thoroughness of updates and the interval since previous
revisions should be noted. Browser or server-based monitoring may be
momentary tracking device or persist for months. In this latter case, the
impact is larger, but perhaps forgotten sooner as initial warnings fade
from attention and memory. If collecting such data affords consumers a
convenience (as Amazon.com and others maintain), then
websites should explain the trade-offs and allow the option to be
disabled. The usual appeal to
“convenience” is moot if the consumer can’t vote on it.
- Verification: Choice, Access and Control of
Use: Rarely
is permission asked of consumers for each new item of data collected or
each new use made of it. Yet this level of attention affords consumers
continuing control over the “life” of data gathered about them and kept by
the various web servers. Can users see the files kept? Can they amend or
delete the contents? Can they control those to whom the data are given?
Few merchants and consumer research organizations permit this with data
collected on-line. In an ideal case,
verification allows individual control at each step of consummating a
purchase or of giving-up personal data to a website and its use
thereafter. In filtering and monitoring software, a prime sales virtue is
stealth, intentionally leaving hapless users in ignorance about the level
of surveillance levied. Certainly, user “participation” in this secretive
relationship is not solicited nor is it voluntary.
- Safety and Security: Data “farming,”
monitoring, profiling and related techniques stereotype individuals and
presume patterns of taste and behavior from categorical indicators. Used
commercially, their purpose is to target and identify specific
users (e.g. likely purchasers of expensive vacations). Used as
employee or child surveillance, their purpose is to measure productivity,
threats or “deviant” behavior (e.g. email to competitors; gay
interest groups, etc.). While one can sympathize with employers seeking to
protect their employee’s time and the desire of schools to keep children
from “bad” Internet sites, users need to know how these data are handled;
who views the results? What its consequences are? And who can be contacted
for redress if monitoring is misused? In short, stealth shifts control
entirely to supervisors and unseen managers. There needs to be balance
with good disclosure to those being monitored.
On-line
privacy in operation is a mix of web-based features and proper notification.
Filter agents play an important role in the control of content to and gathering
of information from the user. Their future use likely will grow, and with it a
need to give consumers good information on what is being left out of their
information diet.
Research and Measurement Needs: In this report, we have presented a static view of
filtering issues based upon manufacturers data. For individuals and groups
threatened by filtering, there is a strong need for tracking data, based on
actual performance of the filter. Some early efforts of this sort are
notable. The Internet Filter Assessment Project (TIFAP) initiated in 1997 by
librarian Karen Schneider used volunteers from around the country to test
filter products against test terms and websites.[28]
Noted were both failures to filter offensive materials and the propensity for
socially beneficial sites to be excluded. Results were qualitatively reported –
instances of gross errors and product failures.
But
given the magnitude of the Internet and the range of products, these tests did
not have statistical generalizability and have been outdated by the substantial
turnover in both filtering products and Internet content itself. For general
consumers, filter accuracy, adaptability, updating become important points over
the life of the product. Have filters become more subtle, worse or more
pervasive with time? Are they targeting new key terms and web sites? Have
once-banned terms become acceptable? Does a culturally sensitive product
purchased in 1996 still work acceptably today? Such information must be solid
and credible to industry, academe and government policy-makers. Certainly it
must earn the attention of those who build and promote filtering products. Some
key issues:
·
Internet Monitoring: As suggested above, this task is difficult and fault-prone for the
makers of filtering products. It will not be much easer for those who review
the action of these products on the web. Language, images, hyperlinks, motion
and sounds all present barriers to filters that commonly rely on text and
simple, categorical meanings. We know,
too, that web pages change swiftly with time. Though time “slices” or snapshots
of the Internet are becoming available through The Internet Archiving Project,
the data files are massive and difficult to manipulate.[29]
These are huge problems that partly explain the lack of comprehensive filter
testing.
·
Analytic Efforts: How should websites be categorized? How can one efficiently
manipulate huge datasets efficiently and for a tolerable cost? How will these
tools evolve as the Internet and its supporting industry change with time? In
short, the web awaits standards of description that have stability and wide
acceptance, yet are flexible enough to catch new trends.
·
Benchmark Test Sets: How would filters be fairly and consistently tested as gateways into
the Internet? Can consumer groups, filtering advocates and critics produce test
runs that show differences among search engines, global shifts in filter terms
or phrases, or in websites “banned” with URL or system blocking?
·
Policy Review: Numerous websites have
appeared that promise a “safe harbor” to children. Contents and links on each
have been pre-selected as safe or fulfilling particular moral mandates. Recently, a score of key Internet service
providers have organized GetNetWise.org, a site centralizing information
about filtering and giving assistance to those wanting to try out these tools.
Supervised by the largest Internet providers like America Online, software
providers like CyberPatrol and a number of civic organizations such as
the US Chamber of Commerce and the Internet Content Rating Association, the
site is potentially a powerful force in the adoption and use of filters. As of
this writing, the organization is working out its acceptable use policies.[30]
Clear policy is needed to alert those choosing filters to a sponsoring
organization’s pet themes and possible biases – and to stand critical review by
groups negatively affected.
Will
assessment aid come from without? Software manufacturers serve themselves in
sidestepping evidence of their products’ clumsiness. Government seems ready to
embrace this software – faulty or not – as an alternative to first amendment
tangles and constituent pressure. And increasingly large Internet service
providers seek easy, low maintenance solutions to their oversight
responsibilities. Few of the proposals
mandating filtering or blocking consider the quality of software to be
used in much detail, if at all. It is probably from consumer groups, academics
and organizations who stand to be harmed by the status quo that useful
criticism will come.
A
Growing Threat to Minority Voices? Multicultural content on the web has grown as
improved access and training broaden those able to create websites. The
Internet affords a content diversity that possibly has not been seen since the
birth of radio in the 1920s. As media historians know, much of radio’s original
variety – from churches, schools and community groups – was lost by early
government favor for commercial interests who went on to homogenize content to
a handful of profitable formulas for mainstream audiences.[31] As web commerce becomes dominant – again
assisted by Federal government policies – and there is increasing resort to
filters as a way of soothing mainstream fears over content, a similar risk
arises.
Recently, the US Senate passed a bill requiring all Internet service providers (ISPs) with more than 50,000 subscribers to offer upstream filtering.[32] Other bills have been introduced in Congress to mandate filters in libraries and schools.[33] Regulators – from legislatures to the courts – find it difficult to categorize the novel character of the web, complicating policies for socially constructive management of the Internet. Is the web broadcast, telephone or publication? Complexly, it is all of these, yet not any one. The strength of the Internet is in its mutability as multiple media, going quickly from one form to another as a one- or two-way device available to most anyone with modest equipment and a telephone line. With about 40 million Internet-connected US households (1999),[34] there is a mass reach platform for thousands of groups denied access to costly commercial media like television. Just as importantly, it is a way for minority opinion to reach out across great distances to gather their community on a world scale. Filtering, blocking and monitoring – absent adequate disclosure – stall these efforts, handing control of cyberspace to commercial forces or special interests that owe little save to their investors and content ideologues.
Notes