C. Dianne Martin, RSAC President, George Washington University

Joseph M. Reagle, Jr., WC3 Group. Massachusetts Institute of Technology

There are approximately 750,000 on-line users below the age of 18. A recently pronounced goal in the United States for the National Information Infrastructure (Nll) is to enable it to provide a level of education to all students that surpasses the highest levels of education available today. Throughout the history of the Nll, education and research were a key motivation for the development of the technology, first as the ARPANET, then the Internet, the NREN, the Nll, and as part of the United States Department of Education project GOALS2000. Many of the recent initiatives (from about 1989 onwards) have also focused a great deal on the educational capabilities of these networks for K-12 (grade school) students. In addition, a significant reason for the presence of young people on the Internet has been the explosive growth of on-line services and Internet access, especially through services such as America On-line (AOL), CompuServe, and Prodigy. Ironically, this surge of new users has also brought an increase in the availability of adult-oriented content and services, much of which is considered inappropriate for young people.

For those that find this alarming, the situation is further complicated by other Internet controversies involving censorship, anonymity, and government control; the decentralized nature of the Internet; and ill informed media attention. Hence, those who are sincere about preventing censorship on the one hand and enabling legitimate parental control on the other hand are left in a difficult position. One solution that has been proposed that will meet the dual goal of non censorious content selection and screening has been content labeling. Several different labeling schemes now available allow Internet content providers to either self label or to be labeled by third parties with respect to any number of attributes. The areas of greatest concern relate to attributes such as sex, violence, nudity, and language.

In 1994, Senators Joseph Lieberman (D-Conn.) and Herbert Kohl (D-Wis.) chaired a number of Senate hearings regarding the increasing levels of violence in computer games. To address these concerns and to deflect possible government regulation of this media, two major content classification systems for interactive electronic entertainment were developed in the United States These are known as the Recreational Software Advisory Council (RSAC), developed by a coalition of over 25 organizations led by the Software Publishers Association (SPA), and the Entertainment Software Rating Board (ESRB), sponsored by the Interactive Digital Software Association (IDSA). Both were established in 1994.

Both groups are independent, non-profit organizations, but the two content advisory systems are fundamentally different from each other. The RSAC system is a content-based advisory system based upon self-disclosure using an interactive ratings package. The ESRB system is an age-based advisory system based upon the decisions of a rating board. The RSAC system has been used mainly by manufacturers of computer games, while the ESRB system has been used for both video platform games such as Sega and Nintendo and computer games.

To understand the RSAC labeling system, it is first necessary to understand content advisory systems in general. The basis of any rating system is the way in which it classifies content. Federman(1996) has used the terms "descriptive" versus "evaluative" to characterize content labeling methodologies. In addition, Reagle et al (1996) have used the terms "deterministic" versus "non-deterministic" to characterize the labeling process itself. They also introduce the dimension of voluntary versus mandatory to the rating process. These terms can be defined as follows:

No rating system is purely descriptive or deterministic. Rather, each system varies with respect to where it falls between extremes. Our usage of these terms is with the understanding that no system is completely without bias or arbitrariness. Most people are familiar with the Motion Picture Association of America (MPAA) rating system in which a board of reviewers examines the content and then issues an evaluative, non-deterministic rating. The process is non-deterministic because, while general rules of thumb may guide the reviewers' decisions, the process itself is opaque and the results are sometimes at odds with other ratings. It is evaluative because the ratings do not describe the content of the film, but what age group may see the film.

In contrast to the MPAA, the RSAC system is voluntary with specific deterministic criteria by which content is rated in a descriptive manner. Content producers, such as video game makers, answer a detailed questionnaire (either in paper or electronic format) about their content with respect to violence, nudity, sex, and language. RSAC then processes the questionnaire, registers and returns the consequent rating to the company. The company is able to use that label in advertising or on their product. The label consists of a number, between (0-4), for each of the four categories. A rating of All (0) represents the minimum amount of objectionable material. The system is represented in graphical form by a thermometer. The number, or the temperature of the thermometer, informs the customer about the specific content of the package as is demonstrated below in the RSAC advisories for violence:

RSAC Advisories on VIOLENCE

0: Harmless conflict; some damage to objects

1: Creatures injured or killed; damage to objects; fighting

2: Humans injured or killed with small amount of blood

3: Humans injured or killed; blood and gore

4: Wanton and gratuitous violence, torture, and/or rape

The RSAC system does not say for whom the content is appropriate, it merely describes the content with respect to characteristics that may be of concern to parents. Since content providers fill out the questionnaire, it is a self-labeling and voluntary system. To ensure public confidence in the RSAC system, the content producer is contractually obligated to rate the content accurately and fairly. Every month a number of registered titles are randomly sampled. Producers who have willfully misrepresented the nature of their content may be fined up to $10,000 and may be required to recall their product from the shelves. Using this system, RSAC has rated over 350 game titles with 94 companies including the popular "Myst" by Broderbund, "Doom II" by id Software, and "Dark Forces" by LucasArts. Only two companies have ever requested an appeal, and so far no suits have been filed for misrepresentation.

During the year leading up to the passage of the Computer Decency Act at the end of 1995, a number of Internet specific labeling activities occurred: 1)the U. S. Senate Judiciary Committee heard testimony regarding the "Protection of Children From Computer Pornography Act of 1995" (S. 892); 2) the Information Highway Parental Empowerment Group (IHPEG), a coalition of three companies (Microsoft Corporation, Netscape Communications, and Progressive Networks), was formed to develop standards for empowering parents to screen inappropriate network content; 3) a number of standards for content labeling were proposed including Borenstein's and New's Internet Draft "KidCode" (June 1995), and 4) a number of services and products for blocking inappropriate content were announced, including Cyber Patrol, CyberSitter, Internet Filter, NetNanny, SurfWatch, and WebTrack.

By August, much of the standards activity was consolidated under the auspices of the World Wide Web Consortium (W3C) when the W3C, IHPEG, and twenty other organizations agreed to merge their efforts and resources to develop a standard for content selection. The result of the agreement is the Platform for Internet Content Selection (PICS) standard that allows organizations to easily define content rating systems and enable users to selectively block (or seek) information. It is important to stress that the standard is not a rating system like MPAA or RSAC, but an encoding method for carrying the the ratings of those systems. Those encoded ratings can then be distributed with documents or through third party label bureaus.

To aid the rating of large sites, labels may apply to whole directory structures (hierarchies) of a Web site if the label is appropriate to all the content. Labels can also be put on individual web pages or individual assets on a web page. This flexibility to rate at different levels is referred to as the granularity of a particular rating. The following example demonstrates a label for an RSAC label of langage (l=3), sex (s=2), nudity (n=2) and violence (v=0):

The PICS encoding specifies the rating service, version number, the creation and expiration date, the page, the rater, and the ratings themselves (other options may be specified but are not shown). Multiple labels can exist for any page. Labels can be included in html documents within the meta-tag, they can be fetched from the http server using the http get command, or they can be fetched from label bureaus. Hence, the author of a homepage could include a variety of labels on the page itself (ie, the RSAC, MPAA, or Golf-Fan systems). The http server on which the page resides could have a label or labels for that particular page, and a third party label bureau like the "Good Housekeeping Seal of the Web" could be queried for its opinion of the quality of the Web page.

In the rapidly evolving market of the Internet, label systems and services have a significant stake in maintaining the public confidence in the authenticity of their ratings. Malicious users who falsely label content could damage the reputation of a service, a rating system, or PICS in general. To prevent the manipulation of labels or the content to which they apply, PICS includes the capability to ensure the integrity of a label using message integrity checks (MICS) and its authenticity using digital signatures. In this way, compliant browsers can ensure that a document has not changed or been manipulated since the labeling of the document and that the label is genuine. An important part of PICS compliance is the requirement that PICS compatible clients read any label system definition from a user accessible configuration file.

In April 1996, the RSAC rating system was adapted for Internet content under the name RSACi using the PICS encoding standard. The RSACi system is a Web-based questionnaire that queries the user about the content of a Web page or directory tree based upon the content categories shown in Figure 1.

_________________________________________________________________________

___________________________________________________________________________

Upon completion of the questionnaire, a PICS meta-tag similar to the one shown previously is returned to the user to be placed in the file header. There is also the option to place the RSACi symbol on the web page. The service does not currently provide message integrity checks or digital signatures. This service is currently free to anyone interested in labeling the contents of a web site. It is expected that many of the attributes of the previous RSAC system will be extended to RSACi, including the sampling of sites for labeling veracity and compliance with the terms of service that a user agrees to before receiving the label.

RSAC'S potential role in the labeling of Web content is complex. Just as the production and distribution of Web content is more than a matter of placing an html document on a server, RSACi and other PlCS-compliant rating systems are more than the voluntary insertion of labels into documents by their creators. This simple act is only the first step in a strategically and technically complex flow of information from origin to destination. This section presents an analysis of RSAC'S relations to the production and distribution of content.

The production and flow of content is neither a vertically integrated production chain - the same people who create the content do not necessarily provide the conduit and browser - nor is it a purely distributed and segmented market. Although this market is highly compartmentalized, the need for market efficiencies will drive the creation of strategic alliances and standards between functional domains (such as on-line companies and browsers). This consequently affects the delivery paths and quality of content. Included in this rapidly evolving market are content producers, content hosts, other rating services, bots, search engines, directories, filters, Internet Service Providers (ISPs), on-line services, protocol developers, and browser/software companies (see Figure 2).

Content Producers: Commercial and non-commercial developers of Internet information and web sites; they can range from single individuals to huge multinational corporations. They may or may not have incentives on their own to provide content advisories using a system like RSACi with the information they produce.

Web Farms/Content Hosts: Web farms and content hosts provide server services to individuals and organizations that lack the means or interest to support their own server. As a defense against charges of harboring objectionable material without proper safeguards, these entities may encourage or require content developers to self label. For example, CompuServe has endorsed the RSACi system through an implementation with CyberPatrol and has encouraged individual and institutional content developers on its systems to employ the RSACi labeling system.

Search Engines and Agents: Search engines and agents lay outside of the direct path of content flow - one does not need a search engine. However, they often provide an important value added service in channeling and selecting information. As such, search engines may gain from being compatible to PICS because label information may improve searching and indexing capabilities. This in turn may be a further incentive to content developers to adopt RSACi and other PlCS-based rating systems.

Bots: Bots travel from site to site retrieving information of interest to their owners. Since bots are personal, discriminatory spiders, their ability to search and retrieve content with content labels has implications similar to that of search engines As they gain the ability to communicate with each other (one could now call them "agents"), PICS compliant labels could become the language for communicating about the preferences of their owners.

Internet Service Providers (ISPs): Internet Service Providers provide the means for connectivity from one point on the Internet to another. They have been viewed by governments as convenient points of control. Legislators have been eager to make ISPs legally responsible for the material they carry. Since ISPs have been a focus of much of the controversy, they have been very interested in adopting or supporting content labeling systems such as RSACi.

__________________________________________________________________________

__________________________________________________________________________

Browsers: Browsers are used to access information on the World Wide Web. At the time this was written, Microsoft has incorporated the RSACi PICS implementation into its most recent browser product, the Microsoft Internet Explorer. The value of such an agreement for browser companies is that it addresses parental and institutional concerns about restricting access to inappropriate material. One point of particular interest is that while many of the PICS recommendations will be implemented by these and other browsers, the companies have thus far declined to implement signature verification of the labels, an omission that may put the trustworthiness of RSACi and other PlCS-compliant systems at risk.

On-line Services, Firewalls, Proxies, and Intranets: These categories include both publicly accessible (AOL, CompuServe, Prodigy) and private/corporate networks. This market has been particularly concerned with inappropriate material. While ISPs have argued for common carrier status, on-line services have invested significantly in the creation of a family orientated image. Hence, they have been the quickest adopters of content selection and screen software - such as SurfWatch and CyberPatrol. Corporations are also concerned about the inappropriate activity on their networks, and some are seeking the ability to monitor or screen the activities of their employees using systems like NetShepard. Network services such as Intranet servers, firewalls, and proxies are also points of control for the dissemination of information to an organization.

The relationships between all of these entities may be looked at as more of a plumbing system made of reservoirs (containing a variety of liquids), conduits (with a variety of delivery capacities, operating pressures, and flow rates) and control systems (upstream versus downstream regulation), with filtering mechanisms interposed at various points in the plumbing. At each step, information may be redirected, collected, or amplified by a value added service. Companies can take advantage of strategic opportunities for increasing market efficiency and strengthening their position in the market. Given this interesting information flow structure, the relevant question is which domains (and their boundaries) will be of the greatest significance to labeling services?

Some browser filtering systems have similarities with the RSACi system in that they are PICS compliant and content descriptive, but they may differ in significant ways. In the case of SafeSurf, it’s rating system provides an example of a PICS compliant system that is more evaluative than the RSACi system:

1) it includes an appropriateness rating with regard to age

2) it provides descriptive labels that have highly judgmental definitions and descriptions.

Other methods for content filtering include mechanisms like SurfWatch which maintains lists of URL's with objectionable content. NetNanny has filters which block objectionable material (such as curse words) in real time. Although non-RSAC filtering mechanisms may be synergistic in some cases (meaning they may be able to cooperate at some levels), these blocking technologies are different from the RSACi system because they:

1) require proprietary software

2) are labor-intensive

3) are not extendible to other areas of concern or interest

4) realize no economies of scale as the volume of content grows

5) employ standards that are obscure, somewhat arbitrary, and ultimately restrictive

Instability: The process of content screening and selection will continue to be highly unstable for the near future. One must remember that it is only within the past year that many of these standards and services became available to users of the Internet. As an example of the tremendous pace of events, consider the case of CompuServe. CompuServe has offered SurfWatch as part of its Internet in a Box, a suite of Internet access applications including software from Spry. A competitor of Spry, SpyGlass, has now bought SurfWatch!

Digital Signatures, Intellectual Property and Market Brand: Elsewhere we discuss digital signatures with respect to the PICS standard. To engender pubic trust in labeling systems, any organization like RSAC must ensure that its labels correspond to the content, and that no unauthorized content developers use their labels and their respective icons. On the Internet, while trademarked GIFS may be of some advantage in creating brand recognition, the important "content" with respect to selection software will be the validity of the rating that is accessed by the content seeker. How easily can this text be misappropriated? If a digital signature is provided by RSAC and checked by the browsers for authenticity, it is very difficult. If digital signatures are not incorporated, it can be misused very easily. One could create such a label for an adult Web service without consulting the RSAC questionnaire, and one may do so with malicious intent. Hence, simple encryption technologies would seem to provide the only protection to widely-used labeling systems.

International Issues: The threat of governmental censorship of electronic media provided the main impetus for the formation of RSAC and the development of PICS. Until this point, we have only considered this issue with respect to the United States. However, an oft cited characteristic of the digital realm is its global scope. This can increase the difficulty of developing a content labeling system because the cultural norms of violence, language, sexuality, and political freedoms differ across the globe, and there are no cultural boundaries in cyberspace. Hence, content which may be considered appropriate within one culture, may be considered inappropriate to others. Governments have been attempting to legislate technical infrastructure requirements because of indecency or cultural concerns.

An immediate difficulty with evaluative labeling systems is that what may be appropriate for one culture may be highly inappropriate for another. Fortunately, the PICS system allows for multiple rating systems, services, and label bureaus. As an example of a potential problem, consider the aversion for Nazi propaganda by the German government. Without requiring draconian regulation of infrastructure or ISPs, Germany could require that all browsers and lSPs use a labeling system and label bureau for filtering information pertaining to Nazism. All PICS compliant browsers must be able to read label system definitions from a configuration file, and the government could be responsible for developing the appropriate rating and labeling services. However, this technique can also be extended even further by totalitarian nations such as China to filter sensitive information, if all access is required to go through gateways that employ filtering software.

Regardless, RSACi has an advantage in the international market because systems that use straight forward content description rather than age appropriate evaluations will have greater applicability and adaptability across multiple cultures. While there is some cultural bias within the RSAC system, efforts to extend the system while keeping it very content oriented would allow it to have international scope. Some countries may associate different icons or names with the ratings differently, but the numeric value of a descriptive rating would stay the same. Potentially, this would extend usage of the RSACi system beyond the United States to become an international content labeling service.

A common saying among those that study the Internet is that, "three months are one Web year." However, there are a number of observations one can make about content labeling today. One observation is that this market is extraordinarily dynamic. Many of the filtering companies discussed in this case study are one to three years old. Some of the companies will likely go out of business, or be purchased or bought by larger content or infrastructure organizations - as has happened with SurfWatch.

The dynamic nature of the Internet leads one to realize the importance of balancing healthy competition with cooperation on sensitive social issues between the entities discussed. With the chaotic development and flow of information on the Internet, it is also important that standards such as PICS are being adopted at each level of information delivery to bring some sense of order and control to concerned users. It is in this spirit of cooperation that disparate organizations such as RSAC and Microsoft have worked together to use the PICS encoding system to develop a content labeling and blocking mechanism and to make the system available as widely as possible. The ultimate goal of such content advisory systems is to provide a technical alternative to government regulation and censorship of the internet and to empower members of the public to make informed decisions based upon their own value systems about the appropriateness of content when accessing the Web.

Federman, Joel. Media Ratings: Design, Use and Consequences. Mediascope, Inc. Studio City, CA. 1996.

Reagle, J. M., Jr., Evans,M. and Shareck, P. "RSACi Case Study". Electronic Commerce and Marketing Course, MIT’s Sloan’s School of Business Management, Boston, MA. 1996.

RSAC Homepage, http://www.rsac.org/ and The Recreational Software Advisory Council background web page http://web.mit.edu/reagle/www/commerce/singles/rsac7.html /, 1996.

Much of the background information in this article was developed as part of a case study on RSACi during an internship at RSAC by Joseph M. Reagle, Jr., Michael Evans, and Patrick Shareck for an Electronic Commerce and Marketing Course at MIT’s Sloan’s School of Business Management, Boston, MA during the spring of 1996.