III. Content Blocking - Technical Aspects

Introduction

The earlier sections of this report have highlighted a trend towards self-regulatory frameworks to deal with illegal and harmful Internet content. One unifying feature of all such self-regulatory models is the perceived need for comprehensive technical blocking means for content. The primary focus in this section is the role of technical blocking methods in the context of the content regulation debate rather than an in-depth review of the technical strengths and weaknesses of existing types of blocking software.

Are such filters an essential adjunct to self-regulation or are they merely an optional piece of software for those users who decide they wish to proactively tailor the nature of the content they receive to their specific requirements? Current trends in the European Union and Australia would suggest the latter with a strong focus being placed on the PICS technology (see below).

This section of the report first seeks to examine technical approaches to content blocking and some of the issues which arise. It then looks in more detail at the Platform for Internet Content Selection ("PICS") which appears to be the emerging as a de facto industry standard and finally examines some of the challenges PICS still faces.

1. Technical Approaches/Issues
   This section assumes that the user's own ability to terminate an unwanted connection is insufficient - that the user must be alerted to the nature of a site (perhaps because of "adult" material) or must be refused access on legal or ethical grounds.

   In order for unwanted material to be blocked there are a number of steps which must be carried out:
   

   1. The offensive material must be susceptible of being identified;
   2. here must be software implemented at the appropriate point in the system architecture which is capable of carrying out such identification; and
   3. Having identified the material the appropriate response must be given (words blanked out error messages conveyed password prompt sought etc.).
   4. Each of these areas faces its own challenges even assuming that one can determine with certainty what material ought to be blocked.
   1. Material must be identifiable

      At present commercial blocking software operates on the basis of relatively simple identifiers of offensive material. In particular the software available today is capable of blocking specific manually blacklisted URLs or sites (such as the CyberNOT blacklist) or automatically blocking pages containing certain key words.

      Image recognition is still in its infancy and is likely to remain so for the foreseeable future.

      Any increase in the sophistication of blocking technologies or in the extent of their use has clear cost implications. These must either be borne directly by the consumer in terms of additional client software or in terms of recovery by an ISP of its cost of running proxy servers and filtering software.

      The difficulties of identification being limited to URLs labels and words are several:

      1. URLS

         Firstly, there are many examples of the reaction to the blocking of specific URLs being the appearance of numerous mirror sites (in the case of the German ISPs blocking Radikal (see Section II above) around 50 mirror sites were recorded). Secondly it is relatively easy for URLs to be changed more rapidly than the means of blocking them where they are for example used by a small number of paedophiles who can be individually notified of a new and rapidly changing URL. Responsiveness to such changes is more easily achievable with a proxy server rather than a client based filter.

      2. Labels

         For a labelling system to be effective the labels must be ubiquitous and capable of being read by the software implemented at the point of blocking. One of the major questions is how to ensure that labels accurately reflect the content they are labelling; any labelling exercise is largely subjective and it is necessary for the reader to have an informed understanding of the editorial stance both of a labelling system (i.e. what the labels imply) and of the labelling service (i.e. who is attaching the specific labels). There have recently been objections that some blocking systems (for example CYBERsitter ) have taken overtly political decisions such as blocking access to feminist sites in addition to sites that focus on topics such as adult issues and illegal activities. The introduction of the PICS standard (see below) is an attempt to overcome these problems.

      3. Words or expressions

         For software to be able to filter words or expressions or indeed (when such becomes commercially available) to identify types of images (naked flesh and so on) these need to be available in plain text or not encrypted in a manner which renders them opaque to the blocking software. In theory this makes them relatively easy to circumvent. However in practice the benefits of opacity are often outweighed by the disadvantages. Most of the resources that people want to block are commercial (usually pornographic sites) and these sites actively want to be seen especially by search engines. The techniques they use to make themselves visible to search engines such as padding pages with 'invisible' keywords are the very techniques which also make them obvious to blocking software. It is only where sites are happy to be invisible to search engines etc. that encryption can be readily implemented in order to avoid blocking technologies.

   2. Software must recognise for blocking purposes

      Conceptionally it ought to be possible for blocking software to be applied to any client with Internet connectivity or at a suitable point in the system (for example a proxy server or cache operated by an ISP or by a corporation in conjunction with a firewall). There will however be cost implications in terms of perceived system performance at the point at which the software is implemented.

      Clearly the software must recognise the relevant factor which enables identification (URL, label, word). This requires that the information to be identified is in a form compatible with the blocking software. The PICS standard endorsed by the W3 Consortium is an attempt to introduce an industry wide convention setting out a framework for describing ratings and labels so that software from different vendors can effectively exchange labels (see below). Prior to this development software ratings systems have had interoperability limited by the difficulty of ensuring that labels produced under one software could be read by a client operating another software.

   3. Appropriate response required to offensive material identified

      While 'access denied' is the most likely response to an item identified by a piece of software as offensive or illegal there are a range of more sophisticated editorial options such as individual words or images being removed. These may have unintended effects so that blocking at the paragraph level is likely to be more constructive . It could also be possible to trigger alarm or monitoring systems allowing sysops or others to become aware of attempts to obtain certain types of material. The use of this type of information clearly gives rise to concerns relating to privacy as well as possible criminal and civil liability for contributing to subsequent misuse or failure to report misuse.

2. PICS

   Since the commencement of this report the first generation of software filtering tools has arguably been superseded by the PICS technology. PICS is a set of open technical specifications for creating rating systems and compatible filtering software for Internet content which has been developed by the World Wide Web Consortium.

   This allows content providers to create rating labels for Internet content and embed them within that content. These labels indicate specific aspects of content such as offensiveness of language explicitness of sex and the degree of violence. Rating services on the other hand determine the substance of the labels by setting the rating criteria or values. The labels can be applied by content providers themselves ("self-rating") or by independent organisations ("third party rating"). In the first instance the labels are embedded in a site's HTML or expressed as part of the HTTP stream between the client and the site. In the case of third party label bureaus the labels can be expressed standalone and matched to a URL. Filtering software installed on a user's computer (or indeed the web browser) is used to automatically read the labels and block content that does not fit the "permissible content' criteria specified by the Internet user.

   The flexibility of PICS allows a single Website to have multiple labels applied by different rating systems. Users are free to choose which rating systems to install on their computer. They can also choose whether to block access to content that has no label at all or to override the block after viewing the material.

   In assessing the long-term viability of PICS there are two quite distinct factors which could both respectively seal its fate: technical shelf life and acceptance by industry and society. Industry in this context means the service providers and the content providers the browser and filtering software developers. Society means the users (who may also be content providers) the parents of users, policy makers, regulators and the conventional morality of the wider community. If PICS is to become the effective adjunct to self-regulation as is widely predicted a number of developments will be required in the short to medium term.

   1. Technical Shelf Life

      There appears to be widespread consensus that technically PICS is a major step forward in the field of content labelling. Indeed the PICS standard is fully operational and already being implemented by one major browser (Microsoft's Internet Explorer 3) and - to a lesser or greater degree - by most commercial filtering products (Cyber Patrol, PlanetWeb, Net Nanny, Net Shepherd and Safe Surf to name but a few).

      However, Internet related technical innovation continues at breakneck speed and there is always the danger that such innovation could at some point render PICS obsolete (or at least significantly less effective). The Leitmotif of PICS however is flexibility and it is this flexibility which is likely to ensure that it adapts to deal with new technical challenges. The PICS standard is the subject of ongoing development by the W3C and indeed now offers "metadata" or descriptive functions which will make it as useful for locating content as for blocking it. It therefore promises to offer further refinements and options developing "organically" to meet the needs of the industry (see below at Section b(3)).

      This report predicts that the industry will see a technical harmonisation of all commercial filtering software in the short to medium term with PICS becoming the global technical infrastructure for content labelling. However it is vital to distinguish what has been termed the "plumbing" of content labelling from the rating element which provides the framework for the moral and cultural value judgements to be applied to the content.

   2. Acceptance by Industry and Society

      If the technical viability and robustness of PICS is not currently in dispute the widespread uptake and acceptance of PICS is certainly contingent upon the availability of effective user friendly rating systems which meet both user and content provider requirements in terms of simplicity and flexibility and which - ideally - are universal in their application. Universal in this context means the capacity to balance local cultural requirements and specificity with an open infrastructure permitting compatibility with other systems and the mapping of one system's ratings to another. Until such systems become available it is likely that much Internet material will remain unrated with the result that users will be denied access to the vast majority of Internet material if they install PICS compatible filtering software and configure it to accept only rated pages. This is of course likely to prohibit the widespread uptake of PICS in the short term.

Prerequisites for effective rating
1. A Comprehensive System

   Should ratings be expressed as scores on a number of categories (e.g. violence language nudity sex) or should they be converted into recommended suitability for an age range as with films and videos? The RSACi system devised by the Recreational Software Advisory Council ("RSAC") is based on the above four categories and there appears to be some consensus that this approach is preferable to an age based system for reasons discussed at (2) below. RSAC has recently produced a revised methodology which sub-categories the above four in much greater detail however arguably other categories such as racism, blasphemy, drug use, hate, etc., must be covered by a truly comprehensive system. The need for comprehensive systems does not of course preclude specialist systems being developed in tandem which are tailored to specific moral ethical cultural or religious etc. requirements (e.g. the Jewish faith).

2. A Global System

   It is generally acknowledged that rating systems cannot stand alone; they must be capable of using and interpreting other systems to build up a wider catchment of rated sites. Unique local systems may work for small groups of users with a particular interest but a stand-alone national system is unlikely to overcome the "chicken and egg" problem that users will not switch on a system which only reaches a few sites while site owners will not rate with a system which has few users. The UK is currently formulating a proposal for such a system and hopes to collaborate with bodies in the US Europe and Australia all of which have expressed interest in this work (see Section 2 below).

3. An Accountable System

   A frequent rhetorical question heard in the context of the rating debate is "who will rate the raters?" There is concern that the discretionary self rating systems which PICS permits may be open to abuse in terms of negligent or bad faith rating by content providers.

   It has been stated that "One of the most critical aspects of any self-regulatory regime is the lengths that it goes to ensure that people aren't cheating the system" . While this statement was made in relation to self-regulation of computer games it applies equally to Internet content. Indeed RSAC's Internet rating system RSACi currently the leading database of PICS labels freely available to the public operates a system of spot checks to ensure self raters rate accurately. If they do not RSAC has contractual rights allowing it to take legal action against those who wilfully mis-rate. This seems a good model for self-rating.

   In the case of third party rating services the reputation and credibility of such systems will rest on their track record for providing consistently accurate ratings in accordance with their chosen rating system(s). Therefore accountability will be less of an issue. However no doubt in time there will be rating services to rate the rating services! Ultimately however de facto standard systems will emerge as users decide which systems best meet their individual needs and this is precisely the aim of PICS; to provide users with maximum choice and control.

4. Content Producer Cooperation

   It was stated recently that 'the biggest problem (with PICS) is that although many companies are now supporting PICS in their business the folks who create Web site development tools don't seem to have even heard of the concept . However the same commentator predicted that once Web development tools allow PICS labels to be generated semi-automatically the task will appear less onerous and time consuming to content producers.

   The prevailing view amongst content providers is that they will be willing to label content provided rating is made quick cheap and easy.

   In addition the above presupposes that an acceptable rating system is in place providing them with a framework for rating.

   If third party rating services do become the most popular method of obtaining labels then self-rating may gradually become redundant. This is not likely in the short term however.

5. Browser Compatibility

   For PICS to really take off it needs to be supported in all major browsers as many users may not want the additional expense of purchasing a filtering product. This is not currently the case although Netscape promises that a future version of Netscape navigator will support PICS details to be announced fairly soon.

6. User Confidence

   The key to ensuring that PICS wins over users is to achieve factors (a) - (e) above. In other words once several rating systems become accepted as industry standards trusted labelling services are established labelling tools exist to facilitate the generation of labels and all major filtering software products and browsers are fully PICS compatible then users will have the requisite confidence that all parts of the PICS equation are in place.

   In the meantime it is vital that the industry concentrate on educating users and raising their awareness of the existence of PICS in tandem with the above developments.

A possible model for a rating system

1. Internet Watch Foundation (UK)

   The recently formed Advisory Board of the Internet Watch Foundation ("IWF") is currently investigating a way forward to bridge the gap between the objectives of having a mechanism for rating and blocking content sensitive to the needs of UK culture and of having an internationally accepted system to promote widespread usage and allow users access to the maximum amount of global content. The report focus on the work of IWF as it appears to be the most ambitious scheme for a universal rating system undertaken to date. The proposal is based on the difference between describing and blocking content and has been summarised as follows by one of the Advisory Board members.

   1. Develop an agreed international methodology for describing content

      There needs to be a common basis for describing content. This will first require establishing a set of factors of concern - e.g. nudity, language, violence, etc., and second an agreed system for calibrating a scale of the extent to which any factor is present. From this can be developed a common methodology for content producers to place their content within the matrix of factors.

      While no such a matrix can ever be value free (the selection of the factors of concern inevitably is a result of different cultural religious and social influences) the aim is to have a common international basis for describing content which is as objective as possible. It may therefore be necessary to include factors that some cultures consider unimportant to obtain maximum agreement.

      Such an objective methodology must crucially choose a calibrated scale of how much a factor is present that is unrelated to the age of potential users. Different cultures have very different views about what is suitable at different ages. A simple numeric scale against agreed criteria is preferable.

      Content producers will rate their web sites using the agreed methodology and insert a PICS label into the URL. This will create a vast international data set describing global Internet content.

   2. Adopt universally available PICS compatible blocking software

      Users need to access the data set of content labels and can do so using browsers and blocking software compatible with PICS. The current version of Microsoft Internet Explorer with the RSACi labels has already demonstrated the possibilities of this approach. Individual users around the world have the capability to set their browser against all the factors agreed in the international data set. Such an approach gives maximum freedom of choice to users. For example a Muslim in Saudi Arabia could set a very restrictive block on all the factors of concern including one that recognises modesty of clothing and a user in Northern Europe could take a more relaxed approach on all or some of the factors.

   3. Make available a variety of "blocking profiles"

      While the maximum choice of factors option outlined above is attractive some users and some cultures will find this bewildering. Some users will be anxious to have advice as to what constitutes a 'safe' setting and in some countries there may be a desire to relate the description of content to existing recognised rating systems. These desires can be met if software profiles are produced for individual countries or cultures which draw on the information in the data set. Profiles are in fact a technical modification of PICS discussed below in (3).

      To take a hypothetical example of a profile in the UK a software block could be devised based on the widely recognised age related video classification system. The classification body could examine the data set and determine for example that an approximation to a 12 age rating was a score of 2 on language, 1 on nudity, 1 on sex, etc. Thus a UK profile could be built which matched the data set to pre­existing video classification criteria.

      While such a process will inevitably mean losing the fine tuning offered by the user setting such factors individually it may provide an additional guide and tool for many people that they find helpful. Moreover it does allow sensitivity to cultural differences.

      An analogy which may help understanding this process is the choice consumers make in adjusting the settings for a central heating system sold in a variety of countries. Such systems allow a number of different factors to be controlled -thermostat temperature setting length of time boiler operates whether it comes on once or twice whether water only is heated or both water and heating etc. Most users accept the manufacturer's recommended settings for their particular climate and needs knowing that they can adjust individual factors if they want. Others may adopt settings suggested by energy conservation organisations. In the content blocking case users might be offered the opportunity to consider adjusting all the settings individually or accepting a recommended profile.

      In time third party rating services will undoubtedly offer 'off the shelf' profiles as part of their service.

2. Ongoing challenges for PICS

   The question is whether the W3C developers can keep up with Internet developments and adapt the PICS technology swiftly enough to allay user concerns that (1) PICS can effectively filter all content and (2) that if it cannot they are not denied access to large amounts of valuable material merely because it is unrated.

   1. Rating Information Push Services

      One of the most significant recent Internet technologies is Information Push where a user subscribes to an information "channel" and content is downloaded to the client machine automatically. Companies that are providing this kind of service include PointCast Castanet by Marimba BackWeb and Tibco's TIBNET. The method used to get the content to the client varies (not necessarily HTTP and HTML-based) as does the structure of the channels. This raises the following questions:

      • How would the actual channels themselves be rated in particular within the current PICS framework?
      • How would the labels get to the user?
      • How would their rating of individual "articles" within a channel be achieved?
      • If the user has a preference file for blocking channels/articles how can we ensure this is compatible with all Information Push services?

      These issues need to be investigated and the W3C PICS working group is already addressing them.

   2. Newsgroup Ratings

      It is the stated intention of IWF in the UK to produce composite ratings for all newsgroups. This will require a ratings service to sample groups and produce the ratings and a ratings bureau to maintain a rating file. It has been proposed that IWF should facilitate these actions using volunteers - both from ISPs and independent raters. Although there are ways of reducing this mammoth task (currently about 22,000 news groups) it is still a daunting undertaking. The outcome of the work may either be a commercial product/service or freely available to the UK industry and its subscribers.

      The IWF is also spearheading the INCORE (Internet Content Rating for Europe) proposal which has been submitted to the EU Commission as part of a bid for EU funding. One of the three key aims of INCORE is to address the existence of illegal material on Usenet newsgroups at European level and develop a rating system for Usenet. Several technical approaches are being considered some of which are based on the PICS technology.

   3. Profiles

      A ratings profile is a set of filtering rules that allows the user to select or block content from the Internet based on its URL and PICS label. Profiles are intended to be simple to use and portable allowing the user to configure their filtering software easily and share their profiles with other users. There are several ways that profiles could be used:

      1. the user could install a pre-determined profile from a trusted organisation whose values they agree with - this would make it easy for the user to select Internet content without having to write their own profile;

      2. profiles could be sent to search engine servers proxies and other Internet services as part of the user's query or request. This would allow the server to only return information that fits with the user's profile preventing invalid links getting to the user;

      3. users could use many different software filters on their machine but use the same portable profile with all of them.

      Profiles are written in picsRULZ a common profile specification language. It allows various types of rules to be written that block or allow access based on the URL of the resource and its PICS label.

      The technical specification for PICS profiles still requires some fine tuning however this work is in progress at the W3C.