Experiences from the Internet Content Rating Association (ICRA)
How not to fail with a global labelling system
GAM Task Force Roundtable
Düsseldorf 11 - 12th March 2010
Slides at http://philarcher.org/icra/
Original document also available.
I was:
- ICRA's technical guy from April 2000 - November 2008.
- Responsible for various initiatives including:
- European projects designed to promote labelling;
- new labelling and filtering tools;
- developing better labelling technologies.
- The most visible ICRA person in Europe (after 2002).
I now divide my time between:
World Wide Web Consortium (W3C), working on mobile Web & augmented reality;
i-sieve Technologies: providing quantitative and qualitative sentiment analysis.
ICRA was designed to:
- Create a Web site labelling system that was neutral and objective and therefore international.
- Be interpreted locally according to an individual's cultural preferences.
- Support self-labelling and, by extension, self regulation.
- Be implemented in browsers with a parent-friendly interface.
ICRA had:
- Substantial support from the online industry.
- Political and financial support from the European Commission.
- Political support from various governments.
ICRA failed
Although the buck doesn't stop with me, I'm not passing it on either.
Reasons to be cheerful about GAM
- The timing is good - Patrice showed lots of political support;
- the technology for Stefan's "Youth Protection API" is well-defined (linked data);
- GAM has attracted the support of film & game classification bodies.
Comparing ICRA and GAM
Let's look at:
- what ICRA got right;
- what it got wrong;
- what others did and didn't do.
What ICRA got right
The theory was that self-labelling would:
- establish a direct communication channel between the content provider and the parent;
- provide for parental control without the need for, or the accusation of, censorship;
What ICRA got right
The vocabulary.
- Descriptors were binary.
- Each defined characteristic was present or not.
- First published in 2000, the vocabulary was updated in 2005 and was scheduled for a further update in 2008.
- We learned from experience, making the vocabulary easier to use and less open to misunderstanding.
The vocabulary was widely recognised as being good.
What ICRA got wrong
We believed that other people would create filtering templates - i.e. a set of
rules for the types of content that would be blocked and allowed - and that
parents would be able to choose the template offered by organisations they trusted. For example:
- film classification bodies;
- other child protection organisations;
- youth organisations;
- brands wishing to promote their family-friendly values;
- ISPs;
- individual citizens.
Importantly, each template provider would act as a channel through which labelling would be promoted.
What ICRA got wrong
It didn't happen.
With hindsight it's obvious that no brand is ever going to put its name on a filtering product that isn't able to offer
strong guarantees of accuracy and efficacy.
What ICRA got wrong
Attempting to offer consumer software.
What others did and didn't do
- Very few people actually labelled their site. We never did have accuracte figures but my
personal best guess is that there were never more than about 10,000 labelled sites (a tiny number in Web terms).
- A handful of minor software companies recognised ICRA labels in their filtering products but there was never enough data for the
big filtering companies to need to pay much attention to labelling.
- The majority of ICRA member companies never labelled their site.
The problem with labels
What do you do about unlabelled sites? You either block them all or allow them all.
You can create lists to work alongside labels but to do that you need automated classification.
So you may as well just use automated classification and not bother with labels.
The problem with labels
Why bother?
Things ICRA tried
- A standalone simple tool for generating labels (today you'd call it an App or a widget).
- A cheap membership scheme aimed at Web developers who would label all the sites they created in return for recognition and publicity.
- A validation service (subject to a small fee we'd check the label was accurate and, if so, publish this assertion as data).
So we have to get smart
Labels can be useful if:
- Web sites that carry a label offer users a better experience than ones that don't;
- labelling a Web site requires no effort at all by the content provider;
- labels can be trusted.
I would also argue that for labels to be successful, they should be created for a variety of
reasons and read by a variety of software, at least some of which are unrelated to child protection.
So we have to get smart
Labels can be useful if:
- Web sites that carry a label offer users a better experience than ones that don't;
- labelling a Web site requires no effort at all by the content provider;
- labels can be trusted.
I would also argue that for labels to be successful, they should be created for a variety of
reasons and read by a variety of software, at least some of which are unrelated to child protection.
More detail on these thoughts is availabe in a separate document (170 Kb PDF).
These slides at http://philarcher.org/icra/.
Phil Archer
phil@philarcher.org
http://philarcher.org