from the demon-drink dept
Against a background of the UK government teetering on the brink of imposing an opt-out Web filter "for the children", here's yet another example of how automatic categorization of sites for blacklists gets it wrong, as recounted by the UK's Open Rights Group (ORG):
someone used blocked.org.uk to tell us about another church (St. Mark's in Southampton) that is blocked -- this time on [the mobile operator] Vodafone. We have confirmed that it is also blocked by Orange. The site is blocked on O2's highest blocking setting, but not on their 'default safety' service.
So why might a church be classed alongside sinful purveyors of alcoholic beverages? ORG has a suspicion:
Using O2's very handy 'URL checker', we have established that they classify the site as 'alcohol'. It is likely that this is the category that has led to its blocking on other networks, but this is not confirmed.
It is likely that the reason for this categorisation is the use of the word 'wine' on the church's website. The church is part of the 'New Wine Network of Churches'. Their website explains that this means they "have the aim of 'Equipping Churches to see Jesus' Kingdom Grow'". Their use of the word 'wine' is not related to selling or the use of alcohol.
Although it seems that the site has now been unblocked, that's only because it was "manually reviewed". As ORG points out:
It's yet another example of how internet filters make simple and costly mistakes which often result in 'over-blocking.' Our report from May this year collected more examples of this. Since then we have seen political parties, technology news websites, and more recently a number of maternity health sites all blocked by mobile networks. It can be tricky and slow to get sites removed from block lists (although mobile networks say this is improving).
That last point is important. No system is perfect, and errors will always be made. But what matters is how quickly the mistakes are corrected. Unfortunately, the evidence so far is that not only are such automated filters unreliable when it comes to evaluating sites, but the correction mechanisms are pretty awful too -- a worrying combination.