Debunking Automated Copyright Filtering
A couple of stories of interest. Cory’s latest Guardian column explores all of the practical hurdles an automated system must overcome to preserve the same balance of values as a manual system. If perfect control were in anyway relevant to the discussion of copyright enforcement, the issue would be relevant. Sadly, years of attacking perfect control advocates within the DRM debate have not translated to any traction with what Big Content no doubts sees as a separate debate. So we have to make all the same arguments again about the cost to the average, law abiding consumer for a dedicate minority of scofflaws who will circumvent every attempt at perfect control, often just to show it can be done without any real interest in the content to be protected.
Meanwhile, the risk of chilling if not outright excluding fair uses is substantial. This discussion is not any different from the risk assessment for any automated system. Facial recognition for catching terrorists. Signals intelligence for identifying threats. Any time you remove human discretion from the equation, to do what we do well, to deal with exceptions, then the cost of a false result raises substantially. It doesn’t matter if the implementer of the filter includes a complaint system or not. The damage is done in the cementing of our permission culture.
I made a similar complaint about using micro payments. If I want to exercise my fair use right, to promote or comment, as a form of free speech and practically a free as in beer benefit to the creator, why should I have to pay such costs?
The second story is that the EFF has assembled a set of videos it thinks should not be blocked by automated systems. This is a tough opinion to back since fair use rights are only protected after the fact, assuming a complaint has been raised, via an affirmative defense that has to be argued by human beings.
It would be easy to suggest that filter makers could tune their systems to this set, but I think that misses the point. No sample set will ever be large enough for that exercise to be useful and with the current rules on fair use, machine intelligence just isn’t up to the task to begin with. How can any algorithm we know how to implement today correctly identify the character of the original work and the nature of a derivative work? How could it encompass market factors, wholly external to the individual piece in question?
Here’s a simple question, if this sort of discretion is so hard to code, then why aren’t we hearing many, if any, in Big Content suggesting a discussion on evolving new standards of fair use that are a reasonable compromise or offer greater advantages to both sides?
