False Positives, Real Problem – The Limits of Filtering as a Solution to Copyright Infringement

The question of how to deal with technological change lies behind much debate around copyright at the moment. While copying and distributing works used to require investment-heavy equipment, now anyone can do it, making it undoubtedly easier to pirate films, songs and music.

Concern about piracy is often linked with criticism of the major Internet platforms which allow for users to upload and share content. It may help that they are usually easier to find, and have deeper pockets, if ever there is to be compensation or a settlement.

While the link between the biggest platforms and piracy is not as clear as some make out, it is understandable that the wealth and profit margins of the biggest platforms can lead many to a feeling that something is wrong. Nonetheless, there is a difference between using a work legally but not paying enough, and not using it legally in the first place.

The Answer Depends on the Question

In terms of responses, these tend to depend on the framing of the problem. If the issue is market power (i.e. that rightholders are too dependent on the power of the platform to be able to negotiate), then the answer is competition law. If it’s that major platforms are using their global presence to avoid tax, then the answer is better international coordination and cooperation to stop this.

But then there are those who portray things in terms of major tech companies making money from advertising around pirated content. Ironically, their response to a situation ‘created’ by technology seems to be more technology.

To use ‘filters’ trained with artificial intelligence to recognise copyrighted works, and stop them being uploaded. Or, if a platform refuses to use them, legal action. This is the position promoted by some in the ongoing European Union copyright discussions.

There is a human rights argument against such an approach – primarily that expression on the internet via platforms should not be conditional on a prior check. There is no truly free expression under such a system, just as there wasn’t in the past in authoritarian regimes that steamed open letters to read them before they arrived at their destination. See IFLA’s response to the call for comments by the UN Special Rapporteur on Freedom of Expression for more on this.

Tech for Bad, Tech for Worse?

The technological issues tend to compound the human rights ones. This is because, while progress is being made on developing artificial intelligence that can recognise copyrighted works, there seems to have been a general disregard for their ability to replicate the balance that exists in law, at least where there are exceptions and limitations.

The evidence that tools deployed are able to identify legitimate uses under exceptions and limitations to copyright, for example satire, criticism, incidental use or broader fair use is limited.

To use language from medical research, the number of false negatives (occasions when a filter misses an infringing use) may be low. But this is not the same as ensuring a low number of false positives, where a filter wrongly identifies a legitimate use as infringing (see examples in this article).

Moreover, the incentive to reduce this number of false positives will not be high when there no price to pay for preventing legal uploads, but the costs associated with false negatives are significant. Of course this is not a concern if the goal is to stop infringing content making it to the Internet, regardless of the collateral damage done. For those who do worry about this, notice and take-down appears to offer a better way forwards, at least for now.

The Library Dimension

For libraries, there is an immediate risk to those who encourage users to be creative and upload their work to platforms. Those operating repositories – for example Open Access repositories or collections of open educational resources, as well as sites such as Wikipedia – could well find themselves obliged to apply such (faulty) filters.

More broadly, the application of filters without consideration of their impact on exceptions and limitations to copyright, or freedom of speech in general, runs counter to library values.

Be Cautious, not Cavalier

So filters should not be the subject of misplaced faith, but rather of caution. This isn’t the only case of course – blockchain too risks undermining the operation of exceptions and limitations, as well as of anonymity. Perhaps one day technology will improve, but until then, we should beware of miracle cures, and not lose sight of key ethical and human rights principles around freedom of expression.