Copyright is a joke. Scroll through any social network, and you’ll likely see hundreds of obvious copyright infringements within a few minutes. But as we argued last year in our report on Fixing Copyright, the age of mass infringement may very soon become an age of mass enforcement. With so much infringement taking place online, the digital tools to detect and then prosecute infringers are now rapidly improving.
This poses plenty of risks, which is why reforming it now should be a matter of great urgency - with copyright in some cases lasting for over a century, putting off reforms is likely to have major and unforeseeable impacts many generations from now.
So far it’s the way that copyright will interact with newer technologies that is attracting attention from policymakers. The Intellectual Property Office (IPO), for example, recently completed a public consultation on Artificial Intelligence and Intellectual Property.
One of the questions the IPO asked was about the protection of computer-generated works. These are, astonishingly, already protected by the Copyright, Designs and Patents Act of 1988 – our legislators then were remarkably forward-thinking to include the technology at such an early stage. But there is now a concern that the protections given to such works last far too long. AI can work rapidly, and with minimum human effort, so does it really make sense for it to be protected for half a century?
As with the vast majority of creative work, the value of any computer generated work is likely to be entirely realised within the first few years. Given the long duration, and the rapidly changing nature of what AI can do, many in the industry are concerned that a term of 50 years risks causing costly and confusing lawsuits many decades from now, which would eventually end up harming the development of AI rather than incentivising it. So there is a hesitant consensus that the duration of protection should be shortened.
But how short? Over the next few decades, the lines are likely to be blurred between what is an AI-created work coverable by copyright, and what is an AI-assisted invention or design, which would have to be patented or registered. Indeed, those lines are already heavily blurred. So to introduce greater certainty for creators, users, and licensees – always a boon to investment and the flourishing of any industry – we suggested to the IPO that the copyright protection for computer generated works be shortened to 20 years, to bring it in line with the duration of patents.
On a related note, the IPO asked for input on whether AI can invent things on its own, and may thus be entitled to a patent for its creations. We think not. Patents exist to create an artificial monopoly over an invention, to incentivise greater investment, as well as to provide a detailed record of the invention for others to be able to improve upon it. But AI as it currently exists is always a tool for a human, the ultimate inventor. The human is the only one who needs any incentivising, and even supposing we were to create an Artificial General Intelligence able to invent things independently, it would need no incentive or monopoly. It is not human. It would be like giving a patent to a god.
So far as AI pertains to patents, then, it is always a tool. And the law is clear that inventors must be natural persons. But where there may be an issue is over the use of AI to assist in the process of invention. It should be clarified, we argued, that using AI to assist in the process of invention would in no way invalidate or impair the claims of a human to their inventions. After all, we should want to encourage the development of such tools. Increase the rate of invention, and we potentially boost the underlying rate of economic growth.
One proposal we heard was that the developers of an AI, which someone else then uses to invent something, should have some kind of claim over the patent. But we strongly disagree. That would be like giving the manufacturer of a hammer, used in building a house, a claim to the house. Or like giving Microsoft a claim to all copyrighted material written using Word. It would seriously complicate the intellectual property regime and stifle the use of AI-assisted invention. We very much hope the IPO won’t entertain any such proposal.
Last of all, the IPO asked about how copyright affects text and data mining – in training a neural network, for example, AI sometimes makes permanent copies of material that may be copyrighted, analogous to how a human brain remembers an image or song. As things currently stand, there is an exception from UK copyright for when text and data mining is used for non-commercial purposes. In our report Fixing Copyright, we argued that the exception should be extended to all purposes.
The law as it currently stands incentivises many companies to be reticent about sharing any details of how they train their algorithms. They rightly fear that revealing any details may open them to copyright infringement suits – whether justified, accidental, or spurious. The law thus militates against AI transparency.
And it militates in favour of bias. Even if AI companies follow copyright law to the letter by trying to train AI on only databases that are free or easily licensed, this curtailment of material on which to train the AI heightens the possibility of bias. It is a problem that will be faced especially keenly by smaller companies or start-ups, who may be limited to using only cheaper, low-quality, or free data.
The only people who stand to lose out from extending the exception for text and data mining are those who charge for the use of databases. In our original report we argued that they could protect the value of their databases by simply restricting access to them with paywalls, much as they already do. But the IPO has come up with an even better option – one that both extends the exception and allows database rightsholders to assert and maintain their rights without having to use paywalls. This option would be to extend the exception to all uses, but for database rightsholders to be able to opt-out their data in a machine-readable format, the standard of which would be set by the IPO. The few who care would thus see no diminution of their database rights, while AI would simultaneously be able to mine the vast majority of other works that are copyrighted, but whose owners do not seek to capitalise on. It is a clear win-win option, and one we wholeheartedly support.