THIS WEEK IN CULTURE + THE CULTURE BUSINESS
News Briefing: Generative AI and controversial copyright exceptions
The Chair of Parliament’s Culture, Media & Sport Select Committee, Caroline Dinenage MP, last week wrote to ministers Lisa Nandy and Peter Kyle over reports that the new UK government might be considering introducing a copyright exception for the benefit of generative AI companies.
The last government considered doing something similar, but ultimately abandoned the plan in the face of a backlash from the creative industries.
But what are copyright exceptions, what would this exception mean for AI companies and the creative industries, and why does Dinenage call it a “disastrous policy proposal”?
Shortly after we published this briefing, thousands of creators signed a statement insisting that the use of existing creative works to train generative AI without permission must not be permitted – ie strongly opposing any new copyright exception. You can read and sign that statement here.
GENERATIVE AI RECAP
As outlined in this TW Explainer, the term generative AI refers to AI models that are able to generate original content, including text, image, audio, video and music.
These models are trained by being exposed to existing content. So AI companies copy large quantities of existing content onto their servers to form a training dataset.
That training data is often copyright protected content owned by third parties. In order to copy that content onto their servers, the AI companies should get permission from – and therefore negotiate a licensing deal with – all the relevant copyright owners.
But quite often that doesn’t happen. Even though the creative industries are adamant that permission is required.
COPYRIGHT EXCEPTIONS
Copyright law gives creators control over the outputs of their creativity – so they can control the reproduction, distribution, rental, adaptation, performance, communication and making available of their work.
However, copyright law also includes some exceptions, scenarios where the copyright owner doesn’t have control, meaning third parties can make use of their content without getting permission.
The rules differ from country to country, though common exceptions include things like parody, news reporting and critical analysis.
Some copyright systems have an exception covering text and data mining. Many AI companies argue that that exception applies to AI training. Which, they insist, means they can use copyright protected content as part of their training processes without getting permission.
UK copyright law includes a copyright exception for text and data analysis, but only in the context of “non-commercial research”.
However, in 2022 the government proposed introducing a wider text and data mining exception that could be used by commercial AI companies. This proposal prompted a backlash from the entire creative industries, so much so ministers ultimately dropped the plan.
They then convened a roundtable involving representatives from both the creative industries and the tech sector in the hope that a code of practice could be agreed around AI and copyright. But without success.
Now the new government is trying to decide on a policy around copyright and AI. That work is being led by AI minister Feryal Clark, alongside Peter Kyle, the Secretary Of State For Science, Innovation And Technology, and Lisa Nandy and her team at the Department For Culture, Media & Sport.
THE NEW EXCEPTION – WITH OPT-OUT
There has been speculation in recent weeks that a text and data mining exception for AI companies is being considered again, though this time probably more in line with the exception that already exists in European Union law.
That allows copyright protected material to be used in AI training without getting permission, but also allows copyright owners to opt out from the exception, meaning that their content cannot be used.
Many copyright owners have already opted out – or ‘reserved their rights’ to use the legal term – from the EU data mining exception.
Introducing an ‘exception with opt-out’ in the UK is unlikely to win many fans on either side of the debate. Copyright owners will be strongly opposed to any new exception at all, even one that includes an opt-out or other restrictions.
Meanwhile, for AI companies, the opt-out makes the exception fairly useless, particularly if they are looking to use commercially published content in a training dataset. Because pretty much all major corporate copyright owners will opt out immediately, with smaller copyright owners likely following suit, given the huge amount of discussion there is around these issues.
Indeed, those copyright owners that have opted out from the EU exception have usually done so in a way that anticipates and pre-emptively opts out of any further data mining exceptions that might be introduced in the future
CAROLINE DINENAGE’S LETTER
In her letter to Nandy and Kyle, Dinenage urges ministers to not put introducing a new data mining exception back onto the agenda.
Such a “disastrous policy proposal”, she wrote, would allow AI developers to “scrape creative works from the internet to train their systems without permission and without paying the human creators whose work AI seeks to emulate and compete with”.
“I have deep concerns that the new government is seeking to resurrect this flawed notion of a ‘text and data mining exception’”, she went on, adding that doing so “would remove any motivation tech companies would otherwise have to work with the creative industries to devise commercial models that safeguard the incentives and reward for human creativity in the AI era”.
She then concluded, “The only thing government actually needs to do in this area is oblige tech companies to be transparent about the creative work they are using to train their systems so that the two ‘sides’ (creative and tech) can get around the table and discuss commercial models”.
It’s not clear if and when the government will formally propose a new data mining exception, it may not be until early next year, though the creative industries are already primed to campaign against it.
FURTHER RESOURCES
You can read Caroline Dinenage’s letter here.
TW Culture now offers a training course on AI for creative businesses which can be delivered in-house for your team – either online or in-person. A two hour and half-day version is available. Click here for more information.
TW Culture’s sister publication CMU has been covering generative AI in the context of the music business in great detail – you can follow all that coverage here.
CMU also worked with the UK’s Music Managers Forum on a guide to AI and music as part of the Dissecting The Digital Dollar project.
Here are some other resources about AI and the creative industries…