Last month, UK technology secretary Liz Kendall backtracked on plans to let AI firms use copyright-protected work without permission. The government now no longer has a “preferred option” on copyright reform, after previously issuing a proposal that tech companies would be able to train on copyrighted work unless creatives opted out.
At a glance, the U-turn represents a win for creatives who won’t be forced to opt-out to prevent their works being used to train AI models. However, the governments policy still presents uncertainty for both creatives and AI vendors.
On the one hand, the UK government hasn’t completely ruled out an exception for training on copyrighted works. At the same time, it’s offered next to no guidance to help AI vendors develop models in compliance with the law. The protections of creatives and the responsibilities of tech companies have been left in complete ambiguity, which could have negative implications for the UK tech and startup scene.
How over-regulation could damage the UK tech scene
Since the UK government U-turned on its “preferred option,” technologists have been deeply concerned that the lack of clarity on copyright law could stifle innovation in the country. For example, Tech UK is calling for a text and data mining (TDM) exception that would enable copyrighted works to be used for commercial TDM purposes. Currently, TDM activity is only permitted for non-commercial research purposes, which limits the scope of data available to train AI.
“This report is based on the misconception that the UK’s copyright regime can exist in splendid isolation from the rest of the world. In fact, these recommendations are a recipe for ensuring that the UK is an AI taker rather than an AI maker. The most likely scenario based on these recommendations is that AI models continue to be largely trained in the U.S. and that the ultimate legal framework of AI copyright is determined by the U.S. courts,” Anthony Walker, deputy chief executive of Tech UK said in a blog post.
The Tony Blair Institute has also called on the government to move toward a more overt pro-AI stance. One report, written by Jakob Mökander, director of science and technology policy at TBI and others called for a TDM exception with an opt-out saying: “we believe it is sound policy for legal, economic and geographical reasons.”
TBI argues that prohibiting models from being trained on publicly available data would be misguided and impractical.
“To argue that commercial AI models cannot learn from open content on the web could be close to arguing that knowledge workers cannot profit from insights they get when reading the same content,” the report said.
From this perspective, a hardline approach to copyright could see AI vendors and developers withdrawing from the UK market. Such a position doesn’t seem fair to rightsholders who’ve had their content scraped and used without permission or compensation.
Creatives, however, have a very different view on the type of legal guardrails they’d like to see.
Where creatives stand on AI copyright law
At a high level, creatives have responded positively to the change, but many remain wary that the government could implement copyright exemptions in the future; if the UK can u-turn once, it can do it again.
“We have cautiously welcomed the UK government’s U-turn on AI use of copyright-protected work, but creators in the UK are still a long way from having the protection they need,” Anna Ganley, CEO of The Society of Authors trade union told 150sec.
“While the government has dropped its preferred option of a text and data mining (TDM) exception with an opt-out, they haven’t said what might replace it. Nor did they rule out any future copyright exception that would make it easier (if not completely free) for AI developers to train their models on copyright-protected content without consent or compensation,” she added.
So far AI appears to have had a negative impact on the UK creative scene. According to a Society of Authors report, which surveyed 10,000 creatives, 86% of authors have suffered reduced earnings and 32% of illustrators have lost commissions as clients turn to AI.
The experiences of such creatives make it doubtful that enabling a free-for-all for training on web content wouldn’t further harm rightsholders. It also doesn’t seem ethical to enable AI vendors to train on work which the author hasn’t consented to. Instead, the government should look toward incentivising vendors to embrace licensing models, which fairly compensate creatives.
Ganley says the UK government should implement consent first licensing rather than scraping, plus a ban on misrepresentation of authorship as well as mandatory transparency on works used for model training. If AI is to be trained on creators works, they should be compensated.
The UK can’t sit on the fence
As the AI race rages on, the UK must decide what side of the fence it’s going to sit on: it simply can’t aim to please both creatives and technology advocates.
The country should also look to compromise by providing the clarity and legal mechanisms necessary to give AI vendors access to a high volume of training data, while giving rights holders the ability to protect their work, and seek out licensing models if they so wish.
At the moment, there’s no clarity for either group — just a spectre looming that there could be a TDM exception added at a later date. This results in a situation where creatives are waiting cautiously to see if the government subverts their copyright, while developers are left uncertain what data they can train their models on.
The UK government needs to offer further clarity on just what the parameters of legality are around training models on content. The government’s changing pace is still leagues away from offering a cohesive policy — if it’s taking a pro-creative stance and giving creatives new rights, it should be clear about these restrictions, so that AI firms can build ethical and legally-compliant systems.
Featured image: Wesley Tingley via Unsplash+
