Paul Sanders

How should UK copyright law apply to AI model training? The text and data mining (TDM) exception with opt-out for rights holders proposed by the government offers a balanced, workable approach bit.ly/3XEQvdh The application of UKcopyright law to AI training remains contested, causing uncertainty for artists, rights holders, AI developers and society at large. Bold, pragmatic solutions are thus needed to provide legal clarity, support the creatives and unlock investments Unfortunately, the debate around AI & copyright is too often framed as a zero-sum game. This misrepresents both the challenge and opportunity before us. The real question is not whether AI will transform the creative industries (it already is) but how to make this transition beneficial for all stakeholders A new report by the @InstituteGC argues that the UK can and should be home to both flourishing creative industries and cutting-edge AI development. The report also provides a holistic and detailed policy framework to achieve that end The UK government’s proposal for a TDM exception with opt-out is a good starting point, but will require careful implementation to be effective in practice. Moreover, it is only part of the puzzle. To make the UK a leader in arts and AI, the government should also: 🌍 Strengthen international AI preference standards for rightsholders, to ensure that the opt-out mechanism is robust 🪟 Improve transparency across AI supply chains, incl. reasonable disclosures from AI developers about training data 👫 Establish standards for AI & creativity, incl. guidelines on how to distinguish between human and AI-generated works 💸 Increased support to creatives in the AI age, incl. more dedicated funding and the founding of a new Centre for AI & Creative Industries   Why does our analysis support this policy option over alternatives like 'do nothing' 'strengthening copyright' or 'a broad TDM exception'?   1) It's important to separate AI outputs from model training. While AI outputs should not be allowed to reproduce unlicensed copyrighted works, prohibiting AI model training on publicly available data would be misguided 2) Rights holders should have control of their data, hence the opt-out. But there are better ways to support creators than strict copyright laws, incl. support to hone skills, access resources and develop new business models 3) A whole-of-government approach is needed to deliver the AI Opportunities Action Plan. Restrictive copyright laws could push AI development out of the UK, undermining the Gov’s growth agenda 4) Stricter UK copyright laws would not prevent online data from being used to train AI elsewhere. Closing that loophole through extraterritorial enforcement could lead to AI tools not being accessible in the UK 5) Far from heralding the end of human creativity, AI presents new ways of being original. We should enable and celebrate all forms of expression Credit to the author team: @ZandermannKL, @josephcbradley, @AmandaBrockUK & @MickGrierson – plus @FERNANDOGARIBAY for foreword!   Huge thanks also to all experts who provided comments on drafts (input ≠ endorsement) incl. @whoisgallifrey, @LeeTiedrich, @philiptorr, @lilianedwards, @Prorata_ai, @natejhake, @UKAIofficial, @KR21org, @ChrisMammenIP – and many others