“The unlicensed use of creative works for training generative AI is a major, unjust threat to the livelihoods of the people behind those works, and must not be permitted.”
– AI Training Statement / https://www.aitrainingstatement.org/
Over 11,000 artists, writers, actors, musicians and others in the creative industries have signed the above statement, expressing their opposition to works being used to train Generative AI models. They include Thom Yorke and Julianne Moore (as highlighted by The Grauniad), but also – some of my own picks from the huge list – Tracy Chevalier, Margaret Drabble, Joe Goddard, Geoff Barrow… and The Reverend Canon Dr Gordon Giles, Canon Chancellor of Rochester Cathedral, Chairman of The English Hymnal Company.
The statement and petition was created by Ed Newton-Rex, who was a lead at Stability AI, but resigned over it’s own position on ‘fair use’ of copyrighted material. He said:
“There are three key resources that generative AI companies need to build AI models: people, compute, and data. They spend vast sums on the first two – sometimes a million dollars per engineer, and up to a billion dollars per model. But they expect to take the third – training data – for free. […] When AI companies call this ‘training data’, they dehumanise it. What we’re talking about is people’s work – their writing, their art, their music.”
Quotes from this piece in The Guardian
I’ve been turning over what I think about the statement today. I am fully supportive of it (and have signed it myself)… but would also want to have some deeper thinking behind it, and would like to see a more extensive position statement about what they are looking to achieve.
As far as I can see from the site, there is no elucidation of ‘the statement’, nor anything offered about what should be done.
This does matter. Earlier in the summer, I was part of a research team funded by BRAID (Bridging Responsible AI Divides) tasked with looking at the impacts of GenAI on those working in the Creative Industries. We sent out a wide-ranging survey, worked with the Musicians Union, BECTU, the Society of Authors, and UK Music and hosted three in-person workshops with people from across the creative arts – from directors to translators to musicians, voice artists, children’s authors and beyond – to really drill down into the issues that they were facing.
To summarise: ‘must not be permitted’ doesn’t even begin to cover it.
Firstly, there was an acute sense that some kind of reparations ought to be justified, given that past works had already been stolen to create the current models. ‘Don’t do it any more’ isn’t seen as sufficient.
Secondly, what about the impacts of licensed works? We spoke to an actor and voice artist who was struggling financially. The offer of £20,000 to take his voice and be able to use it in a licensed way to train GenAI systems was a huge temptation. But… the legal use of that licensed voice would materially then impact his future earning ability, and that of his fellow artists. Paying a few people in great need to legally license their creative talent isn’t good either – and the statement doesn’t cover this.
Finally, though I worry that this is rather too Old School. Many of the artists who are signatories are in positions of huge privilege and wealth. Their livelihoods are not under threat. What worries me is that the position of the statement could be taken as wanting to cut off any benefits of AI in opening access to new and quite radical forms of creativity. They’ve done very well out of the system as it is; they don’t want the system changing.
There’s two dimensions to this. For Thom Yorke, for example, the biggest impact of AI is not going to be on him if his music is used to train AI models. The far bigger impact is going to be workflow automation systems further along in the pipeline that see people displaced. This is not about his creative talent being ripped off – it’s about someone’s labour being automated out, someone who is very likely a lot less visible, and a lot less well paid.
Secondly, what should the position be on how this impacts new artists, who are not yet established and don’t have the major labels / publishing houses etc backing them and funding their endeavours? What seems missing from this statement is about the creative power to take new advances in tech and make brilliant new things. What is the position on that? Where is the redistribution of wealth generated by the mega-tours of huge artists to support the smaller venues that are suffering?
And finally, the first signatory is Björn Ulvaeus. The man is actually now a hologram, performing twice a day in London! Can we hear a little about the economic disruption and innovation of that? Around £8m a year for a virtual, AI-driven show… £8m that’s not then going to other, less established artists? In twenty years, will we just have – playing in perpetuity – ABBA, The Beatles, The Rolling Stones… and nothing new under the sun?
AI is itself a Voyage. I hope that this statement isn’t just about pulling up the drawbridge.
Grab a copy of God-Like here. Or if you’re in the West Country, come see me speak in Frome on 22nd November – grab a ticket here.
Comments
2 responses to “Not OK, Computer?”
Thanks, Kester.
One can’t help thinking that much of this is too little too late. While there may still be scope for retrospective legal claims on historical scraping of people’s intellectual property (see New York Times or Getty Images ongoing legal actions for more details), in most cases the generative AI tools have already digested ALL of ‘our’ information. This means that Radiohead-like music or generative image output in the style of [insert artist here] is already possible without any capability to blocking or preventing it happening (unless the legal cases win out). A main issue with AI generative output is the specific definition of what is original or where authorship lies. According to a US court, works created by artificial intelligence are not protected by copyright because they are not produced by a human. This would appear to bolster the position of human creatives against other humans that are aiming to profit from derivative work made by AI (and why some publishers are keen to have their authors verify that AI was not used in their writing process, in order to protect the publishers’ business model). This defence of originality relies on the current wording of a US copyright law. So, if AI can produce a Radiohead-like track, the resultant output is not protected in the same way as a human’s. For now.
It is worth noting that AI generators have no conscious intention and are producing original output from human prompting and based on the training data. This has been likened to the way artists may learn or develop their own work with reference to those artists who have come before – an accepted part of art [and science]. When this process goes wrong (is too similar to an existing work), the central legal question is the charge of plagiarism. This is a much harder area to work through as historical examples indicate that already successful and rich individuals mostly sue others, with the presumption that they do so because they have the financial means to support the action, but also to seek financial damages in that defence. Proving that something is a direct copy or merely ‘inspired by’ is not always so clear cut and if I was an AI platform, I would claim no responsibility for a user’s output, in much the same way as social media companies claim they are not publishers and not liable for users’ posts – so-called Section 230 defence. This is the first obstacle that the signatories to the AI statement face if they really are serious about taking this further. To get even close to doing something about this ‘AI problem’ will involve first legal actions against deep-pocketed corporations and then overwhelming numbers of individual lawsuits against users, which will be extremely costly and with little consequent monetary reward. Of course, maybe the co-signatories of artists, musicians and creatives will sink their collective wealth in an endeavour to protect the noble pursuit of human art in the name of and in defence of human expression. The legal questions that need to be resolved in this whole area are complex and any consequent laws may be ultimately unenforceable.
As you allude to, signatories are supporting what [currently] appears to be an empty gesture and issuing an ultimately useless primal scream of futile despair, which is sometimes at least something!
Thank you James – excellent thoughts! Totally agree with you here.