Unleashing AI, Leashing Creativity

Background

The UK Government under Keir Starmer is looking to “unleash” artificial intelligence (AI) across the UK to woo tech companies into the UK. The Labour ‘government is aggressively pushing for AI companies to expand operations in the UK, whether by setting up AI Growth Zones or prioritising the needs of AI over people during an ongoing energy crisis for renewable energy solutions.

A part of this strategy involves reworking copyright law to allow tech companies creating generative AI easy access to copyrighted work. In an open consultation, the government proposes to alter UK Copyright laws to allow for a general exception for data mining. This proposed exception extends to use of material for commercial purposes. This exception does not extend to accessing the work: the entity seeking to use the copyrighted material for data mining must still have lawful access to the material. This assumes that the enforcement around copyright law is both ethical and effective, neither of which is a given. Recent revelations that Meta downloaded nearly 36 terabytes of pirated data from shadow libraries to train their AI models highlight large corporations’ utter disregard of copyright law when they need to violate it.

Copyright and Capitalism

Discussions about copyright reform would be incomplete without briefly commenting on the purpose of copyright. While copyright and intellectual property rights more generally are often suggested to be necessary/important to incentivise creativity and innovation, a deeper look into their origin and functioning reveals a different story. Much like the enclosures of the commons, intellectual property rights have served to enclose the “intangible commons of the mind” (Boyle, 2017). Rather than incentivise creativity and protect the rights of the creator, copyright serves to commodify creative labour and render it exploitable by capitalists for profit (Arditi, 2020). Publishing houses purchase the (copy)right to reproduce creative work (as owners of the means of reproduction and marketing, protected by the force of law) while the authors effectively become contract workers for the publishers.

While copyright commodifies the art produced, alternatives exist that touch on undoing the enclosures of creative works. Creative commons (CC) exists as an alternative system for creators to share works while retaining certain optional rights: attribution for the work (BY), prevention of derivative work (ND), to require users making derivative work to share such work under identical terms (SA) or use it solely for non-commercial (NC) purposes. The CC system prevents the exploitation of creative labour by releasing creative work to be accessible by all. The means of reproduction are no longer held by the capitalist but released to the public at large. This does not entail a lack of remuneration for the creator; indeed, many creators today choose to share their work with the public and seek direct support from users/audiences through payment services such as liberapay. Having works shared under “non-commercial” or “share-alike” licenses specifically serves to preserve the commons as an ideological space for creative work.

General Exemption for Data Mining: Creative Closures

Should the government choose a general exception to copyright for data mining for any purpose (including commercial purposes), AI companies effectively gain access to any copyrighted material that they have legal access to. Even with the requirement for them to have accessed the work legally (which is not much of a restriction given the obscene power differences involved in legal processes against big corporations), a simple subscription to a publishers database or library would be sufficient for an AI company to use the copyrighted material for data mining.

A general exemption for AI data mining also directly impacts works released under CC licenses. The government’s proposed reform of copyright law would allow data mining for any material that is available on the internet. This would effectively include all works released under CC licenses. Affording commercial purposes as one of the uses for data mining negates the SA and ND clauses on CC licenses. Data mining is often used by AI companies to train generative AI models (such as ChatGPT) that learn from the data mined in order to “generate” work for users. Generative AI learns the underlying patterns in the training data and uses these patterns to output text that matches responses to the prompt.

The work so generated is effectively work derived from the mined data, or derivative work that would not be permissible under a CC-ND license. Additionally, even if it is assumed that the CC-ND work was not used for the derivation, the use of the work for data mining by a commercial AI company would violate a CC-NC licence. This is further complicated by the fact that non-profits can be converted into for-profit companies; OpenAI is currently looking to do this. This means that data mined by the company for its generative AI that may have been licensed to OpenAI on account of it being a non-profit will now be used for commercial purposes, without the need for OpenAI to re-license the work. It is unclear whether AI trained on a data can be “untrained” on data that it no longer has the license to access, nor clear whether data already mined can effectively be unmined.

Say No to AI: Art by SolgeonSolray @DeviantArt

The government has stated that creators will be allowed to reserve their rights against this general exemption afforded to AI data mining. This would be an “opt-out” system for creators, where the default protects the rights of AI companies rather than the creators, increasing the burden on the producer, should they wish to protect their work. Such a system benefits large corporations (who would have the means to protect their own data) while burdening creators without the same access to resources as large AI companies. Additionally, while the government suggests a machine-readable system like robots.txt to prevent the use of restricted material in the opt-out system, it makes no mention of creative commons licences, despite the fact that these already exist as machine-readable codes for creators. It is imperative that the government clarify in any reform to copyright law that SA and ND clauses would suffice as valid “opt-out” mechanisms against any general exemption, should one be enacted.

The effect of a general exemption to copyright for data mining by AI companies will ultimately be the enclosure of creative commons. Contrary to the government’s belief, this will not fuel innovation or spur creativity; it will stifle creativity while exacerbating inequality. Large AI companies will “enclose” creative work, using the general exemption to mine creative work for generative AI that is then licensed to users on subscription fees. What we could otherwise pay to independent artists for creative work, whether for works of art for personal use (such as paintings), or professional use (logos, trademarks, events) will now be paid to large corporations for the “convenience” of having quickly-produced derivative work. Land enclosures in England in the fifteenth century fuelled the rise of the capitalist economy, spreading from the UK across the globe through colonialism, eroding the rights of people globally. A general exemption given to data mining for AI will fuel a similar enclosure on creativity, eroding our rights to intellectual commons and shared knowledge even more than copyright.

Action Points

The enclosure of creative commons is not inevitable: we can object to our works being enclosed by big tech, both by refusing to pay for generative AI and by protesting the erosion of our rights by the government. The consultation on this policy is open till the 25^th of February; responses to the government may be done through citizen space or by directly emailing responses to copyrightconsultation@ipo.gov.uk. Please do make your views heard; the citizen space survey allows for an easy way to respond to the call. The more people respond to the call, the more likely we are to be heard.

If you wish to protest a general exemption given to copyright in the survey, I would recommend responding “No” to question 4 (“Do you agree that option 3…is most likely to meet the objectives set out above?”) and “Option 1: Strengthen copyright requiring licensing in all cases” to question 5 (“Which option do you prefer and why?”) in the survey.

References

James Boyle, The Second Enclosure Movement and the Construction of the Public Domain, in Copyright law 63 (2017).

Big Tech