Article 6FV4R New French AI Copyright Law Would Effectively Tax AI Companies, Enrich Collection Societies

New French AI Copyright Law Would Effectively Tax AI Companies, Enrich Collection Societies

by
Mike Masnick
from Techdirt on (#6FV4R)
Story Image

This blog has written a number of times about the reaction of creators togenerative AI. Legal academic and copyright expertAndres Guadamuzhas spotted what may be the first attempt to draw upa new law to regulate generative AI. It comes from French politicians, who have developed something of a habit of bringing in new laws attempting to control digital technology that they rarely understand but definitely dislike.

There are only four articles in the text of the proposal, which are intended to be added as amendments to existing French laws. Despite being short, the proposal contains some impressively bad ideas. The first of these is found in Article 2, which, as Guadamuz summarises, assigns ownership of the [AI-generated] work (now protected by copyright) to the authors or assignees of the works that enabled the creation of the said artificial work." Here's the huge problem with that idea:

How can one determine the author of the works that facilitated the conception of the AI-generated piece? While it might seem straightforward if AI works are viewed as collages or summaries of existing copyrighted works, this is far from the reality. As of now, I'm unaware of any method to extract specific text from ChatGPT or an image from Midjourney and enumerate all the works that contributed to its creation. That's not how these models operate.

Since there is no way to find out exactly who the creators are whose work helped generate a new piece of AI material usingaggregated statistics, Guadamuz suggests that the French lawmakers might want creators to be paid according to their contribution to the training material that went into creating the generative AI system itself. Using his own writings as an example, he calculates what fraction of any given payout he would receive with this approach. For ChatGPT's output, Guadamuz estimates he might receive 0.00001% of any payout that was made. To give an example, even if the licensing fee for a some hugely popular work generated using AI were 1,000,000, Guadamuz would only receive 10 cents. Most real-life payouts to creators would be vanishingly small.

Article 3 of the French proposal builds on this ridiculous approach by requiring the names of all the creators who contributed to some AI-generated output to be included in that work. But as Guadamuz has already noted, there's no way to find out exactly whose works have contributed to an output, leaving the only option to include the names of every single creator whose work is present in the training set - potentially millions of names.

Interestingly, Article 4 seems to recognize the payment problem raised above, and offers a way to deal with it. Guadamuz explains:

As it will be not possible to find the author of an AI work (which remember, has copyright and therefore isn't in the public domain), the law will place a tax on the company that operates the service. So it's sort of in the public domain, but it's taxed, and the tax will be paid by OpenAI, Google, Midjourney, StabilityAI, etc. But also by any open source operator and other AI providers (Huggingface, etc). And the tax will be used to fund the collective societies in France... so unless people are willing to join these societies from abroad, they will get nothing, and these bodies will reap the rewards.

In other words, the net effect of the French proposal seems to be to tax the emerging AI giants (mostly US companies) and pay the money to French collecting societies. Guadumuz goes so far as to say: in my view, this is the real intention of the legislation". Anyone who thinks this is a good solution might want to read Chapter 7 of Walled Culture the book (free digital versionsavailable), which quotes from areportrevealing a long history of corruption, mismanagement, confiscation of funds, and lack of transparency [by collecting societies] that has deprived artists of the revenues they earned". Trying to fit generative AI into the straitjacket of an outdated copyright system designed for books is clearly unwise; using it as a pretext for funneling yet more money away from creators and towards collecting societies is just ridiculous.

Follow me @glynmoody onMastodon. Originally posted to WalledCulture.

External Content
Source RSS or Atom Feed
Feed Location https://www.techdirt.com/techdirt_rss.xml
Feed Title Techdirt
Feed Link https://www.techdirt.com/
Reply 0 comments