The House of Lords chamber, where an amendment to the UK’s heavily debated AI bill has passed 272-125.
That’s according to outlets from across the pond, which have described a new amendment to the proposed law as “a heavy defeat” for the government. As we previously noted, said law as written would have given gen AI developers the green light to train their models on protected works without prior permission.
Instead, it’d be up to rightsholders themselves to “opt out” of training. While the government painted the system as part of a “blueprint to turbocharge AI,” the majors, Merlin, and a multitude of artists criticized the measure as a serious blow to the creative community.
As described by The Independent, politician (and film director) Beeban Kidron in more words criticized the proposed law as enabling tech giants to steal protected media and then undercut creatives. A related amendment to the bill is said to have passed in a 272-125 House of Lords vote.
Digging into the amendment itself, the suggested change would empower copyright owners to obtain “information regarding the text and data used in the pre-training, training, fine-tuning and retrieval-augmented generation in the AI model, or any other data input to the AI model.”
Additionally, AI developers would be compelled to “provide an effective mechanism to allow copyright owners to identify all individual works that they own that are used” in training, per the amendment.
And the same owners would reportedly have to approve beforehand – not proactively opt out of – training usages. The text describes several other obligations for AI developers (including in relation to bots) as well.
Seemingly every gen AI business operating in the UK will be on the hook under the amendment as written; the requirements would apply to any model that “has a significant number of United Kingdom users” or that counts the UK as one of its “target markets,” the document shows.
Stateside, the battle over AI training is as heated as ever, referring both to high-stakes legislative proposals and ongoing suits. Even individual platforms like SoundCloud are finding themselves at the center of training debates.
And while it happens to be a key argument from the AI side, evidence strongly suggests that developers based in countries with less robust IP protections are, in fact, training their models on protected works. Put differently, there are more than a few angles to consider when it comes to the unprecedented technology’s ongoing fallout.
Share on:
You must be logged in to post a comment.