Copyright laws shouldn't apply to AI training, proposes Google

Google has told regulators in Australia that its AI systems should be able train on any data, unless the copyright owners expressly opt-out.

If generative artificial intelligence (AI) is to be useful, it needs enormous sources of data to train on, and if you're using that data, you should pay for it. Unless you're Google, in which case you think you're an exception — and you have a record of trying to bully your way out of paying.

According to The Guardian, Google has presented a case to Australian regulators that it be allowed to do what it wants and, okay, maybe publishers should be able to say no. But that's on the publishers, not Google.

Google's submission, seen by The Guardian, calls for Australia to adopt "copyright systems that enable appropriate and fair use of copyrighted content to enable the training of AI models in Australia on a broad and diverse range of data while supporting workable opt-outs for entities that prefer their data not to be trained in using AI systems."

Reportedly, this is similar to arguments Google has presented to Australia before, except that it has added what appears to be a reasonable opt-out.

However, requiring publishers to explicitly opt-out of any AI training on their data means the publishers having to know that their work is being mined. Then since the regulators are making plans for all AI providers, it also means that it may not be possible to prove whether a company has ceased mining the data or not.

More, an AI company could in theory delay the process so that by the time it stops mining, it has already used all of the data for its training.

According to The Guardian, Google has not specifically said how it believes such a system could work.

Separately, Google is one of a consortium of Big Tech firms based in the US that has recently pledged to establish best practices for the AI industry.