Co-founder of Atlassian Scott Farquhar has said the benefits of large language models outweigh the issues raised by AI training its data on other people’s work for free.Photograph: Mick Tsikas/AAPView image in fullscreenCo-founder of Atlassian Scott Farquhar has said the benefits of large language models outweigh the issues raised by AI training its data on other people’s work for free.Photograph: Mick Tsikas/AAPAnalysisScott Farquhar thinks Australia should let AI train for free on creative content. He overlooks one key pointJosh TaylorTechnology reporterCEO of Tech Council of Australia and Atlassian founder’s argument hinges on whether AI goes on to create something ‘new and novel’Follow our Australia news live blog for latest updatesGet ourbreaking news email,free appordaily news podcastAustralia should adopt US-style copyright law to allow artificial intelligence tosuck up all creative contentor risk harming investment in the industry in Australia, according to the Atlassian founder Scott Farquhar.Farquhar, the Tech Council of Australia chief executive,told ABC’s 7.30 programon Tuesday: “All AI usage of mining or searching or going across data is probably illegal under Australian law and I think that hurts a lot of investment of these companies in Australia.”This was because, he said, Australia doesn’t have fair use exemptions coded into copyright law like the US does.Farquhar’s claim overlooks that this is not a settled issue in the US, and could have devastating effects on creative industries.Companies developing AI, including Atlassian, Google and Meta, want a text and data mining exemptionput into copyright lawto make AI able to train on all human works in perpetuity without paying for it.Sign up: AU Breaking News emailFarquhar’s argument is that it is not theft of people’s work unless the AI is used to “copy an artist directly”, such as creating a song in their style.“I do think people would say that, hey, if people are going to sit down with a digital companion, an AI song creator and they collaboratively work with an AI to create something new to the world, that’s probably fair use.”Farquhar said the benefits of large language models outweighed the issues raised by AI training its data on other people’s work for free.Can AI deliver economic nirvana? Only if workers can monitor and shape how it’s used | Peter LewisRead moreHis argument hinges on whether that AI then goes on to create something “new and novel” which is referred to in copyright as being transformative – creating something new.He said he would have no issue with someone taking what he had created and using it as long as it was “transformative”.“If someone had used my intellectual property to compete with me, then I think that is an issue, directly with me. If they’d used all the intellectual property of all the software on the world to help people write software better in the future, I think that is a fair use.”US law is not settled on AI training being fair use. The US Copyright Office notedin its May pre-print reporton generative AI training that there are dozens of lawsuits challenging AI companies using fair use as an excuse for training large-language models on copyrighted works without paying.In the US, there are also factors to consider whether something is fair use include:Whether the use is commercial or not.The nature of the copyrighted work.The amount used of the copyrighted work.The effect of the use on the market for or value of the copyrighted work.skip past newsletter promotionafter newsletter promotionIn US case law, the transformative nature of what is made of what was taken is the first important factor in fair use, the Australian law firm Gilbert + Tobinnoted in a May publicationbut it is not the only factor to consider, with the impact on the market of the copyrighted work being key. The US supreme court has twice described this as “undoubtedly the single most important element of fair use”, according to the US Copyright Office report.“The copying involved in AI training threatens significant potential harm to the market for or value of copyrighted works,” the report said.“Where a model can produce substantially similar outputs that directly substitute for works in the training data, it can lead to lost sales.“Even where a model’s outputs are not substantially similar to any specific copyrighted work, they can dilute the market for works similar to those found in its training data, including by generating material stylistically similar to those works.”The Copyright Office stopped short of recommending legislative intervention in the US, noting that voluntary licensing were already under way with some AI companies, and allowing licensing would allow AI innovation to continue to “advance without undermining intellectual property rights.”Farquhar’s argument that how generative AI uses copyrighted works would stack up if there were proper guarantees that all usage would be transformative and would not affect the markets from which they’re drawing from.In many of the industries, AI could have devastating effects. In news, for example, AI summaries in Google search already mean people click through to stories less frequently for information, and referrals from AI chatbots compared to the amount of times that AI crawls a page are even worse.To argue that fair use for AI based on the US law is something Australia should aspire to overlooks that it’s hardly settled law, and is being hard fought in the courts. Rushing to give the tech companies what they want in the name of innovation for one new industry could come at the expense of many other industries.Explore more on these topicsArtificial intelligence (AI)AtlassianComputingLaw (Australia)Intellectual propertyanalysisShareReuse this content