The term artificial intelligence (AI) is arguably a misnomer. That's because the technology is based on massive data sets of human-generated content that algorithms devised by computer scientists assemble into what users ask for with their prompts. Already, parts of officialdom have ruled that while AI used as a tool is OK, wholly-generated output won't be considered as new and original work.
Where the lines in that shades of grey area end up being drawn will be decided in legal battles the next few years, and it'll be a a crucial war for AI companies. They depend on fresh, human created data for their image, video, audio and text generators. Humans who create the data are often not happy about their work being used to train AI models without their permission and compensation.
It doesn't take too much pondering to understand why people would object to well-funded AI companies taking their creations, and even likenesses and characteristics as humans, to build technology that's said to put them out of work, ultimately. The New York Times has a story (paywalled) about two voice actors discovering they had essentially become cloned by AI. That is, their voices are allegedly used by an AI company without their permission.
The two are now suing the AI company in question, supported by the Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) union, seeking class action status to cover other potential litigants as well.
It will be an interesting court case to follow: the AI company's denying any wrongdoing, but rights holders have long established precedents on their side, and are powerful enough to influence the law itself, like they did in New Zealand with the anti-filesharing amendments to the Copyright Act. If in doubt as to how powerful rights holders are, try adding a YouTube video with a copyrighted music track and see what happens.
Already, the United States Authors Guild umbrella body for writers has scored a win in its copyright court case against Microsoft-backed OpenAI, with the latter admitting it had deleted two large data sets containing the text of some 100,000 books used to create its generative pre-trained model GPT-3, saying it was "Internet corpora". That's a fancy way to say "stuff scraped off websites".
OpenAI says the data sets were not used to train later models such as GPT-3.5 and GPT-4.
Microsoft and OpenAI are in the legal gunsights of The New York Times as well, with the tech companies alleged to have used copyrighted material from the publisher to train AI models. The Authors Guild meanwhile are keeping a close watch on the situation, and has issued AI Best Practice for writers covering several issues the new technology has brought in.
Now, Sony Music Group (SMG) has set a beautifully crafted legal trap for AI companies. SMG and its affiliate organisations are banning the use of their artists' copyrighted music for AI training, without explicitly sought permission - and for a raft of other digital uses.
Staring down the legal barrels of SMG's big lawyers is enough to make any tech startup squirm, as they're facing one of the world's three largest music publishers with more than enough in the kitty to fund any court battle, and as per above, well-established precedents backing them.
And massive amounts of money are at stake here. Last year, well-known image library Getty went after the company behind image generator Stable Diffusion, Stability AI, asking for US$150,000 per infringed image, or an enormous US$1.8 trillion in total in damages. With the looming licence breach cases and the mounting cost of generating AI imagery, Stability AI has now been forced to shed staff and its chief executive has exited.
Getty has, meanwhile, embraced AI with a deal that compensates creators whose work is used to train the generative technology, providing indemnity for image use. Similarly, OpenAI this week reached a deal with Reddit to add the content from the social news aggregator and discussion forum to the ChatGPT chatbot.
That AI is headed towards a legal morass shouldn't come as a surprise to anyone, as it reflects the nature of the Internet where so much content is freely accessible. Well, the content seemed to be freely accessible, but the AI companies should've checked more closely what the terms were.
What could happen over the next few years is that only tech giants such as Microsoft and Google are able to move forward with AI as they have the deep coffers to indemnify customers against rights holder claims. AI startups, meanwhile, face a choice of using a limited number of data sets specifically vetted to be free to use, or to build their products and services on top of tech giants' infrastructure to avoid being sued. Adding the high cost of running resource hungry AI hardware, and we could see the industry being dominated by just a handful of heavy-weights over the next few years.
We welcome your comments below. If you are not already registered, please register to comment
Remember we welcome robust, respectful and insightful debate. We don't welcome abusive or defamatory comments and will de-register those repeatedly making such comments. Our current comment policy is here.