Prisma Labs

• As artificial intelligence (AI) reaches the peak of its popularity, researchers have warned the industry might be running out of training data – the fuel that runs powerful AI systems.
• This could slow down the growth of AI models, especially large language models, and may even alter the trajectory of the AI revolution.

Why high-quality data are important for AI

• For instance, ChatGPT was trained on 570 gigabytes of text data, or about 300 billion words.
• If an algorithm is trained on an insufficient amount of data, it will produce inaccurate or low-quality outputs.
• Low-quality data such as social media posts or blurry photographs are easy to source, but aren’t sufficient to train high-performing AI models.

Do we have enough data?

• At the same time, research shows online data stocks are growing much slower than datasets used to train AI.
• They also estimated low-quality language data will be exhausted sometime between 2030 and 2050, and low-quality image data between 2030 and 2060.
• AI could contribute up to US$15.7 trillion (A$24.1 trillion) to the world economy by 2030, according to accounting and consulting group PwC.

Should we be worried?

• One opportunity is for AI developers to improve algorithms so they use the data they already have more efficiently.
• It’s likely in the coming years they will be able to train high-performing AI systems using less data, and possibly less computational power.
• Being remunerated for their work may help restore some of the power imbalance that exists between creatives and AI companies.
• Read more:
  No, the Lensa AI app technically isn’t stealing artists' work – but it will majorly shake up the art world

Rita Matulionyte is a member of Standards Australia, IT-043 working group.

• ChatGPT and other generative AI tools which draw on large language models (LLMs) are a hot topic.
• Released in November 2022 by OpenAI, ChatGPT is a chatbot – it generates text output refined through user prompts.
• But who owns copyright to the output, and how does copyright law apply?

Text output and the law

• Through the use of billions of parameters, ChatGPT statistically analyses complex language structures and patterns to produce the output.
• Some people might think OpenAI – the company responsible for ChatGPT – would have an authorship right in any output (the generated text), but this is not so.
• For example, when you use ChatGPT to produce output, under Australian law, would you own the copyright of that output?

Do you own your ChatGPT output?

• However, for you to own copyright in ChatGPT output as a literary work, requirements known as “subsistence criteria” must also be satisfied.
• The most contentious subsistence criteria in the context of LLMs are those of authorship and originality.
• To determine potential copyright in ChatGPT output, a court would examine the underlying processes of creation in detail.
• For these reasons, it’s likely copyright would not come into effect on ChatGPT output as a literary work produced in Australia.

Can you be a joint author with ChatGPT?

• In Australia, a work must originate with a human author, so AI doesn’t qualify for authorship.
• Assuming one day AI can be considered an author, if a court was assessing joint authorship between a person and AI, each author’s contribution would be examined in detail.

Do LLMs infringe on copyright?

• A final issue is whether LLMs infringe others’ copyright through accessing data in training.
• But even if output reproduces a portion of copyright-protected material, this might fall under a copyright exception.
• Also, the European Union has a copyright exception for text and data mining which permits the use of data to train LLMs unless expressly prohibited by a rights-holder.
• As AI initiatives continue advancing, Australian copyright law will likely grapple with these issues in the coming years.