Australian’s Leading Scientific Research Body Enlists AI to Write Articles

An unnamed Chinese hacker uses a computer at an office in Dongguan, in China's southern Guangdong Province, on Aug. 4, 2020. (Nicolas Asfouri/AFP via Getty Images)

Australia’s national science magazine has taken on the use of artificial intelligence (AI) to write articles, in a move that has stirred controversy in the world of publishing and journalism.

Cosmos, now published by the Commonwealth Scientific and Industrial Research Organisation (CSIRO), received a grant from the Walkley Foundation Meta News Fund for a project that aims to investigate the opportunities and risks of using AI to write science-based stories.

However, ethical questions have been raised around the use of funding intended to support journalism being used to aid AI development, with some ambiguity surrounding the approval of the project.

Last month, Cosmos published six AI-generated stories on various topics with the program built with funding supplied by Meta, which had been allocated for caretaking to the Walkley Foundation to support journalism in Australia.

According to a Walkley Foundation statement online, Meta allocated up to $2.5 million (US$1.6 million) in funding for its Digital Innovation Fund in 2023, which was open to journalists and publications to apply for.

The move to approve funding towards AI has sparked confusion around the decision. Cosmos’s former editors Ian Connellan and Gail MacCallum told the ABC they had not been aware that the publisher at the time, the Royal Institute for Australia (RiAus), was applying for funding for the AI service in late 2023.

“We had no knowledge of the proposal to employ AI as a background writer-creator,” Connellan said. “As editor-in-chief, I would have said this is a bad idea.”

The ABC reported that Will Berryman, the executive director of RiAus at the time, applied for the grant with the intention of it being used as an aid for journalists and that it wasn’t seen at that stage as a tool to generate news.

Concerns around AI replacement of reporters have been a growing issue in Australia.

Last month, Nine Entertainment Co. staff went on strike citing various concerns including a perceived lack of protections against AI misuse.

Walkley Foundation Responds

The Walkley Foundation responded to concerns around the allocation of funds to develop Cosmos’s AI program in an online statement on Aug. 8.

In the statement, the foundation said reports had been “misleading” and inaccurate” to suggest the foundation directly funded Cosmos for the AI program.

CEO Shona Martyn said the foundation had an “arms-length” administrative arrangement in caretaking Meta’s funding for journalists.

“The Walkley Foundation, its staff, directors and judging board play no part in its funding or judging,” she said.

“The Walkley Foundation requires strict adherence to legal and ethical standards. If grant recipients breach their contractual responsibilities, we act accordingly and seek repayment.”

The Walkley Foundation statement listed a panel of six judges who gave the green light for the funding, including Hal Crawford from Crawford Media Consulting, head of Sky News Elise Holman, editor Stephen Hutcheon, ABC managing editor for standards and compliance in the content division Sasha Koloff, former News.com.au editor-in-chief Lisa Muxworthy, and journalist Steve Pennells.

The Walkley Foundation said journalists had expressed concerns about how their work was being used by AI, given Cosmos’s program was reliant upon former work written by employees and freelancers.

It said the foundation required all recipients of grants to honour and pay reporters and copyright holders, but did not explain how this might be quantified if stories were used as a database for AI.

CSIRO Publishing Responds

A CSIRO Publishing spokesperson told The Epoch Times that Cosmos had a long history of being a trusted source of scientific information, and that facts were checked despite the use of AI.

The editorially independent publishing arm took on the publication of Cosmos in June.

As part of this transition, seven staff from Cosmos moved to the CSIRO, and an additional science journalist has since been hired.

“This experimental project is scheduled to run from March 2024 to February 2025,” the spokesperson said.

“It involves assessing the possible usefulness (and risks) of using a third-party pre‑trained large language model (Open AI’s GPT-4) to assist our science communication professionals to produce draft science explainer articles.

“Given the phenomenon of ‘hallucinations’ or ‘made-up’ information, part of this project involves testing an approach which effectively requires the AI to check the correctness of its output.”

The spokesperson said the experimental project included a step where the generated story content was fact-checked using the database of Cosmos’s previous content, using the approach known as retrieval-augmented generation (RAG).

The experiment does not involve training OpenAI’s GPT-4 model (which was pre‑trained by OpenAI).

“All articles are also fact-checked by a trained science communicator and edited by the Cosmos publishing team,” the spokesperson said.

“Published articles generated in this way are clearly and transparently labelled with an appropriate notice.”

The spokesperson said as the project progressed, CSIRO Publishing would continually review learnings and methodologies.

“This ongoing review could involve testing changes in how we program the tool, how we choose to use the tool, and whether any further usage or development of the tool is to continue after the end of the project,” they said.