We conclude that, when supported by appropriate safeguards, LLMs have substantial potential to complement existing science-communication efforts.

The article is open access: doi.org/10.1016/j.co...

...science skepticism around issues such as climate change and vaccination.

Across studies, there is little evidence that large language models spontaneously produce conspiracy theories—but there is evidence that interacting with them can reduce science skepticism and misinformation beliefs.

I enjoyed the opportunity to discuss this work with such an engaged group of scholars working on collective action research.

For those who are interested in the full paper, it’s available as an open access article in the Journal of Environmental Psychology: doi.org/10.1016/j.je...

Our findings show that while extreme climate protests can reduce support for the activist group carrying them out, they can also increase public concern about climate change and spur intentions to act—a tension that many activists and organisers face.

We should have another article coming out in the not too distant future summarising the various research we and others have done using this paradigm on these kinds of topics if you'd like me to share it with you when it is published.

Thanks for sharing these links, Ketan. I imagine that each subsequent version is generally getting better, but will never be perfect (pictured is how it was explained to us). But yeah, definitely share your concern about the corporate aspect of the technology.

For anyone interested, here is a link to bypass the paywall and download a PDF version of the article: rdcu.be/eHcNY

Quite possibly! However, it should be noted that the effects on reducing scepticism among sceptics were susceptible to decay at follow-up. So, it might be that repetition of the intervention, or other factors are needed, to make the decline more durable.

In the paper, we’re explicit about several limitations of the approach. However, we've now conducted several studies that are similar to those in this paper (on different science-related topics), with none of them finding inaccurate information being generated by ChatGPT on these topics.

That's a fair concern, one that we address in the article itself. We used a validated process in which responses were fact-checked by another LLM (Claude), and a substantial subset of responses manually fact-checked by independent climate experts. Results found negligible evidence of misinformation.

This paper is the product of a team effort from members of the Net Zero Observatory. I’m grateful to have worked alongside such talented colleagues: business.uq.edu.au/research/net...

Across two studies, we found that short conversations with ChatGPT led climate sceptics to report lower confidence in their sceptical views and small increases in pro-environmental intentions.

You can read the article here:
doi.org/10.1038/s415...

Thanks to my supervisors, progress review panel, fellow members of the Net Zero Observatory, and other friends and family for their support throughout this journey.

And in what feels like a rite of passage, I now also seem to have my very own UQ web profile: business.uq.edu.au/profile/1925...