There’s arguably no face, voice or collection of exuberant, patterned shirts more recognisable than those belonging to Dr Karl Kruszelnicki.
The bespectacled boffin has been answering curly listener questions about science, with characteristic excitement and passion, for more than 40 years. Despite a seemingly tireless work ethic, Kruszelnicki, now 77 years old, can’t be everywhere all at once.
Those questions now come in waves, across social media platforms at all hours of the day. “Sometimes I get 300 requests a day on Twitter to answer an involved question about climate change,” Kruszelnicki says.
Particularly on X (formerly Twitter), he says he would often engage with users who don’t believe climate change is real or urgent. He hoped there might be a way to change the minds of this group of people, who he says have been bombarded by misinformation in places such as the Murdoch press for the past 30 years.
After speaking with longtime friend and technology journalist Leigh Stark, the pair settled on an idea: an AI-powered Digital Dr Karl. Using a large language model (LLM), they’re creating a chatbot designed to sound like Kruszelnicki that provides users with evidence, backed by trustworthy sources, that the climate crisis is caused by humans and is an urgent problem to solve.
“I cannot answer all the questions by myself and people want questions answered. The only way I can do it is develop this digital AI,” he says.
Kruszelnicki’s achievements as a science communicator are unparalleled: in Australia he’s considered a National Living Treasure, he won the Unesco Kalinga prize, he wrote dozens of books and is the one and only Julius Sumner Miller Fellow at the University of Sydney, a position he has held since 1993.
He believes AI can help convince those who don’t believe in the severity and causes of the climate crisis – even if there are outstanding questions around the ethical use of AI, its training data, accuracy and its own environmental impacts. “I think with climate change, we are at a stage where the perfect is the enemy of the good,” he says. “We’re certainly not going to become unethical or become like the forces of evil.”
Digital Dr Karl runs on an open-source LLM developed by Mistral, a French company considered one of Europe’s challengers to OpenAI and Google.
To create Digital Dr Karl, Stark has taken Mistral’s base model, then trained it on a corpus of Kruszelnicki’s climate science resources acquired through his own research for his own books and writing on climate. It includes academic papers, consensus statements and original articles from publications including the New York Times, the Guardian and RenewEconomy to build out its knowledge, just like the real Kruszelnicki has.
“This is an AI that’s been trained on the 40,000 PDFs I’ve gathered over the last 40 years,” he says.
Stark says questions over copyright are valid, and he would like to ultimately have the chatbot trained exclusively on data, but says “we’re not at that point yet”, emphasising “this is beta, this is really early stuff” and the intent is to build something “based solely on data”.
Putting Digital Dr Karl to the test
Taking the Guardian through a demonstration of Digital Dr Karl, Stark reveals the AI interface is similar to ChatGPT, and users can type in a single query about climate change to kickstart a conversation.
Stark types in “climate change is a hoax” and the Digital Dr Karl replies a few seconds later in a stilted and tonally inconsistent recreation of Kruszelnicki’s voice. It wants to know if we are suggesting climate change “is a fabricated idea”. We are only able to answer yes or no. We respond yes, at which point the AI quotes Barack Obama on the effects of climate change.
As the conversation continues, Digital Dr Karl displays data, such as graphs showing atmospheric carbon dioxide over the last 170 years. But it also seems to mix metaphors and “hallucinate” (the terminology used in AI research for “make up”) some of the numbers for atmospheric carbon dioxide rise.
Stark describes the AI as both an “alpha” and “beta” version, and he is working to improve the AI voice, but expects Digital Dr Karl will release this in October. Kruszelnicki says he has already spent $20,000 of his own money since February to develop the AI: “This is purely philanthropic – I do this because I see this as my duty, in return for 16 years of free university education that I received.”
after newsletter promotion
Kruszelnicki plans to run his digital self for 100 days because “it’s a nice round number”. He will also deliver 100 TikToks, one a day, alongside the project and each one will push people to his Digital Dr Karl, Stark says. After 100 days, the pair will switch off their AI and “work out what the fuck just happened”. At that juncture they will do a survey with the hope that the result is that “more people are open-minded and believe in climate change”, says Stark.
Kruszelnicki says they’re just “trying to do the Mark 1” and will see what they find before deciding whether they go on to a Mark 2.
There are some hints about what could happen. Mounting evidence – academic and anecdotal – suggests LLMs can influence emotion, opinion and belief.
In September 2024, a study in the journal Science showed conversations with a chatbot could reduce participants’ belief in their chosen conspiracy theory, including everything from the Kennedy assassination to the illuminati, by around 20% on average. The effect persisted for two months after the conversations took place.
Thomas Costello, assistant professor of psychology at American University and lead author of the Science study, says the AI is persuasive because it can rapidly access and strategically deploy information in conversation. “The back and forth is useful because [reasoned] dialogue and debate is excellent at surfacing the crux of disagreements and kicking the tyres of each side,” he says.
Costello has also co-authored another study, yet to be peer-reviewed but available online, suggesting a similar effect is seen when AI models, tailored to respond to specific concerns from a user, address climate scepticism and inaction. One of the key elements though, is that these AI agents are not based on any real person, and to shape belief, users must be willing to engage in conversation.
But even if Digital Dr Karl can change minds, it contains the same outstanding issues as other LLMs. Kruszelnicki and Stark hope to alleviate the concerns around AI’s environmental impacts.
“We’ll run the website entirely off solar panels and you don’t need a lot of energy,” says Kruszelnicki.
Stark says Digital Dr Karl is running off a very small amount of computer memory on a $12,000 Mac and it theoretically can run on renewables. “If we can get several of these computers running off of a solar battery or basically solar panel and a large battery, then we can effectively run this on renewables.” However, with more users, Stark says scaling it could be a challenge – he expects up to 2,500 people will be accessing Digital Dr Karl at any time.
“We’re going to be keeping an eye on it, on every response that it makes,” Kruszelnicki says. “And if it goes bad, we’ll pull the plug.”