There is probably no face, voice or collection of more recognizable exuberant patterns than those belonging to Dr. Karl Kruszelnicki.
The scientific communicator with glasses answered questions from the auditor completed on science, with a characteristic excitement and passion for over 40 years. Despite an apparently tireless work ethics, Kruszelnicki, now 77, cannot be everywhere at the same time.
These questions now come in waves, on social media platforms at all hours of the day. “Sometimes I receive 300 requests per day on Twitter to answer a question involved in climate change,” explains Kruszelnicki.
In particular on X (formerly Twitter), he says that he would often engage with users who do not believe that the climate crisis is real or urgent. He hoped there could be a way to change the mind of this group of people, who, according to him, has been bombed by disinformation in places such as Murdoch Press for 30 years.
After talking with a longtime friend and technological journalist Leigh Stark, the couple settled on an idea: a Dr. Karl Digital powered by AI. Using a large language model (LLM), they create a chatbot designed for It looks like Kruszelnicki who provides users with evidence, supported by reliable sources, that the climate crisis is caused by humans and is an urgent problem to solve.
“I cannot answer all questions for myself and people must answer questions. The only way I can do it is to develop this digital AI, “he says.
Kruszelnicki’s achievements as a scientific communicator are unprecedented: in Australia He is considered a living national treasureHe won the UNESCO Kalinga Prize, he wrote dozens of books and was the only and only Julius Sumner Miller scholarship holder at the University of Sydney, a position he has held since 1993.
He thinks that AI can help convince those who do not believe in the gravity and the causes of the climate crisis – even if there are questions unanswered about the ethical use of AI, its training data, its precision and its own environmental impacts. “I think that with climate change, we are at a stage where the perfect is the enemy of good,” he said. “We are certainly not going to become contrary to ethics or become like the forces of evil.”
Digital Dr Karl operates on an open source LLM developed by Mistral, a French company considered as one of the European challengers in Openai and Google.
To create Dr. Karl Digital, Stark took the basic model of Mistral, then trained it on a corpus of climate science resources from Kruszelnicki acquired thanks to his own research for his own books and writing on the climate. It includes academic articles, consensus declarations and original articles of publications, notably the New York Times, The Guardian and Renewecomy to develop his knowledge, just like the real Kruszelnicki.
“This is an AI that has been formed on the 40,000 PDFs that I have gathered in the past 40 years,” he said.
Stark says that the questions on copyright are valid, and he would finally like the chatbot to be formed exclusively on the data, but says “we are not yet at that time”, highlighting “it is the beta version, it’s really early things” and the intention is to build something “based solely on the data”.
Put Dr. Karl Digital to the test
By taking the Guardian through a demonstration of the Drict Drin Karl, Stark reveals that the AI interface is similar to Chatgpt, and users can grasp only one question about the climate crisis to launch a conversation.
The Stark types in “Climate change is a hoax” and the Dr Karl Drictor responds a few seconds later in a recreation of the voice of Kruszelnicki, tonally inconsistent. He wants to know if we suggest that the climate crisis “is a fabricated idea”. We can only answer yes or no. We answer yes, how the AI quotes Barack Obama on the effects of the climate crisis.
While the conversation continues, Digital Dr Karl displays data, such as graphics showing atmospheric carbon dioxide in the past 170 years. But it also seems to mix metaphors and “hallucinating” (terminology used in AI research to “compose”) some of the numbers of atmospheric carbon dioxide.
Stark describes AI as an “alpha” and “beta” version, and it strives to improve the voice of AI, but expects Digital Dr Karl to publish this in October. Kruszelnicki says he has already spent $ 20,000 on his own money since February to develop AI: “It’s purely philanthropic – I do it because I see this as my duty, in exchange for 16 years of free university education that I received.”
After promoting the newsletter
Kruszelnicki plans to run his digital self for 100 days because “it’s a good round number”. He will also display 100 tiktoks, one per day, alongside the project and everyone pushes people to his Dr. Karl Digital, says Stark. After 100 days, the pair will turn their AI and “Loulera what the fuck has just happened”. At this point, they will investigate the hope that the result is that “more people are open -minded and believe in climate change,” said Stark.
Kruszelnicki says they “simply try to do the Mark 1” and will see what they find before deciding if they go to a brand 2.
There are some clues on what could arrive. Evidence of assembly – academic and anecdotal – suggest that LLM can influence emotion, opinion and belief.
In September 2024, A study by the journal Science has shown that conversations with a chatbot could reduce the belief of participants in the theory of the chosen conspiracy, including everything, from Kennedy assassination to Illuminati, about 20% on average. The effect persisted for two months after the conversations took place.
Thomas Costello, deputy professor of psychology at the American University and principal author of the scientific study, said AI is convincing because it can quickly access and strategically deploy information in the conversation. “The back and forth is useful because [reasoned] The dialogue and the debate are excellent to deal with the node of disagreements and kick the tires on each side, “he said.
Costello also co-written Another studyBut to be evaluated by peers but available online, suggesting that a similar effect is seen when AI models, suitable for specific concerns of a user, approach skepticism and climate inaction. One of the key elements, however, is that these AI agents are not based on a real person and to shape belief, users must be willing to engage a conversation.
But even if Digital Dr Karl can change mind, it contains the same problems pending as the other LLM. Kruszelnicki and Stark hope to relieve concerns about the environmental impacts of the AI.
“We will manage the website entirely on solar panels and you don’t need a lot of energy,” says Kruszelnicki.
Stark says that Digital Dr Karl performs a very small amount of computer memory on a Mac of $ 12,000 and that it can theoretically work on renewable energies. “If we can run several of these computers from a solar battery or a solar panel and a large battery, we can effectively perform this on renewable energies.” However, with more users, Stark says that scaling could be a challenge – it expects up to 2,500 people will access Digital Dr Karl at any time.
“We are going to keep it an eye on every response that does,” says Kruszelnicki. “And if it goes bad, we will eliminate the catch.”
