Dr Karl Taps AI Chatbot to Persuade Climate Sceptics
For over four decades, Dr. Karl Kruszelnicki has been Australia’s quintessential science communicator, instantly recognizable by his distinctive voice and enthusiastic explanations that demystify complex scientific concepts for millions. At 77, despite a tireless work ethic, Kruszelnicki faces an insurmountable challenge: the sheer volume of questions inundating his social media channels, particularly concerning the climate crisis. He reports receiving up to 300 daily inquiries on platforms like X (formerly Twitter), many from users entrenched in climate skepticism, a group he believes has been subjected to decades of misinformation.
Driven by a profound sense of duty and the understanding that he cannot personally address every query, Dr. Karl has embarked on an ambitious new endeavor: the creation of an AI-powered Digital Dr. Karl. This initiative, developed in collaboration with technology journalist Leigh Stark, leverages a large language model (LLM) from French company Mistral, a rising challenger in the AI landscape. The chatbot is designed to emulate Kruszelnicki’s speaking style while providing evidence-backed answers to climate-related questions, drawing from a vast repository of credible sources.
The training data for Digital Dr. Karl is formidable, comprising an estimated 40,000 PDFs meticulously gathered by Kruszelnicki over 40 years. This extensive corpus includes academic papers, scientific consensus statements, and articles from reputable publications such as The New York Times and The Guardian, mirroring the breadth of the real Dr. Karl’s own research. Kruszelnicki views this project as a philanthropic undertaking, having invested $20,000 of his own money, driven by a commitment to public education. He acknowledges the broader ethical questions surrounding AI, including data accuracy and environmental impact, but argues that in the face of the climate emergency, “the perfect is the enemy of the good.”
Early demonstrations of Digital Dr. Karl reveal an interface akin to popular chatbots like ChatGPT. When prompted with a skeptical statement such as “climate change is a hoax,” the AI responds by seeking clarification and then proceeds to present data, including graphs illustrating atmospheric carbon dioxide levels over the past 170 years. However, as an early-stage “alpha” and “beta” version, the chatbot exhibits some limitations, including a somewhat stilted voice, occasional mixed metaphors, and instances of “hallucination”—a term in AI research referring to the generation of plausible but factually incorrect information, such as fabricated numbers for CO2 rise. Stark and Kruszelnicki are actively working to refine the AI’s voice and overall accuracy ahead of its anticipated public release in October.
The project is conceived as a 100-day experiment. During this period, Kruszelnicki plans to release 100 daily TikTok videos, each directing viewers to Digital Dr. Karl. Following the 100 days, the team will meticulously analyze the chatbot’s performance and public engagement, with a primary goal of assessing whether it has fostered greater open-mindedness and belief in climate change among users. This “Mark 1” approach is a deliberate trial to gauge efficacy before considering further development.
The underlying premise of Digital Dr. Karl’s persuasive potential is supported by emerging academic research. A September 2024 study published in the journal Science demonstrated that conversations with a chatbot could reduce participants’ belief in various conspiracy theories by approximately 20%, an effect that persisted for two months. Thomas Costello, lead author of the study and an assistant professor of psychology at American University, attributes this influence to AI’s capacity for rapid information access and strategic deployment within a conversational framework. He emphasizes that reasoned dialogue is crucial for identifying the core of disagreements. While another study co-authored by Costello suggests similar effects for climate skepticism when AI agents address specific user concerns, it notes that these effective AI models were not based on real individuals, and user willingness to engage was key.
Addressing the environmental footprint of AI, a growing concern, Kruszelnicki and Stark plan to power the Digital Dr. Karl website entirely with solar panels. Stark notes that the chatbot runs on minimal computer memory, theoretically enabling it to operate on renewable energy sources. While scaling the service to accommodate an anticipated 2,500 simultaneous users could present energy challenges, the team is committed to continuous monitoring. Kruszelnicki affirms their oversight, stating, “If it goes bad, we’ll pull the plug.” This ambitious project represents a unique blend of science communication, artificial intelligence, and a deeply personal commitment to combating climate misinformation, venturing into uncharted territory to see if a digital doppelgänger can indeed shift public perception.