Between the political and technological hurdles to achieving a global energy transition, the climate crisis can often feel deeply overwhelming. But articulating a solution to what is arguably the greatest potential catastrophe humanity has ever faced is no problem for ChatGPT—or at least, so the chatbot makes it seem.

ChatGPT has the potential to disrupt and transform industries from computer science to media, and has quickly infiltrated college campuses, where students have taken advantage of its various uses to complete assignments. As of January, the platform reached roughly 100 million active users in record time. Released free of charge to the public late last year as part of a testing phase, the chatbot comes from OpenAI, a California-based artificial intelligence research and deployment company in which Microsoft is a major investor. The chatbot is designed to provide human-like responses to a wide range of queries. Now, Google is hoping to piggyback on Microsoft’s success as it prepares to publicly release its own AI chatbot, Bard.

In the past few years, the relationship between AI and the climate crisis has become a growing area of focus both within climate science itself, including for climate modeling, and for climate activists at groups like Extinction Rebellion and Greenpeace, which have targeted Big Tech companies like Microsoft, Google, and Amazon for providing Big Oil companies with AI tools. How exactly AI chatbots could shape the climate sphere remains to be seen, but based on the release of ChatGPT, AI experts and climate activists already have a few ideas—and many concerns.

For members of the climate movement, ChatGPT offers a quick and easy tool to churn out language that could be adapted and packaged into advocacy materials, from press releases and op-eds to flyers. Iris Zhan, cofounder of Fridays for Future Digital, imagined using the chatbot to generate content for direct actions, like song lyrics or rally speeches, and to create better social media posts. “AI can really help you put your words in an organization and in a way that can resonate with people, and I think that’s pretty cool—especially for social media.” Zhan also suggested that the chatbot might be a useful sounding board and help inform persuasive arguments to reach diverse constituencies, from climate deniers and lawmakers to coal miners and fossil fuel CEOs. “It can give climate activists an idea of how certain people would respond to certain climate statements or climate news, and I think that might help with climate communications,” said Zhan.

Still, it seems ChatGPT has yet to really take hold in climate organizing. Adam Cooper, an activist with the UC Green New Deal coalition, said he had not seen anyone raise the subject in organizing spaces. Zhan similarly said they had not seen fellow organizers considering ChatGPT for activist use, likely because of the divide between climate activism and climate tech.

Some climate activists are reluctant to embrace the technology, uneasy about potential misuse. Zhan, for instance, voiced concerns about AI tools’ control by Big Tech and certain wealthy individuals like Elon Musk. Climate communications expert and activist Dr. Genevieve Guenther cautioned that the chatbot could be more effectively used by these interests to hamper climate action. She imagined that the public relations arm of the fossil fuel industry was already exploring how to use the chatbot to the industry’s advantage. “From my perspective, ChatGPT is going to open up another frontier of climate denial and misinformation,” she said. “Whatever benefits it may afford climate scientists or people in the climate movement are largely outweighed by the costs and the dangers that it poses for climate communication and cooperation among stakeholders in decarbonization.”

Climate activists’ concerns are well-founded. Despite its potential applications to their work, ChatGPT has clear operational pitfalls that may undermine such benefits. As HuggingFace research scientist Sasha Luccioni highlighted, the lack of transparency around ChatGPT remains an outstanding issue. While the chatbot might appear “open” to nontechnical users based on the free public access OpenAI currently provides, the actual model behind ChatGPT remains closed-source, with the company providing little information about the process behind it. “It’s really hard to talk about it from a scientific perspective,” said Luccioni. “It’s kind of like a black box that is maybe great to interact with, but…you don’t really know its limitations. You don’t really know what data it was trained on [or] much about the model itself,” said Luccioni.

What is known is that ChatGPT’s responses employ information scraped from the Internet. But from which sites? Online misinformation is a particularly strong concern when it comes to climate change, given its prevalence on social media and the continued legacy of industry-funded climate denial. “There’s always a non-negligible risk that that’s going to come back through the model,” said Luccioni, adding that the chatbot might also provide outdated information it sourced online.

A recent blog post from the Environmental Defense Fund featuring climate scientist Dr. Ilissa Ocko concluded that ChatGPT “missed several critical points” when giving responses to questions about the causes and solutions to climate change. When asking ChatGPT about the fastest way to slow global warming, for example, the bot suggested reducing greenhouse gas emissions, particularly carbon dioxide, but overlooked methane emissions, which account for around 30 percent of the planet’s warming.

For a subject matter expert, receiving inaccurate or inadequate responses might not pose much of a problem. But for the average user, the risk of being misled by ChatGPT—if taken as a primary information source—is much higher. “It’s designed to come up with answers that sound very authoritative, but could be wrong,” said Priya Donti, cofounder and executive director of Climate Change AI, a nonprofit focused on the intersection of climate change and machine learning. Donti emphasized the limits inherent in the chatbot’s sourcing information from English text and being produced by a US-based team. “You’re definitely going to have various lenses in terms of what kinds of data the algorithm is seeing and how the algorithm was designed in the first place that are particularly reflective of…Western white views as opposed to the kind of global way we need to really think about approaching [climate change],” said Donti.

Asemelash Teka Hadgu, cofounder and CTO of Lesan and research fellow at the DAIR Institute, said that both the availability of content, as well as the language in which it is available, could shape the depth and accuracy of responses that a user could get from ChatGPT when talking about climate change impacts in Tigray, Ethiopia, where Hadgu is from, versus in Berlin, Germany.

OpenAI has acknowledged some of these limitations, including that ChatGPT may provide inaccurate or biased responses, and says it is working to address them, in part with a new content moderation tool to filter out harmful information and prevent misuse. An OpenAI spokesperson also shared that OpenAI has already updated ChatGPT based on user feedback, and published information on efforts to improve its behavior, but did not respond to further requests for comment.

Anecdotally, it is clear that some guardrails exist. In response to a prompt doubting climate science, for instance, the chatbot informed me of the overwhelming scientific consensus around climate change, its anthropogenic origins, and the widespread impacts it is already having. Meanwhile, far-right author and commentator Alex Epstein tweeted his frustration about the chatbot’s refusal to produce an argument in favor of increased fossil fuel use. Still, the full extent of ChatGPT’s limitations and its potential impacts remain troublingly unknown.

Whether chatbots are a promising tool for the climate movement or a useless or even harmful one remains an open question, but what is clear is that AI is not going anywhere, and neither are its implications for addressing climate change. Those implications include not only what AI is being used for but also the literal impact of its use on the planet. While the precise carbon footprint of ChatGPT is unknown, it has the potential to require significant amounts of energy both in its production process and in its consumption, and so notably contribute to global emissions. Another model from OpenAI, GPT-3 was estimated to produce about 500 metric tons of carbon dioxide while being trained—the equivalent of roughly 610 one-way direct flights from New York City to Paris. “These large language models are…contributing to the negative effects of climate change,” said Hagdu.

Nonetheless, climate activists and AI experts alike expressed hope for the future of other AI tools, if not ChatGPT, and climate work. Donti is excited about projects that allow climate and AI or machine learning experts to “cocreate something that is leveraging both of those aspects in the development of the model or the workflow itself” and incorporate equity consideration from the outset. Zhan suggested that AI could help curtail the risk of activist burnout as it continues to advance. “The climate movement requires a lot of work,” they said. “I’ve always thought about…how AI would hopefully cut down on exploitation and the amount of labor it takes…and improve activists’ lives.”