Known as ChatNetZero, it is designed to sift through wordy climate plans – and deliver blunt verdicts.
The launch is planned for next month’s New York Climate Week, while world leaders are gathering at UN headquarters.
It is being developed by AI company Arboretica, research lab Data-Driven EnviroLab and researchers behind the Net Zero Tracker project.
Angel Hsu, the research lab’s founder, on Thursday said the app would be the “next evolution” of applying AI to climate science.
A contributor to the Intergovernmental Panel on Climate Change that issues dire warnings about the planet, she said the aim was to improve on existing tools such as ChatGPT.
Scientists have tried asking those existing apps to analyse net-zero pledges but “we were incredibly disappointed with the results”, she told an event hosted by Climate Change AI.
A live demo of ChatNetZero will be presented to a group called Climate Action Data 2.0 during New York Climate Week on September 22. A limited number of beta testers will also get access to a pilot launch the week before by applying on the app's website.
Users will be able to ask it about countries and companies' climate policies.
“We are developing a fine-tuned, specific large language model to help demystify net-zero commitments,” she said.
In one example, ChatGPT gave a fence-sitting answer when asked whether reliance on fossil fuels undermines a net-zero pledge. ChatNetZero stated forthrightly that such a plan “is not pledging net zero credibly”.
The aim is that users will eventually be able to upload documents such as new corporate climate plans and run them under ChatNetZero’s nose.
“A large volume of text is now available but it’s cumbersome to manually read and code,” Dr Hsu said.
“We need to really understand how large language models can help in this space.”
Dr Hsu also backed AI to take a good guess at measuring emissions from countries or companies that are unwilling to share data.
Findings in June by Net Zero Tracker found the number of major companies setting green targets had doubled since 2020 – but analysts said many lacked transparency on emissions, making them “largely meaningless”.
Another recent report said only a third of the UK's corporate carbon emissions were covered by net-zero plans.
Scientists have used AI to crunch existing figures and try to fill the gaps, although Ms Hsu said the training data came mainly from Europe and that it was not always certain why emissions were falling.
“We have to be looking more towards hybrid approaches, which include integration of machine-learning approaches to fill in these greenhouse gas emissions gaps,” she said.
“You can’t manage what you don’t measure. We can’t make solid decisions if we don’t know whether emissions are actually coming down as a result of our policies."