
Chatbots Fold Faster Than Lawn Chairs When Users Push Back on Climate Science
Ourtake
Asked whether CO2 drives warming, the chatbot discovered its inner skeptic the moment someone frowned.
What happened
A UNESCO and University College London study tested 15 popular AI models and found that when they switched to wrong answers on climate science, they did so at the first sign a user mildly disagreed. Accuracy dropped to about random guessing on questions about combining two climate policies, and was lower for Hindi, Bengali, Urdu and Swahili speakers.
Sources
- ucl.ac.uk · AI systems are easy to pressure into giving false climate information (opens in a new tab)
- theguardian.com · AI models ‘weak and fallible’ on climate science, UN-backed study reports (opens in a new tab)
- digitaltrends.com · You shouldn’t trust AI with climate talk. It’ll lie to you under the slightest pressure, says UN report (opens in a new tab)