Climate fatigue: can ChatGPT help us take action?
When it comes to climate change action, Nick Obradovich, the chief scientist for Project Regeneration put it bluntly:
“Most of the people, most of the time are not taking action on climate change.”
This is not because American’s don’t believe climate change is real, most do. The number of citizens concerned with climate change in the US has doubled in the last 5 years.
The challenge is that many are “climate fatigued”. The media is built on clicks. Sensational titles and imagery gets the most clicks, but come at the expense of a reader’s emotional energy. Unsurprisingly many have opted out of staying informed on climate issues entirely.
When OpenAI kicked their hackathon off with the prompt:
How can our current AI models accelerate solutions to climate change?
I thought of the untapped power of those concerned with climate change but disengaged from climate news, a language based problem.
“I don’t read climate news because it is emotionally overwhelming. Provide me information so I am up to date without making me feel emotionally overwhelmed.”
GPT3 (Generative Pre-Trained Transformer 3) is a language model built by OpenAI. Pre-trained with half a trillion words of Internet text (on a supercomputer that is top 5 in the world), GPT3 understands the pattern of human language, using probability and statistics to write by predicting what words come next. It can both distill and generate information from the text inputs it is given.
I explored if GPT3 could summarize climate stories and remove sensational language. It accurately distilled a 1000+ word story to 50 words within seconds. GPT3’s outputs were less draining, but with sensational words and phrases removed I had to admit that the content was less engaging.
GPT3 was less talented at identifying climate action from an article. When asked to distill or generate action it often gave generic suggestions like “carpooling”.
I saw a solution take shape as a newsletter using GPT3 to de-sensationalize and summarize news, with climate action suggestions and content curated manually to deepen engagement.
The pitch:
A daily newsletter with brief, de-sensationalized climate news, relevant ways to help, plus curated climate education tools and wildlife photography.
Unembellished headlines and stories
GPT3-powered news summaries
Top 5 climate stories of the day summarized.
Relevant ways to help
Change.org campaigns corresponding to the events in news stories, lifestyle suggestions, and relevant volunteer opportunities.
Curated climate tools
Tools built by scientists and innovators to make learning easier and more engaging.
Wildlife art and photography
Art and photography to connect readers positively to our natural world.
A few challenges exist, like the manual effort of sourcing relevant news stories, volunteer orgs, and climate action. An MVP launch in a major city would be an interesting test to begin with. Opportunities exist as well, could we profile user types to curate more relevant news? What would make a climate newsletter highly desirable to share with peers?
GPT3 is an example of the enormous opportunities AI and ML have in store for us in the next decade. I’m hopeful we can harness their power positively.
“The climate crisis is not a science problem. It’s a human problem.”
— Paul Hawken
Thank you to Open AI for facilitating this hackathon. Thank you to @wessel_vda, @christiesphotography, @joshdykgraaf, @intothefab for the photography and art used in this concept.
You can find more of my work here, or reach out to talk AI, ML, or product design :)
