Why read when you can cloud?
What I learned using generative AI to analyze climate-change communications
Like many of us, I’ve been fascinated by the new crop of generative AI systems emerging recently. The potential to transform how we communicate is significant.
As someone seeped in climate-change communications, I am curious to see how we can use these systems to improve how we talk and write about humanity’s biggest challenge. In this two-part blog, I’m going to show how I’ve been interacting with one system in particular and the insights it’s producing.
What do our computer overlords think about the latest IPCC report?
The great hulking behemoth of climate comms is arguably the Intergovernmental Panel on Climate Change (IPCC) reports. I used Humata to analyze the latest Summary for Policymakers—the most readily intelligible and widely-read of all the IPCC reports. Humata is a large language model tailored to academic/research publications. A recent recommendation in the New York Times turned me onto it, so I was curious to see how it handles the big fish of climate comms.
The first thing to note is that Humata’s user interface is a bit buggy. It was initially unable to read the PDF of the AR6 Synthesis Report Summary for Policymakers (the most recent one, released in March 2023). I opened a help ticket with the company and it took them a few days to get back to me, although they were ultimately able to foist it into the system.
Humata’s platform also tripped up on a few queries I tried to feed it, asking me to refresh and input my query again (and again and again). Although it eventually managed to provide answers, it’s definitely still ironing out the kinks.
A climate-comms co-pilot?
The scientific nature of climate change means that reports like the Summary grab a lot of attention when released. Yet few have the time or interest to read every word, and even those that do often struggle over the rather prolix way the IPCC presents information. Platforms like these could provide a helpful tool for journalists and other climate-watchers in digesting major findings.
Indeed, I generally found Humata useful in deciphering, decoding, and analyzing the Summary. My insights fall into three main categories: navigating the report, generating insights and improving climate communications. I’ll cover each one in turn:
How can people use it to navigate reports like these?
A report of this magnitude tends to cover a lot of ground, and it’s easy to get lost. The lack of a table of contents adds to the feeling that you’re stepping into a thicket without a clear guide. Humata is like an indexer on steroids. Say you were interested in finding specific climate mitigation strategies. There is indeed a section toward the end entitled “Current Mitigation Progress, Gaps and Challenges”, but the large blocks of text punctuated by parenthetical phrases indicating confidence detract from readability. It also doesn’t help that the authors pepper insights about mitigation throughout other sections on the report.
Humata, however, produced a succinct ten-point list of mitigation strategies identified in the report—a clear improvement over typing “mitigation” into Adobe Acrobat’s text search field and parsing through the 107 results. Humata also appends footnotes that include the specific section of the report from which it generated its finding, although this feature is still bit buggy (some of the footnotes were incomplete, for instance).How can people use it to draw insights from reports like these?
Just as the IPCC report’s weight and significance can have a maze-like effect on readers, so too can it turn the document into a kind of Rorschach test. One’s own climate predilections are readily reinforced by whatever cherry-picked reading one engages in. So why not ask an ostensibly emotion-free observer like Humata to do the reading for you? Here are two prominent effects I noticed:
It drew a conclusion about the report findings that I would not have made myself. Ask people what they think the most significant single impact of climate change will be in the future, and they will likely extrapolate from the current slew of disasters making headlines. Pick your nightmare: searing heat, out-of-control wildfires, raging storms. Climate experts (including, presumably, IPCC report authors) would likely hesitate to name a single impact, at the risk of downplaying other potential consequences.
So when I asked Humata, “What is the most significant impact from climate change that this report predicts could happen?”, it singled out the loss of the Greenland and West Antarctic ice sheets over the coming centuries. There’s no question that this impact would be devastating and massively alter the face of the planet. Humata rightly pointed out several reasons why it chose this answer, including sea-level rise and further destabilization of the global climate system.
Yet nowhere in the report do the authors rank this consequence above any other in terms of significance. They mentioned it, but only alongside several others. It makes me wonder just what kind of logic Humata employed to come to this conclusion. To me, this finding bolsters the case for better explainability for AI systems—a major bone of contention in the developer world.It helped me see the report in a new light. In this line of work I think a lot about whether to impart hope or despair, opportunity or risk, boosterism or alarmism. A recent episode of the Energy Transition Show—one of my favorite podcasts—took to task the doomerism that colors much of climate communications. There is no question that existential dread is a reigning theme of climate work, for better or worse.
It was with this context in mind that I asked Humata to delineate the report’s stated benefits of keeping the rise in global temperature below 1.5°C. What it came back underscored this glass-half-empty mindset: “The report states that limiting global warming to 1.5 degrees Celsius compared to pre-industrial levels would bring significant benefits in terms of reduced impacts and related risks, as well as reduced adaptation needs. However, specific details about these benefits are not provided in the summary for policymakers” (italics mine).
Indeed, the report’s (relatively brief) section on benefits (page 25) reads more like a litany of disasters avoided or downgraded, rather than true benefits. I consume a lot of techno-optimistic chatter about the energy transition—including the potential riches out there for the taking—and I’m equally skeptical when I encounter anything that smacks of sugarcoating. But when Humata came back with a proverbial shrug when asked about benefits, it hammered something home for me: that there’s a certain doomerist “vibe” to the IPCC reports that takes a bit of scrutiny to discern amid the ostensibly neutral, scientific, detached tone.
In my next post, I will look at how a generative AI system like Humata could potentially improve the IPCC report. What does it know about creating sharp and effective content and communications that esteemed climate scientists do not?