Eric Berlow’s client was new to climate grantmaking. The organization, a long-time supporter of human rights and democracy, wanted to know how its recently launched portfolio fit into the climate philanthropy landscape. But it had no idea where to begin.
“We don’t actually know who to talk to,” Berlow recalled the organization’s staff telling him. “We don’t know who’s already doing what and who could we talk to for advice.”
His client, which asked to remain anonymous, was aware of existing data tracking projects such as Climate Policy Initiative’s analysis of a decade of climate finance. Such work traced big-picture trends, but left many questions unanswered.
“You can’t double click on the bar and be like, who actually got the money? What are they doing? Who should we talk to? Who are the founders?” said Berlow, an ecologist and climate researcher turned data scientist, who now leads Vibrant Data Labs, a social impact data science company.
That request ultimately led to the Climate Finance Tracker, which launched late last year with the goal of not only answering such questions within climate philanthropy, but also creating open-source tools that can be used in any philanthropic sector. It may be the first of a new generation of tools within the field that use artificial intelligence in one form or another to make grantmaking data more accessible.
With the world abuzz with debate over ChatGPT and other such platforms, this new database offers an early glimpse of how the broad array of computing processes that employ artificial intelligence might be used to bring order to the chaotic array of terminology and categories employed by philanthropy. At the same time, at least one leader in the field — while hopeful that this spectrum of technologies could be helpful — has concerns over their use, wary of errors and biases seen in past AI-related projects.
Berlow and his co-founder Jay Hirschton said they share the concerns over the potential for bias in artificial intelligence, and designed their approach accordingly. They stress that their tool should not be confused with the large language models that are currently in the news; rather, the team used different AI models in limited roles and with human oversight, while primarily employing decade-plus-old data processing techniques.
How AI helped create this tool
Berlow and his team at Vibrant Data Labs began by creating a list of more than 150 keywords related to climate mitigation and adaptation. They pulled from sources such as Wikipedia pages, the websites of groups like Project Drawdown and research reports from places like the Climate Policy Initiative.
After licensing Candid’s grants data, the team scanned it for instances of grants from the past five years that used keywords from their list. To prevent false hits — such as a grant description that uses the phrase “political climate” — they manually reviewed 10% of those results. They then used that subset to train a Primer.ai engine to weed out unrelated results. The team used another artificial intelligence model from that company to identify frequently used words in grantees’ self-descriptions as suggestions for their collection of climate-related keywords, which was used to assemble the tool’s visualizations. In instances where AI engines were used, the team scrutinized the results in detail, Berlow said.
The resulting tool compiles more than $200 billion in grants and private investments — the latter data coming from Crunchbase.com — from the past five years, including more than 3,400 grants. In all, the project cost about $1 million, according to Berlow, with funding coming both from grants (from the Cisco and Hopper-Dean foundations) and contracts (from ClimateWorks Foundation and others). The tracker is fiscally sponsored by One Earth and hosted by ImpactAlpha.
The database could grow further still. Berlow said he has raised roughly half of a $3 million budget to expand the tracker, including adding public sector funding and money from outside the U.S.
Berlow, who once gave a viral TED Talk titled “Simplifying Complexity,” also wants to use that funding to improve the open source tools powering the database. His hope is that others with power over where philanthropy — and investment — goes will use those tools to cut and slice the numbers to yield new insights.
There’s certainly room for remixes. For instance, as currently set up, the dots on the tracker’s map are sized based on how much funding they represent, but the dollar totals they represent are not available to users.
“It was meant to just be a visual Rolodex,” Berlow said. “This isn’t the be-all and end-all, but the tools created could help others that have a lot of influence on how money flows.”
“I just don’t trust machines to do that. Not today. Not in this body.”
Environmental philanthropy’s longest-running grants database is the Environmental Grantmakers Association’s Tracking the Field program. Launched in 2007, it contains more than 150,000 grants that have been hand-coded by staff over the years, but is accessible only to EGA members.
Tamara Toles O’Laughlin, the association’s president and CEO, said she generally feels that “more information is good” and would like to see more tools that use data to drive change for those in harm’s way. Yet this new entrant also sparks doubts for her.
“It’s always valuable for us to have data about how we work so that we’re not just working blindly and in silos, but I have some questions around the ways that the tool works,” she said. “I hope that they accurately reflect what’s happening and are not a really beautiful form of disinformation.”
For instance, Toles O’Laughlin is concerned that the size of dots on the map, which represent current funding levels, may unintentionally reinforce the prominence of major green groups. She’s also concerned about how the tool evaluated equity-related funding.
More broadly, while Toles O’Laughlin trusts the good intentions of Berlow and others involved in the project, she doubts whether any type of artificial intelligence can avoid deeply rooted societal racism, sexism and other biases. Various AI systems have produced such outcomes in the past, including algorithms that select Black men when asked to identify images of criminals, and facial recognition technologies that struggle to identify people of color, among other examples.
“I just don’t trust machines to do that. Not today. And not in this body,” said Toles O’Laughlin, who is a Black woman.
Berlow and Hirschton said they understand these concerns, and emphasize that not only do the tools they used differ from those that have generated these sorts of biased results, but that human review was used throughout their process. For instance, Berlow said when the team used a “black box” model — an AI system that cannot explain its own results — to weed out unrelated search terms, they evaluated and revised the results, and took similar precautions with the other systems they used.
“We need these kinds of tools”
Cisco Foundation’s support of the tracker demonstrates the audience and demand for this type of data. One of the first grants made by its new climate fund was to Berlow’s project, fulfilling a decades-long dream of Peter Tavernise, who leads the climate impact and regeneration portfolio of the foundation’s climate initiative.
Tavernise had long imagined an interface that tracked grantmaking, but conversations years ago with possible providers had gone nowhere. After Cisco committed in 2021 to spend $100 million in grants and investments over a decade to address the climate crisis, Tavernise was tapped to lead the grantmaking. An encounter with Berlow led to a $65,000 grant out of what Tavernise told me was “enlightened self interest.”
“Why something like this hasn’t been invented before now? I don’t understand that,” he said. “We don’t have 20 years to come up with impact metrics… so we need these kinds of tools.”
It is a model that may spread much further, particularly if concerns can be addressed. Berlow has fielded inquiries from across philanthropy, ranging from interest in extending the climate data tool beyond the U.S. to the possibility of doing a similar mapping for child mental health.
The tool has emerged as alarm bells are ringing louder than ever for climate action. Earlier this month, the U.N. secretary general warned that “humanity is on thin ice” after the IPCC’s latest climate report offered yet another alarming assessment of the consequences of failing to lower emissions.
With a steady flow of donors entering climate philanthropy, demand for data on what’s currently being funded is only likely to grow. This new database is an unusually detailed, free public accounting of such grantmaking, and may prove to be one of the first of a new wave of similar projects.
If those future resources are accurate and unbiased, it could be a tremendous boon for climate philanthropy and for other funding spheres that adopt them. Yet if they codify societal biases, even unwittingly, they will perpetuate them. Philanthropy has been handed some powerful new tools, but it must ensure they are safely put to use.
Editor’s Note: This article was last updated on April 5, 2023 to clarify and correct how AI was used in the creation of the Climate Finance Tracker. We also added the team’s response to concerns about bias in artificial intelligence.
Credit:Source link
