When it comes to artificial intelligence (AI), these are not merely interesting times. They are disorienting ones. The technology is evolving quickly, the claims around it are louder than the evidence, and much of the public conversation is driven by people with financial or institutional incentives to speed adoption.
One of the most important questions can get buried beneath the noise: not how to govern, regulate, or operationalize AI, but when its adoption is justified in the first place.
That question matters everywhere, but it matters especially in the public interest sector. And philanthropy, which has the influence to shape this sector, should support organizations in making sound decisions regarding the use of AI.
Philanthropy can help the sector by beginning further upstream on the issue, starting from the question: Is AI in the public’s interest?
With no settled best practices, we think philanthropy’s answer should be: It depends.
Philanthropy has an important role to play in promoting nuance — and one way it can do so is by encouraging organizations to answer these four questions prior to adopting AI.
1. Is the Status Quo Your Baseline?
It is easy to compare AI to the world we wish existed. Perfectly staffed teams, unlimited time for discernment, and flawless human judgment. In the real world, organizations are under pressure from every direction. They face vacancies, overextended staff, and funding threats.
In this environment, leaders should center the question offered by AI writer Ethan Mollick: “Would the best available AI in a particular moment, in a particular place, do a better job solving a problem than the best available human that is actually able to help in a particular situation?”
He calls it the Best Available Human Standard and we think it applies to the public sector. And given philanthropy’s role in society, it should add to Mollick’s idea by asking, “Does it expand our ability to serve the public good?”
Let’s say a foundation supports a small health clinic that regularly serves individuals who do not speak English. Someone walks in who only speaks Swahili, and the clinic doesn’t have any Swahili speaking receptionists. That clinic might take the approach that it should always rely on professional, human interpreters and request support for additional staffing instead of AI interpreter technology.
Philanthropy can encourage the clinic to ask a more relevant question: does it have access to high quality professionally trained Swahili interpreters in its community? Does it serve the public good to see the person who walks in at that moment instead of turning them away? Is there a third alternative, such as using an AI interpreter in the moment, while sourcing an interpreter for a future appointment?
In supporting the public human good, we need to accept the reality of limited resources. In some cases, that might warrant the use of AI.
2. Where Does the Public Benefit from Human Judgment?
In a recent essay, Cory Doctorow draws a useful dichotomy of how humans interact with technology. In the “centaur” model, humans are empowered by technology but maintain decision-making power. In a “reverse centaur” model, humans delegate judgment to AI and rubber-stamp its outputs. Doctorow frames the Reverse Centaur model as a dystopian one.
The real world, as always, is complicated. Let’s pick an area where the politics around human judgment and automation have shifted over time: determining eligibility for government benefits. In “Why Nothing Works” Marc J. Dunkelman describes how reformers of the 1960s pushed to remove human decision-making from benefits administration.
The issue was that government practices enabled biased decision-making. A white family who barely missed an application deadline might be accommodated, while a black family would not be given the same slack. Reformers fought to have more cut-and-dry rules to guard against bias. Today, we see a reversal of that policy push, with some advocating for more human discretion in government benefits, rather than relying on cold, mechanistic rules.
We are strong believers in the relational nature of social impact work, and there are many cases where AI can automate other parts of the job to give us more energy to devote to the human side. But there may be other areas where human bias creates problems and distrust, and working with AI systems can be both more efficient and more fair.
Part of what the Reverse Centaur model highlights is that we should not be passive recipients of technological outputs. We should deliberately structure our work to allow AI to magnify human judgment and relational work where it is most needed. And in some cases, we should embrace automation as less error-prone or less biased than human judgment.
3. Are You Designing With (Not Around) Frontline Workers?
Those closest to communities understand where AI can have unintended consequences and harm the people they serve. Philanthropy should encourage listening to frontline workers as both good management and an ethical safeguard.
But listening to workers doesn’t mean that it won’t change how we work. Technological transitions are uncomfortable. The goal is not to preserve every existing role exactly as it is. The goal is to ensure that adaptation strengthens, rather than sidelines, human expertise.
Healthy adoption requires structured experimentation and iterating with employees on questions like:
- Where does AI reduce burnout?
- Where does it create new stress?
- What does it free humans to do better?
Organizations can’t get answers to these questions without working closely with frontline workers. Philanthropy should be open to funding this type of due diligence where an organization is considering adopting AI.
4. How Does Greater Efficiency Support Your Values?
AI will almost certainly increase productivity in some areas. Efficiency is great! But it creates a fork in the road.
One of us was recently at an “AI in Healthcare” symposium talking about productivity enhancements around notetaking. It was clear that half of the room were doctors thinking, “AI will let me give my patients better attention!” And the other half were healthcare administrators thinking, “Amazing, now I can make my doctors see twice as many patients.”
Both have a valuable point. Patients would likely benefit from a more personalized and relational medical system. And many would benefit from the cheaper healthcare enabled by allowing doctors to see more patients in a day.
There isn’t one right answer to that tradeoff, and the future of medicine will likely feature some combination of those two approaches.
But it makes one of the dynamics clear here: efficiency is not neutral. It amplifies whatever values already govern the organization.
The Bottom Line
AI is moving too fast for clean and easy answers that generalize across all public interest work. AI adoption in the public interest sector should not be treated as inevitable, nor should it be reflexively resisted.
The risks are real: Hallucinations, bias, and the erosion of human relationships are not hypothetical concerns. But the status quo is fraught with its own issues: understaffed clinics, burned-out caseworkers, families turned away from services that they desperately need.
Philanthropy should help the sector use AI with intention and humility. The question isn’t adoption or rejection — it’s whether the sector that philanthropy supports is applying enough rigor to determine if AI can truly close the gap between public interest ambitions and actual delivery.
Ben Zeno is a bilingual strategist based in St. Louis with experience managing mental health initiatives, cross-sector partnerships, and responsive grantmaking across Missouri. Find him on LinkedIn. Tracy McFerrin is co-founder and principal at Credo Philanthropy Advisors, LLP. Find her on LinkedIn.
