For much of this year, we’ve been focusing with our community on the question: “How do we stay ambitious in the context of crisis and constraint?” With needs rising and budgets falling, AI is opening up new ways to do more with less.
Yet AI is not neutral. As with all technologies, it offers opportunities to deliver genuine social good, but also to cause social harm, in both cases on a scale we haven’t seen before. As social impact professionals, how we navigate this matters.
Over the past three years, and most recently at a summit with the Centre for Human-Inspired AI at the University of Cambridge, we have been asking: what would a “human-first, AI-enabled” approach to social impact look like?
Pulling together insights from across our community, and based on our own experience of building AI tools, we’ve developed a framework for thinking through what human-first means in practice and how it should influence decision-making.
We refer to the framework as SHAPE, and it is intended to help us ask better questions so that people remain at the centre of how we design, deploy and understand the impacts of AI.
This is very much a work in progress, and we are keen to learn from others. How are you framing your approach to AI? How can we ensure that we are best placed to tap into the opportunities of AI and manage its potential risks?
S: System-Aware
A human-first approach starts by understanding the system into which AI is being introduced. AI does not enter a neutral world. It enters contexts shaped by unequal access to connectivity, digital literacy, compute capacity, data quality, governance, regulation, language, and institutional capacity. That matters because AI tends to scale the system it enters. It can amplify what is strong. But it can also amplify what is already unequal or fragile.
Key decision questions: Is AI right for this context, given its constraints and capacity? What else needs to happen at the system level, either first or in parallel, for AI to be useful and fair?
H: Human-Augmenting
For us, this is the heart of the opportunity. The goal should not be to replace human capability, but to strengthen it. In social impact work, people bring qualities that AI cannot replicate: empathy, trust, lived experience, contextual judgement, relationships and the ability to interpret complexity. A community worker or health worker may already have the trust and experience needed to drive change. AI can help them reach further, respond faster and draw on more knowledge. But the trusted human relationship remains the anchor. It is trust that turns insight into action.
Key decision questions: How will AI strengthen people’s judgement, capability and relationships? Are there risks of it doing the opposite by weakening human judgement, bypassing relationships or reducing people’s agency, and if so, how do we avoid this?
A: Accountability-Driven
There is a lot of excitement around AI, and understandably so. New AI models and tools are constantly emerging, with the potential to vastly improve our efficiency and reach, even compared to previous models. But in social impact, novelty is not the test. The test is whether AI improves outcomes for people. That means looking beyond efficiency. It also means asking about harms, financial costs, environmental footprint, long-term effects, safeguards and redress. We need to be accountable for the impacts we want to see, and those we want to avoid.
Key decision questions: How will we know AI improves lives and avoids causing harm? How do we build in space for human review, challenge and correction?
P: Partnership-Led
No single organisation has all the answers here. AI for social impact spans technology, business, development practice, public policy, ethics, community engagement, and data governance. Each actor sees only part of the picture. Technology providers may understand the tools. Businesses may bring scale and resources. Governments may shape regulation and public systems. Civil society may bring trust, accountability and proximity to communities. Progress will depend on us connecting and collaborating as humans.
Key decision questions: Who must share ownership, roles, risks and responsibilities? Who is not at the decision-making table that should be?
E: Equity-Centred
This may be the most important test of all. Too often, communities are treated as sources of data rather than holders of knowledge, agency and rights. In poverty-related work, that is a serious risk. An equity-centred approach means ensuring that affected communities shape the problem definition, data use, design, governance, benefits and access. Inclusive design builds trust, relevance and lasting impact.
Key decision questions: Are communities shaping the problem, design and benefits?
Shaping AI With Care
AI can help us stay ambitious at a time when ambition is urgently needed. It can help us engage more effectively, learn faster and deliver at greater scale. It can help us unlock knowledge that would otherwise remain unused. It can enable more personalised support and service innovation at scale.
But the responsibility is just as real as the opportunity. We need to keep asking not only what AI can do, but what it should do, who it serves, who shapes it, and how we know it is improving lives.
For us, that means being system-aware, human-augmenting, accountability-driven, partnership-led and equity-centred. The future of social impact should not be AI-first. It should be human-first, AI-enabled and shaped by the people whose lives it aims to improve.
