AI for Doing Good: Lessons from the Frontlines of Global Development

AI holds the promise of unlocking new value in almost every domain of human endeavours, and that includes the global development space. I travelled from my sleepy London suburb to the bustle of central Bangalore to attend a gathering on the intersection of developing-world non-profits and AI implementation jointly hosted by The Agency Fund and Tech4Dev. It was a blast. Along with fellow advisors and funders at The Gates Foundation, Veddis Foundation, Bridgespan, Digital Future Labs, Dasra and more, I heard about the work and thinking that the attending organisations have been doing. Here’s what I learned:

1. AI is being met with a combined sense of caution and optimism in the non-profit space.

In my view this is an essential first step: we can’t rush headlong into AI implementations that just don’t work. The organisations invited to the event are part of The Agency Fund’s year long “AI for Global Development” incubation programme, during which they will be collaborating on best practices, developing their programmes, and working to scale up their innovations. Hand-picked by The Agency Fund, who are in my view at the forefront of global development tech innovation, these organisations are amongst those rising to the challenges and opportunities that AI entails. Fundamentally, there are more ways AIs can be used poorly than well, and so a great deal of attention to detail was on display. 

2. Large language models (LLMs) are not one-size-fits-all

The organisations are keenly aware that the training data of many (if not all) LLMs such as ChatGPT and Claude are highly westernised in their cultural norms, and as such there are no out-of-the-box solutions. India is vast and diverse: in his keynote, Manu Chopra explains that if you pick any two Indian people at random, there will be a 92% chance they’ll speak a different language. As such, his organisation Karya are developing AI models by and for communities, enabling AI’s capabilities to be more equitably accessible. Appropriately contextualising each implementation is fundamental to its ability to appropriately engage with, and then deliver value to, the recipients of each project. As such, that leads us to the fact that…

3. Local community buy-in is crucial

If you want to help people, first figure out how to best help them. Unsurprisingly, this often involves making them a fundamental part of the process, leaving behind a “global north helps global south” mindset, and adopting one that asks “how can technology breakthroughs assist people worldwide to empower themselves?”. Digital Green is building an AI assistant for small-scale farmers to address the world’s growing food needs. Namita Singh explained their realisation that there are really no better teachers for local farmers than other local farmers, and so they began a programme sourcing and promoting farmer-to-farmer videos to enable peer learning, sharing and community building. Smart.

4. Implementations must be suited to their context

AI is only as useful as your ability to interface with it, and so some pragmatic thinking is required. Jacaranda Health uses a chatbot to deliver health and pregnancy advice to women in Kenya. Emmanuel Oluoch Olang explained that smartphone penetration in rural Kenya is very low – so they’ve brought AI’s cutting edge capabilities to an old-school tech: simple SMS messages. Women can privately text in either English or Swahili through a Q&A system that educates and informs, leading to improved health outcomes for themselves and their children.

5. Charities in the global development space must focus on outcomes, not capabilities 

Han Sheng Chia from the Centre for Global Development pointed out that, when electricity was introduced to rural Kenya, at first basically nothing changed: what mattered was of course what local Kenyans could in fact do with it as a result, and given they didn’t have electric devices, naturally this extraordinary new technology could do nothing for them. We should reflect on the introduction of AI in a similar light, asking, “how do we expect people’s lives to be improved?”. Bloom, like many of the other funders in attendance, care deeply about impact measurement. Don’t tell us how many people you “reached” or “impacted”, tell us what they can do now that they couldn’t before. Seek to capture real, measurable changes in people’s lives, and your fundraising journey will be that much smoother. 

6. AI is not a value proposition, but an enabler of value

Silicon Valley take notes: as the global development space has been brainstorming how best to implement AI in developing country contexts, it has never been clearer that AI straight out of the box provides very mixed value. At its best, AI serves to help scale up, personalise, and make more efficient processes that were previously small-scale, generic and time-intensive. There are two ways to go here: first, find things that we know already work but are too labour intensive to scale, and ask if AI can become a force multiplier. Second, look at problems we’ve failed to solve and ask, “did they previously fail because we didn’t have AI at hand?”

7. Keep cost-effectiveness as a KPI 

Nonprofits seeking funding are always looking for ways to stand out, and there are many ways to do this. Whilst flashy promo materials, a charming donor lead, and a compelling story will help, fundamentally the bones of the organisation must be strong. For Bloom as with many other funders, this means establishing that there’s a powerful relationship between how much money goes in, and how much good comes out. Being able to calculate the unit of impact created per dollar spent ensures you are good value for money and so a solid bet for philanthropic funding. (If you’re not sure how to quantify units of impact, we at Bloom utilise the Wellbeing Life Year – or WELLBY – outlined here by the Happier Lives Institute.)

8. All is not lost

At the time of writing, the disastrous USAID cuts have sent shockwaves through the global development community, and funding sources seem evermore scarce. Fortunately, many projects continue to make breakthroughs in delivering low-cost, high-impact value to people across the developing world. Communities such as those built by The Agency Fund and Tech4Dev are a genuine bright spot, and I’ll be watching their work closely – I advise you to do the same. 

Finally, if you’re interested in this space, I’d ask you to reflect on what you think ultimately matters for impact: at Bloom we want to put self-reported wellbeing at the centre of the impact world. If every non-profit was measuring their impact in terms of how much better the beneficiaries think their lives have improved, we’d have a genuinely excellent way of knowing and comparing how much good we’re actually doing, and we’d know it directly from the mouths of those who matter most: the people we are trying to help. Capture the improvements in life satisfaction that your work creates, and you’ll have a north star always worth following. Can AI help us do this? Based on what I’ve seen, I’m hopeful.