What happens to civil society when development intelligence becomes cheap?

Contributed by our ED - Craig Kensley

There is a particular kind of anxiety around AI at the moment. What will it replace? Which jobs disappear? What happens when machines can do things we once considered skilled human work?

This weekend, I shared a space with someone who has spent decades leading and repairing complicated institutions. The conversation eventually arrived at AI, as conversations tend to these days.

He offered six things that he believes will distinguish meaningful human contribution as AI becomes more capable: domain expertise, judgement, professional scepticism, curiosity, stewardship and empathy.

I've been thinking about that list, particularly in relation to civil society.

A fair amount of what development organisations do is about to become very cheap. Research, proposals, theories of change, funding scans, policy summaries, workshop designs, monitoring frameworks, literature reviews, reports, communications and first drafts of strategies. AI can already do much of this reasonably well. It will get better.

This is slightly uncomfortable because development has built quite a substantial professional industry around producing these things. We employ people to write about problems, commission people to analyse them, hold workshops about them, produce frameworks for responding to them and then write reports explaining what happened when we did.

None of this is necessarily pointless. Some of it is essential. But its scarcity value is changing.

If a small organisation with decent people and good AI capability can produce in a day what previously took a technical team two weeks, an obvious question follows: what exactly are we paying development organisations for?

Perhaps those six ideas offer part of the answer.

Domain expertise: AI knows an extraordinary amount. Ask it about youth unemployment in South Africa and it will explain the structural drivers. Ask about ECD and it can summarise the evidence. Ask about municipal water systems, disability inclusion or enterprise development and it will produce something quite respectable. The problem is that knowing about something is not quite the same as knowing how it behaves.

Anyone who has worked in development for long enough knows the difference. A programme works in one province and collapses in another. The stakeholder with the impressive title turns out not to be the person who can make anything happen. The community isn't participating, not because it is "hard to reach", but because the programme has misunderstood how people actually organise themselves. The intervention is technically sound and institutionally impossible.

You learn these things by being there.

Expertise in our sector is partly knowledge, but it is also accumulated context. It comes from relationships, mistakes, conversations, political reading and experience. It is knowing that the official organogram and the way decisions actually get made are often two different things. It is recognising the small detail that will derail an otherwise excellent programme because you have watched something similar happen before.

In an age of abundant global intelligence, genuinely local intelligence may become more valuable.

Judgement follows closely behind it because development rarely presents us with clean choices. The funder wants one thing, government needs another, the community sees the problem differently and the implementing organisation has constraints nobody included in the programme design. The evidence points in three directions and the budget points in a fourth.

Eventually somebody has to make a call.

Perhaps this changes what funders should think they are buying. Not another 70-page strategy or beautifully formatted situational analysis, but considered answers to harder questions. What is actually happening here? Where would you intervene? What wouldn't you fund? Which assumption worries you? Who aren't we listening to? What should we stop doing?

AI can help assemble the evidence. Someone still has to take responsibility for the decision.

This is also why professional scepticism becomes more important, not less. AI is extraordinarily fluent. It can produce something that sounds authoritative long before anybody has established whether it is true, which should concern a sector already susceptible to fashionable language, imported frameworks and ideas that become orthodox largely because enough respectable institutions repeat them.

Imagine that tendency at machine scale. The same evidence bases, terminology, international frameworks, datasets and assumptions recycled through thousands of strategies and funding proposals. Everything beautifully written, properly structured and thoroughly plausible. And everything potentially missing the point.

Civil society will need people willing to interrogate the machine's version of reality.

Whose data is missing? What does "employment" mean in this dataset? Who decided what successful participation looks like? What informal activity has disappeared because nobody measured it? Which community has become statistically invisible? What assumption arrived from somewhere else and quietly became universal?

That is not cynicism. It is the discipline of refusing to confuse confidence with truth.

Curiosity becomes interesting for the same reason. AI is exceptionally good at answering questions, which makes asking the right question more valuable.

Development frequently starts with the wrong problem. We think we have an education problem and discover a transport problem. We think we have a skills problem and discover a networks problem. We think we need another entrepreneurship programme and discover that entrepreneurs are already there but procurement systems exclude them. We think a community lacks resilience when the real issue is that we have designed systems which require people to be endlessly resilient.

Curiosity is what leads us to the systemic level.

AI will make us extremely good at optimisation. Once we decide what success looks like, the technology will increasingly help us find faster and more efficient ways of getting there.

The danger is obvious: becoming extraordinarily efficient at solving the wrong problem.

Then there is empathy, which is perhaps more complicated.

Development has always had a tendency towards abstraction. Beneficiaries. Households. Target populations. Vulnerable groups. Numbers reached. We need these categories because institutions need ways to understand scale, allocate resources and measure progress. AI gives us enormously powerful tools for making sense of them.

But eventually the 4,700 people in the dashboard are actual people. Their lives do not organise themselves according to our indicators. They live inside families, communities, informal economies, histories, obligations and systems that frequently make very little sense from the outside.

AI can model empathetic language remarkably well. Development still requires somebody to care enough to notice when the model of the person bears very little resemblance to the person.

Which brings me to the word on the list that has the longest echo: stewardship.

Stewardship asks what we should do with that possibility of the muscle AI offers us.

An algorithm can recommend where resources should go, but someone must decide whether the criteria are fair. AI can identify the most efficient intervention, but somebody should probably ask: efficient for whom? It can analyse communities, score funding applications, recommend investments, identify risks and optimise programmes. Those capabilities will improve quickly and many of them will be enormously useful.

Someone still has to own the consequences.

This may point towards a larger question for civil society. For years, part of our value has been our ability to produce the machinery around development: the proposal, the framework, the research, the programme design, the report and the presentation. That machinery isn't disappearing, but increasingly machines will help us produce it.

For us, the value of civil society therefore moves closer to the things that were always more difficult to put into a deliverables table: proximity, judgement, contestation, contextual intelligence, relationships and trust. The ability to represent realities that don't appear neatly in datasets. The willingness to ask inconvenient questions of funders, governments and ourselves. The ability to recognise when the strategy is working beautifully on paper and failing completely in real life. Stewardship.

I don't think this is necessarily bad news for civil society. It might actually force a useful correction.

Development has never really suffered from a shortage of documents. We have an astonishing number of them. Our harder problems have always been human: knowing what matters, recognising when something isn't working, understanding why, making choices when the evidence is incomplete, navigating power, listening properly, changing course and taking responsibility.

AI may make development organisations dramatically more productive. I hope it does. There are plenty of things we currently spend days doing that deserve to take twenty minutes.

Let the machine help us write the report.

The more interesting question is whether that frees us to become better at the work the report was supposed to be about.

There is a harder possibility here too.

Some civil society organisations will not survive this transition.

AI will make it much easier to distinguish between organisations that produce the appearance of development and those that possess the judgement, relationships, expertise and legitimacy to actually move it.

When knowledge and analysis become cheap, packaging them is no longer much of an advantage.

Funders will have access to much of the same intelligence. They will be better able to interrogate assumptions, compare costs and ask a much harder question of the organisations seeking their money: why you? 

A polished proposal or impressive framework will become a weaker proxy for actual capability.

We are entering a credibility economy and a human economy at the same time.

Looking competent becomes easier, so actually being credible becomes more valuable.

Intelligence becomes abundant, while judgement, trust, relationships, empathy and responsibility remain scarce.

That will call some bluffs. Organisations built largely around the paperwork of development will struggle to hide behind it. Those that survive may be smaller, faster and technologically capable, but they will also need to be more embedded, more accountable and demonstrably useful.

My sense is that AI will turn out to be less of a threat to civil society than a rather unforgiving audit of what we were actually contributing in the first place.

This is how we #GrowZA