Keeping human judgment economically indispensable
As AI systems move from tools to autonomous agents, societies must decide what they are optimizing for.
Hybrid Intelligence proposes a new path: designing human–AI systems so human capability, judgment, and meaning remain central.
AI used to do tasks. Now it takes on whole jobs. Most people’s answer is to move humans to the edges: set the goal, check the output, stay out of the way.
We think that answer is based on a test nobody has actually run.
You’ll often see a study where AI performs better than a professional using AI. But look at what “using AI” meant. Almost always the same thing: the AI hands over an answer, and the person decides whether to take it. That’s the thinnest version of working together there is. Researchers have spent a decade building much better ways for people and AI to work together — and none of them were in the test.
A Multi-Level Socio-Technical Framework for Keeping Human Judgement Economically Valuable
Why better interfaces aren’t enough
To design a good human–AI workflow, you first have to ask people a simple question: in your job, which parts could a machine work out, and which parts need your experience and your read of the situation?
That’s useful information. It’s also a map of how to replace someone.
So if a company has just announced job cuts, people keep the good stuff to themselves. The human–AI version then works badly, which seems to prove that automation was the right call all along. Automation didn’t win. The alternative never got what it needed.
That’s why we work on four things at once:
- the tools people use day to day
- the skills people build while using them
- the companies that decide how work is organised
- the markets and rules that decide what gets paid for
Each one only holds up if the one above it does. Improve just the tools, and everything else drags you back to the default.
Image: The HI Manifesto Series: Interconnected Missions and Companion Papers
The question we’re trying to answer
How do we make sure human judgment still earns its keep, now that AI can not only do the work but plan and organise it?
This isn’t about keeping people around out of kindness, and it isn’t about making someone click “approve” on everything an AI does. It’s about building work where a person’s judgment genuinely changes the outcome — in most of the economy, not a small corner of it.
We want to break work into small steps with a person steering at each one, and the quality comes from how the work is organised rather than from having the biggest model. If we’re right, smaller and cheaper models do the job just as well. That means lower costs, less energy, and less dependence on a handful of giant AI providers.