Research
The problem
About 7% of all humans who have ever lived are alive today, and we're about to add
countless AGI. Yet we can't finish HS2 or the California Bullet Train. Ambitious projects
seem to either grind to a halt under the weight of their own complexity, or, to avoid that,
all the little hacks that make the system work in practice get squashed to make it
understandable
(legible).
This is a problem for every government. The NHS in theory should have standard processes to
give them the efficiencies of scale you expect from a health service that employs 6% of a
population. But the day to day of a GP practice in a small seaside town full of retirees
looks nothing like an oversubscribed inner city one. The same goes for schools.
It's worth comparing how the Roman empire must have run to the information flow in
organisations today. Small letters were sent by boat and horseback from Egypt to Rome every
few months. Whereas today, we're supposed to answer our colleague's Slack ping at 9pm on a
Sunday.
As the cost of communication falls, accountability has an opportunity, and therefore a
tendency, to centralise. But the more that happens, the more accountability wants to simplify
projects. And the more that happens, the more the local nuance which actually makes a thing
work gets smoothed away.
Solution
The focus of my research is to build concrete tools to enable organisations to embrace the
power complexity brings rather than fear what it takes away. Answering, how do we understand
systems holistically enough to steer them, while encouraging all the hacks that make them
practical.
First I published a leading metric of
complexity, based on
information theory. If you can't measure a system's complexity, you can't manage or diagnose
it.
Then I created a sister measure for
self-organisation, to
identify when a complex system is merely chaotic rather than full of sophistication.
AI
AI changes all this, for better and worse. Adding thousands of agents into an organisation
increases its complexity exponentially. It puts extreme pressure on information that has to
travel through traditional hierarchies, both because of the speed agents work at and their
volume.
The good news is agents are exceptionally good at rationalising information, so more parts of
an organisation can stay informed. But the human mind's capacity for understanding complexity
is still fixed. So accountability and execution have to become decentralised again to cope
with AI running at any reasonable scale.
This is why I'm now building Concentric.