Technique Makes Elevation

Case of the Week, Min Wu, PhD · ai-public-health.com

Several months ago, I wrote about learning to pole vault at fourteen,
on a construction-site sand pile, with a borrowed bamboo pole and my
worried father as coach. I closed that post with a question I was not
yet ready to answer:

"What is the planting technique for AI? Inside a real public health
workflow, what is the specific move that converts the kinetic energy
of AI use --- all that prompting, querying, summarizing --- into the
potential energy of elevation: a better decision, a clearer insight, a
sounder judgment?"

This week I want to walk back to the sand pile and answer it ---
because the answer was in the physics of the vault all along.

But first, a scene you may recognize. A public health team adopts AI.
Within weeks, every meeting has an AI-generated summary. Every report
starts with an AI-drafted outline. Every dataset gets an AI-produced
visualization. The team is moving faster than ever. And yet, three
months later, the community health improvement plan reads almost exactly
like last year's. The activity doubled. The outcomes did not move. If
that pattern sounds familiar --- if you have watched a team get busier
with AI without getting better --- then this post is about why.

What I Actually Learned on the Sand Pile

When I was fourteen, I thought the planting technique was one thing ---
get the pole vertical. Do that, and you convert. Fail to do that, and
you fall back on the sand pile.

I was wrong. Even a child's vault on a sand pile has two things
happening at once. First: where do you plant the pole --- shallow in the
sand, or deep into the box (in real vault, the metal slot at the base of
the pit)? Second: what kind of pole is it --- a rigid one that transfers
your energy, or a bending one that stores and releases it? These are
independent. You can plant a rigid pole deep in the box and still not
convert well. You can have a bending pole and plant it in the wrong
place, and it bends over nothing.

Two dimensions, not one. The planting technique for AI has the same
shape.

The first dimension is where AI sits in the workflow. Not how much
AI --- where. Every workflow has peripheral steps (drafting emails,
formatting briefings, summarizing notes) and core steps, where judgment
reads across evidence and produces a recommendation the team will act
on. A team can use AI heavily at every peripheral step and have no AI
anywhere near the core. From outside, this looks identical to a team
using AI well. The pole is being planted in the sand, not in the box.
The energy does not convert.

The second dimension is how AI participates in the work. AI can
transfer information --- produce a summary, a chart, a briefing that
humans read and then act on. Or AI can participate in the action itself,
holding context across steps and driving what happens next. These are
different kinds of poles.

A team can be busy with AI at every peripheral step and still have AI
nowhere near the moment where the workflow actually creates value.

From Bamboo to Bending Pole

Beginners use a rigid pole --- in my case, literally bamboo. A rigid
pole transfers energy: your kinetic energy goes in, gets levered around
the plant, and lifts you as high as geometry allows. That is the
ceiling. Advanced vaulters use a bending pole (fiberglass or carbon),
which does not just transfer energy. It stores it, then releases it. The
pole bends under the runner's momentum, holds that energy for a moment,
then actively drives the vaulter upward as it straightens. A bending
pole is not a passive lever. It is a participant in the conversion.

AI integration has the same ceiling. Most public health AI today is
bamboo --- dashboards, summaries, reports. It transfers information.
Humans read, decide, and act. That is useful, and it is also a ceiling.
The next generation --- what is being called agentic AI --- is the
bending pole. It does not just transfer output to a human. It holds
context, reasons across steps, and drives the next action. The depth of
integration is not a yes-or-no question. It is a progression, from rigid
to bending, from transfer to participation.

But not every bending pole is fit for public health work. In an earlier
Case, My Basement Flooded — and It Wasn't an Information
Problem, I argued that public health AI agents
need three features beyond the classic perceive-think-act loop:
compliance (is the action consistent with laws, guidelines, and what
real people will actually do?), efficacy (will it prevent harm or
save lives, here, now, under this constraint?), and autonomy (can it
act independently while leaving the person's choice and dignity intact?).
Those three features are what make a bending pole fit for public health
--- they are how an agent stores energy without springing it back the
wrong way. A bending pole without them is not safer than a rigid one. It
is faster at the wrong thing.

Why Both Dimensions, Together

The two dimensions are independent. A team can plant at the core step
and still hold a rigid pole whose output ends as a slide no one acts on.
Another team can hold a bending pole that triggers real action --- over
peripheral decisions that were not the real value-creation step. Both
teams are using AI. Neither is converting.

Both dimensions have to be right at the same moment --- the right place,
the right kind of pole. This is why running faster does not help.

The Diagnostic, in Plain Language

Before a public health team commits to a workflow design with AI, two
questions are worth sitting with:

One. Where in this workflow is value actually created --- and is AI
near that step, or only around it?

Two. At that step, what kind of pole is the AI --- a rigid one that
hands information to a human, or a bending one that participates in the
action itself?

Neither question asks how much AI. Neither asks how advanced. They ask
where, and what kind. That is the planting technique.

The Observer's Insight

AI adoption in public health is not primarily a tool problem or a
training problem. It is a workflow-literacy problem. Teams that can see
their own value-creation workflow clearly --- and see where an AI output
would actually travel after it is produced --- will use AI well. Teams
that cannot will use AI fast, at every peripheral step, and wonder why
the outcomes are unchanged.

Technique makes elevation.

A question for you this week: take a workflow you run. On the first
dimension, is your AI use near the core value-creation step, or only
around it? On the second dimension, is your AI a rigid pole ---
transferring information to a human --- or a bending one that
participates in the next action? Reply or leave a comment --- I would
love to hear where your team is planting.

As a member of the Springer Nature Author Affiliate Program, I may earn
a small commission from purchases made through this affiliated link to
my book. Support the author by checking out my textbook, Artificial
Intelligence in Public Health: https://tidd.ly/4mH9389