Two Touches: What Weekend Soccer Taught Me About Designing AI

I turned 50, and I am still playing indoor soccer on weekends. I still have fun — but my game has changed. Somewhere in the last few years, I started telling myself two words before every ball came toward me: two touches.

First touch: receive the ball cleanly. Second touch: pass it, on the ground, at the right speed, to a teammate in a better position than I am. That’s it. No dribbling through three defenders. No trying to be the player I was in my thirties.

Two-touch is not a retreat. It is a discipline. It demands two things that get harder with age, not easier: technique (a clean first touch, an accurate pass) and vision (knowing where your teammates are before the ball arrives).

I simplified my style. I still have my position on the court.

I have been thinking about two-touch soccer while I think about AI.

The Real Problem: Interface Overload

Most of the AI tools I see being built for public health right now are mazes. A user opens the dashboard and faces seventeen filters, four model options, three export formats, and a “methodology” dropdown that unfolds into five more sub-choices. The designers are proud of the complexity — look how much the tool can do.

But the user — a public health officer trying to decide whether to issue a heat advisory by 3 PM — gets lost. Or worse: does not get lost, but spends forty-five minutes navigating when they needed an answer in five.

Now scale this up. Imagine a state health department during a weeklong heatwave. The data is there. The models are accurate. The dashboard has every filter anyone could want. And yet the decision — issue the advisory, at what threshold, for which ZIP codes, by when — still falls on a tired human navigating menus. We used to worry about information overload: too much data to find what matters. Now the tools know what matters, but the interfaces bury it. Call this interface overload — too many paths to the same decision. The system can answer the question. The user can’t find the door.

From Mazes to Straight Lines

A good AI tool plays two-touch. It receives the complex input — messy data, unclear intent, noisy signal (the first touch). And then it passes, on the ground, in the simplest direct path, to the decision the user actually needs to make (the second touch). One touch to receive. One touch to deliver.

This is harder to design than a maze. A maze is what you get when no one made the hard decisions about what to leave out. A straight line requires the designer to know the field — to know which question the user really came to ask, and which complexity is serving the tool rather than the user.

Early software interfaces showed you everything because they didn’t know who you were or what you were doing. We kept that habit long after the tools got smart enough to know both.

The Two-Touch Upgrade: Four Design Principles

When I think about what separates a two-touch interface from a maze, four principles keep surfacing — the same four, across very different tools.

Active, not passive. A maze-interface waits. It shows you everything it can show you and lets you figure out what to ask. A two-touch interface moves first. It opens with the question the user most likely came to answer — Should I issue an advisory today? — and makes that the default view. The user is not searching. The user is confirming or correcting.

Role-based, not universal. A maze tries to serve every possible user from one screen — epidemiologists, health officers, clinicians, communications staff. A two-touch interface knows who just logged in and shows them only what their role actually needs. The epidemiologist sees models. The officer sees a recommendation. Same underlying system. Different first touch.

Task-oriented, not feature-oriented. A maze is organized by what the software can do: here is the filters menu, here is the export menu, here is the modeling menu. A two-touch interface is organized by what the user came to do: draft today’s advisory, review yesterday’s decision, prepare the weekly report. The tasks are the navigation. The features are plumbing.

Spatial, not list-based. A maze buries information in nested menus and dropdowns. A two-touch interface uses the screen the way public health has always used maps — as a surface where relationships are visible at a glance. You see the hot zones, the gaps, the anomalies. You don’t read them off a table. You see them the way you see a teammate open on the wing.

Technique and vision. Same as soccer.

The Observer’s Insight

I will be honest about where I sit with all of this.

I have been sketching design principles like these for a while, but I have not pushed them hard into the field. The reasons are practical. Vendors have built their products around the maze — the complexity is the sales pitch. Health systems are conservative, and rightly so: a simplified interface that hides a critical option is worse than a cluttered one that shows it. Clinical workflows accumulate features the way old houses accumulate additions, and nobody wants to be the one who tears down a wall that might be load-bearing.

So I hold this loosely. I think the maze is costing us more than we are measuring — in officer time, in decision fatigue, in the quiet drift of people who give up on a tool and go back to spreadsheets. But I have not proven it. The field has not given me the room to try.

What I can say is this: the discipline travels. On the soccer court at 50, I cannot do what I could at 30. So I play two-touch, and I still have my position on the court.

Public health AI does not lack capability. It lacks the discipline to turn capability into one clean decision.

The tool should take the complicated touch. The user should be free to play.

Have you ever walked away from a dashboard or AI tool that technically had the answer you needed — because finding it took longer than the decision was worth? What would a “two-touch” version have looked like? Reply or leave a comment — I’d love to hear your story.

— Min Wu, PhD Associate Professor, Zilber College of Public Health, University of Wisconsin-Milwaukee

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