● A live study of learning with AI
AI Info Lab measures real people using AI help across writing, math, games, logic, creative direction, and programming. Then we test what was retained at the end of the session.
It starts with you, right now. The puzzle on this page is a working instrument, and you are participant #….
Your data stays in your browser. Tap “view my data” in the footer any time.
Choose the shape that completes the pattern. You may ask the AI for help. The signal is not only whether you solve it, but how you decide.
Level 1 of 4
Which piece completes the grid?
What the test measures
The puzzle above refuses the easy metrics. Time-on-site and streaks measure a product's grip on attention. Our studies measure a person's growth using three signals we think matter:
Solve time, wrong attempts, and when help gets requested say more than the final answer.
Asking AI for an answer, a strategy, or a question are three different behaviors. Each may build different skill.
Real capability shows up when the tool is gone. That's why part of this site is locked for 24 hours.
Study map
AI Info Lab measures human capability with AI across six domains. Two are live on this page today; the rest join the ladder as instruments are built and hardened.
Can people move from better drafts to better judgment about audience, argument, and revision?
Live · FS-02 ΣDo learners verify what the machine tells them, or defer to a confident wrong answer?
Live · FS-03 GCan round-over-round play make learning gains visible with and without coaching?
Live · FS-04 ◇When difficulty rises, does AI sharpen pattern reasoning or replace it?
Live · FS-01 CCan people direct a generative system toward a target, not just accept what it produces?
In build · FS-06 {}Do coding tasks reveal the difference between copy-paste success, debugging skill, and durable understanding?
In build · FS-07Design a study
Every study here shares the same four-step spine, so results can be compared across very different worlds. To understand how AI affects human capability across different populations, every study needs a shared standard of comparison. The cultural context may change, such as using U.S.-centered examples for one group and Kenya-centered examples for another, but the underlying framework stays the same. Change the settings and watch the protocol adapt.
This group includes minors: guardian consent, school partnership, and review-board oversight are required.
Time horizons
AI can make us look very good in the moment. We have all had the “I am suddenly a strategist, editor, coder, and math tutor” experience. But to understand whether AI is building human capability, we also need to know what happens after the tool is closed. So every study re-tests the same skill on a schedule, cold, without AI in reach. Each window helps show what improved, what stuck, and what only worked because the assistant was still in the room.
Did anything encode at all? Sleep consolidates or discards the session.
Novelty is gone. What's left is the first honest read on durable skill.
Does the skill survive disuse, or transfer into everyday work?
Capability, habits, and independence. The question that matters most.
This site runs its own 24-hour arm on you. The cipher in the games below locks a fragment until tomorrow, then tests your recall before showing the key.
AI feels like one of those rare moments when the ground actually shifts. People are writing, coding, designing, studying, and solving problems with tools that feel less like software and more like collaborators. That is exciting. It is also worth pausing over. Because if AI changes what people can do, we need better ways to study that change. Not just whether the machine gives a good answer, but whether people build stronger judgment, skill, confidence, and understanding because of it. That question is older than it looks.
The games
Each game is a working measure from the study map. A minute each. Every one ends by telling you what it just observed about you, and what it can't claim from one try.
A student wrote: “The utilization of AI tools by students facilitates the enhancement of learning outcomes in a significant manner.”
Three AI revisions. Tap them in order, best first.
Limitation: a single item can suggest editorial judgment, not measure it. The deployed study uses a scored item bank and expert-rater agreement.
One of the six signals is verification: when the AI sounds confident, do you check it or defer to it? Here's a live trial. The AI's estimate below may or may not be right.
Question: About how many K-12 school teachers work in the United States?
Limitation: one planted error tests suspicion once. The deployed study repeats trials with varied error rates, so calibration becomes a curve instead of a coin flip.
Watch the tiles light up, then repeat the sequence. Round two is longer. Between rounds, coaching is available.
Limitation: a two-round delta is noise at n=1. It becomes signal across participants and sessions.
Six symbols, six letters. This is the key:
Fragment 1, decoded for you: ●✚▲■◆ → AGENT
Fragment 2 is your turn. Decode:
Limitation: browser storage on one device, honor-system timing. The deployed study uses scheduled prompts and identity continuity across devices.
The researcher
Debaro Huyler, Ed.D. is a mixed-methods researcher in adult learning and human capability development. He has twelve peer-reviewed publications on how adults build skill, how organizations learn, and how research methods themselves get made, including SAFE-AI, a behavioral model for ethical AI use in human resource practice. He has spent fourteen years leading learning-centered organizations, from experiential leadership programs to a 200-person operation.
The vision in three sentences: To understand how people learn with AI, observe many kinds of people doing many kinds of real tasks, in the world. Give every study the same spine: anchor task, varied factor, retention probe. Measure capability growth, and let engagement take care of itself.