I spent 38 months watching my own judgment change alongside an AI.

Working from that record, I built three instruments to make different parts of that gap inspectable, and designed a participant study to test the question beyond my own record. I also contributed MCAB, a cognitive-autonomy benchmark, to ImpactBench at the MIT Media Lab.

Research Program

From one record to a study beyond myself.

01 Trace — Start with the record.
Longitudinal Self-Study

Ben Study

Coded and classified into a five-phase longitudinal record, read as data, not memory.

Primary source: a handwritten diary kept in Korean.

How often I corrected the AI
first 18 months
1.7%
next 20 months
4.7%
N≈50/mo · 1,805 labeled · 973 conversations · 48,883 messages
View study
02 Measure — Make the gap scoreable.
felt ≠ trace

Felt vs Trace

A writing probe for the gap the Ben Study found between how independent you felt and where your writing stayed.

Open · v2
3 states · 4 dimensions

MCAB

A benchmark that asks whether an AI response compresses, preserves, or expands the user's thinking.

Contributed · ImpactBench
03 Design — Change the interaction.
two readings

Intuition Canvas

Two readings of the same fragments, from different evidence.

Open · Live
4 advisors

Friction Family

Four AI advisors, each biased on purpose, arguing a decision from different sides while you watch.

Open · v0.9.4
04 Practice — Encounter the same question at scale.
70+ markets

Generative-AI Guardrails

Image and video guardrails for Samsung.com's generative-AI rollout, where I first wondered whether the human review was real judgment.

Shipped · EY / Samsung

Autonomy is not how often you leave the frame. Stay inside a good frame and that may be judgment. Bolt from it and that may be reflex. What matters is whether you can tell which one you just did, and whether you leave when the frame is wrong. The instrument for that is built, and an eight-participant pilot is now testing whether it works on people who are not me. What the pilot cannot do is turn a working instrument into a controlled study. That takes a lab where the question is worth asking.