Research Ethics
The stance that shapes how we build, run, and retire experiments.
What we believe
Uncertainty is a feature, not a bug
AI systems don't know what they don't know. Hiding this is deceptive. Exposing it is honest. We design experiments that make uncertainty visible and central to the experience.
Consent is not fine print
We don't bury consent in terms of service. Participation in our experiments requires explicit, informed opt-in. If you didn't actively choose to participate, you're not in the experiment.
Reasoning matters more than outcomes
The value of our experiments is not whether they "work." It's whether they reveal how AI systems reason, where they fail, and what assumptions they encode.
No personalization theater
We don't pretend to know you. Our experiments may use signals to estimate appropriateness, but we don't claim accuracy or construct persistent profiles. Every output is illustrative, not predictive.
Failure is documentation
When experiments miss, we document why. When they raise ethical concerns, we retire them transparently. The log of what didn't work is as valuable as what did.
What we commit to
Users know what the system is doing, what signals it uses, and how it reasons.
We don't pretend systems are more capable, accurate, or personalized than they are.
Consent is never assumed, implied, or inherited from other interactions.
No persistent profiles, cross-session tracking, or identity linkage.
Confidence levels, limitations, and failure modes are visible, not hidden.
Users can leave experiments at any time without explanation or penalty.
When experiments fail or raise concerns, we retire them transparently and document why.
What we don't do
We don't gamify uncertainty
No scoring, leaderboards, or competitive elements. The goal is understanding, not engagement metrics.
We don't A/B test ethics
Consent gates, uncertainty disclosure, and transparency are not variables to optimize. They're non-negotiable.
We don't apologize for misses
When systems get things wrong, we explain why—but we don't perform regret. Misses are expected and documented, not failures to apologize for.
We don't hide behind "research"
Calling something an experiment doesn't excuse poor design, deceptive framing, or lack of consent. Research is a responsibility, not a shield.
Why this stance
The AI industry has normalized personalization theater: systems that pretend to understand you based on shallow signals, deliver outputs with false confidence, and bury uncertainty in disclaimers.
This isn't just bad UX. It's epistemically dishonest.
We build experiments that do the opposite: make reasoning visible, expose limitations, and treat uncertainty as structural rather than incidental.
This isn't a marketing position. It's a technical and ethical commitment to building systems that don't pretend to be what they're not.
How we enforce this
Every experiment gets a consent gate
No passive participation. If you didn't explicitly opt in, you're not in the experiment.
Every experiment exposes its reasoning
What signals does it use? How does it reach conclusions? What's the confidence level? These aren't optional disclosures—they're core to the experience.
Every experiment includes a "why this may have missed" section
We don't wait for users to complain. We preemptively explain why the system might be wrong.
Every experiment has an exit
"Not for me" buttons. "Back to Research" links. No dark patterns to keep you engaged.
Every retirement is documented
Archived experiments show sunset notices. Retired experiments explain what went wrong. Nothing disappears without explanation.