How AI Decision Making Actually Learns — Not From Clicks, But From Whether You Stopped Looping
- David Harvey
- May 15
- 4 min read
Most AI systems learn from what you click.
Thumbs up. Thumbs down. Time on page. Preference ranking between two outputs. These signals are useful. They tell you whether a response felt acceptable. They do not tell you whether it helped.
Ask Omega* is designed around a different signal entirely.
Did the decision stop looping in your head?
That is a harder question to answer and a more honest one to ask. It is also the question the Comfort Index is built to track. And it is the question that drives how AI decision making inside Omega* improves over time
The three loops that make AI decision making smarter
Omega* does not learn in one loop. It learns in three: running at different speeds, measuring different things, and informing different parts of the system.
The fast loop — per decision
Every question submitted to Ask Omega* teaches the system something about how humans frame decisions under pressure. Not the answer — the question.
The specificity of the language. The domain it sits in. The emotional load carried in the phrasing. The gap between what was asked and what was actually needed.
We built a question readiness scorer into the Ask Omega* interface. It scores your question in real time across dimensions of specificity, grounding, and actionability — and gives you a hint for how to sharpen it before you submit.
That scorer is a learning artefact. Every score it produces, every hint it surfaces, encodes what we have observed about which questions produce high-confidence outputs and which produce hedged, uncertain ones. The scorer learns from the gap between question quality and output certainty — and over time that relationship gets tighter.
The medium loop — per cohort
This is the loop that makes Ask Omega* genuinely different from a decision tool that simply returns structured answers.
After every decision receipt, we surface a two-word prompt. Did this help? Two buttons: "Spin stopped" and "Not yet".
That binary — spin_stopped_yes or spin_stopped_not_yet, is the Comfort Index truth test at the individual level. But at the cohort level it becomes something more significant. It becomes a training label.
As 30 users, then 100, then 1,000 run decisions through the system, the aggregate signal tells us which decision types, domains, and lens translations are actually stopping the spin — and which produce high-certainty outputs that do not create felt relief. That gap is where the model needs work.
Most AI training pipelines optimise for accuracy relative to a ground truth. Omega* optimises for something harder to define and more valuable to achieve: the moment a human feels a decision stop spinning.
That is the Comfort Index as a learning signal. Not how you rated the output. Whether you felt different after it.
The slow loop — per clinical iteration
The third loop runs on a different timescale entirely.
The Comfort Index framework, the SHE ZenAI governance layer, and the Q*(E8) ethical calculus. These evolve through collaboration with clinical partners across high-trust decision environments. Evidence from real decisions, compared against what the system recommended, across outcomes that are measurable over time.
This is not an academic ambition. It is the inevitable consequence of building a decision system that takes evidence seriously. If the system claims to help people make better decisions, it has to be willing to be tested against outcomes. The slow loop is how that test happens.
We are in the early stages of this loop. The clinical partnerships are live. The governance framework is built. The evidence is starting to accumulate. This is the loop that will eventually make Ask Omega* publishable in peer-reviewed contexts — not because we are trying to be academic, but because the evidence will exist and it will be worth sharing.
What the founding cohort is actually doing
This is why the first 1,000 founding Pro Users are not just early adopters.
They are calibration partners in the medium loop.
Every decision they submit, every Lens translation they apply, every spin_stopped_yes or not_yet they register — this is the signal that teaches Omega* what helping actually looks like across real people with real decisions in real domains.
We are not collecting this data to train a generic model. We are collecting it to understand the relationship between decision quality, felt relief, and Comfort Index movement — and to make that relationship more reliable over time.
The founding cohort will see the system improve in ways that later users will benefit from but never witness. That is not a small thing. It is the actual work of building a governed decision intelligence system that earns trust rather than assumes it.
The honest answer
How does Omega* learn?
From the gap between what was asked and what was needed. Whether the spin stopped. From clinical outcomes measured against recommendations over time. From 1,000 real people making real decisions and telling us — explicitly, with two buttons — whether it helped.
Not from clicks. Not from ratings. Not from synthetic benchmarks.
Whether a human felt clearer after using it than they did before.
That is a harder standard to meet. It is the right one.
The 7-Day Comfort Index Challenge is the entry point for founding cohort members. Seven days, one governed decision per day, and a visible record of how your Comfort Index moves across the week.
That record is not just for you — it is the signal that teaches the system what helping actually looks like.
Comments