AI Reasoning
Lab.
A playable research environment where learners ask their own question, explore evidence as a three-dimensional world, alter the information available to an AI agent, and compare how its conclusions change across iterations.
Enter AI Reasoning Lab ↗Can AI reasoning become something people learn by navigating rather than reading about?
The project tests whether evidence, contradiction, confidence, and context can become spatial and interactive rather than abstract and textual. Learners enter a three-dimensional world built around their own question, where they encounter supporting and contradictory signals as objects they can inspect, compare, and manipulate. Their movement, choices, and revisions become part of the experiment itself.
The objective is not to reveal a hidden chain of thought. It is to make the conditions that shape an answer observable and changeable.
The first prototype made confirmation bias physical.
The fixed Northstar investment story established the core vocabulary of the lab. Supporting evidence appeared as glowing objects along a path toward a mountain of certainty. The more confirming signals the learner collected, the darker the surrounding world became. A contradiction door eventually appeared, leading to customer-risk evidence that broke the bridge to the original conclusion. This prototype proved that ideas could be destinations, evidence could be objects, contradictions could be obstacles, and bias could be an environmental event.
The first world made bias tangible—but the conclusion was still predetermined.
Users could explore a model of reasoning, but they could not ask their own question, inspect the agent's context, choose which evidence to include, rerun the experiment, or compare how different information changed the conclusion. These limitations became the design brief for the next iteration.
Personal questions create genuine inquiry.
A learner builds stronger intuition when the reasoning world is constructed around a question they actually chose.
Contradiction must exist from the beginning.
Counterevidence should be available as a legitimate path, not introduced only as a scripted surprise.
Seeing change is not enough.
Learners need to manipulate the agent's available information and observe why its conclusion changes.
A learner's own question now becomes the landscape.
In the current lab, research agents assemble supporting and contradictory signals from the start. Each evidence lesson is rendered as a visual object with headline statistics, graphs, bullet points, source attribution, and a confidence response. An observer agent detects when the learner explores only supporting evidence and makes the world darken, opening a contradiction door that invites a fuller investigation.
The learner is no longer walking through our conclusion. They are investigating their own.
The evidence room lets learners change the world the agent can see.
After the first expedition, users enter an evidence room containing every signal the agents discovered. They can include or remove individual pieces of evidence and watch the projected confidence react in real time. When they rerun the same question, the new world contains only the selected evidence. This teaches context dependence without exposing private chain-of-thought reasoning.
Every iteration preserves the conclusion, confidence, and evidence behind it.
The final window compares all runs side by side. Learners see each conclusion, the confidence attached to it, the support and contradiction that were included or excluded, and a plain-language explanation of why the answer shifted.
The agent did not simply change its mind. The learner changed the world of information it could see.
The current lab turns AI reasoning into a repeatable experiment.
The complete flow now runs from a learner's own question through research agents building a world, visual evidence lessons, bias detection, the contradiction door, the evidence room, repeated reruns, and a reasoning history that compares every iteration. Both the live lab and this research notebook remain active and evolving.
Every system in the Decision Systems Lab is engineered to outlive its author : a small, durable piece of operational thinking made legible to the next engineer.
Technical documentation for this system is currently being developed as part of the Decision Systems Lab. This page will evolve alongside the project as research, implementation, and validation progress.