AI to people
The explanation gap
People cannot see why a system recommends something. So they check everything again, or quietly stop using it. The time is not saved; it is only moved. We call this the verification tax.
Cognitive AI for high-stakes decisions
AI that reasons with people, not just for them.
NeuroLink AI links neural networks with human reasoning. Experts can follow what the system thinks, and the system understands what experts intend.
Neural learning continuous field
Human reasoning rules and judgement
One field, read two ways. Drag the link to move between them. Select a square to add intent: the learned field bends towards it.
- Learned field
- Rule or judgement
- Expert intent
The problem
Fluent is not the same as understood.
Modern AI is fast, fluent and often right. But in finance, healthcare and other regulated work, every decision has an owner who must follow it, question it and sign it. Today the connection between people and AI breaks in two places.
People to AI
The intention gap
The system cannot see what people mean: the goal behind a request, the rule that matters in this case, the judgement an expert applies without saying it. So people translate their intent into prompts, and still get answers that miss the point.
The idea
One link, built in both directions.
Our name is our thesis. NeuroLink stands for the link between neural networks and human reasoning. We build that link so it carries meaning both ways.
Neural networks
From AI to people
Reasoning you can follow
Every recommendation arrives with its evidence, the rules it applied and the steps it took, in the order and the terms the expert already uses.
In between: a shared reasoning layer
Where learned patterns, explicit rules and a model of how the expert reasons meet, so that people and machine work from the same picture of the decision.
From people to AI
Intent the system understands
Goals, constraints, priorities and expert judgement become explicit structures the system can represent, follow and learn from. Not just a prompt.
Human reasoning
Example
One decision, step by step.
How a recommendation travels to an expert, and how the expert’s judgement travels back.
AI to people
The evidence, the rule, the learned signal and the recommendation arrive in the order this reviewer works, each with its source.
People to AI
The expert model reflects how this reviewer reasons. The decision stays with the person, and an override teaches the system why.
How it works
Three kinds of knowledge, one line of reasoning.
Neural learning
Reads language and finds patterns in documents, records and history.
Symbolic reasoning
Holds the rules, constraints and policies a decision must respect, in a form people can inspect.
Cognitive models
Capture how experts attend, weigh evidence and decide, so the system works in their shape.
Every time an expert overrides a recommendation, the system records why, and the link gets stronger.
Solutions
Decision support for work that must be explained.
We start in finance and healthcare, where decisions are frequent, consequential and reviewed.
| Area | Typical decisions | The difference |
|---|---|---|
| Finance |
| Reviewers see why a case was flagged, which policy applies and what would change the outcome. |
| Healthcare |
| Teams get complete, consistent documentation and plans they can check quickly. People decide; the system does not diagnose or treat. |
| Beyond |
| Any team whose decisions must be explained to a reviewer, an auditor or a regulator. |
The difference
What changes when AI and experts understand each other.
Decisions people can sign
The reasons arrive with the recommendation, not after a request.
Less blind re-checking
Experts check the steps that matter instead of redoing the work.
Policy applied the same way
Rules are explicit and inspectable, not buried in model weights.
Systems that learn from disagreement
Every override is captured with its reason and improves the next case.
Expertise that stays
How your best people reason becomes part of the system, and helps the next person learn.
Ready for review
Inputs, steps, versions and human approvals are recorded for every decision.
Platform
CogniFlow: the decision-support layer we are building.
CogniFlow sits inside existing expert workflows. It prepares a decision, shows its reasoning, takes in what the expert intends and learns from the people who sign it off. We are developing it with partners in finance and healthcare.
Research
Research that starts with how people reason.
Our work draws on cognitive science, neuro-symbolic machine learning and human-AI interaction. We study how experts decide, and how machines can follow them.
- Cognitive architectures for adaptive AI
- Neuro-symbolic reasoning
- Modelling human intent
- Explanations that fit human reasoning
- Measuring human-AI decision quality
Have a decision that needs reasons?
We work with a small number of partners in regulated, high-stakes work. Tell us about a decision your experts make every day.