AINOS RESEARCH

Better decisions begin
with harder questions.

We study what it takes for AI to support consequential decisions—and where today’s approaches fall short.

See our approach
RESEARCH / PREPRINTJune 2026AiNOS Research

An explanation should help you review a decision.

What does it mean for an AI-assisted decision to be explainable? This paper proposes that a reviewer must be able to reconstruct it, challenge it, govern it, and understand its limits.

WHAT THIS MEANS FOR OUR APPROACH

Design around the decision record: the evidence used, the reasoning applied, the uncertainty that remains, and the authority to act.

Auditable Decision Intelligence: A Decision-Centered Definition of Explainable AI for High-Stakes Decisions

Read the paper Download PDF
Go deeper into the research

This conceptual preprint develops a decision-centered definition of explainable AI. Its four requirements—reconstructability, contestability, governability, and boundedness—shift attention from a fluent explanation to what a reviewer can actually inspect and control. It is a research proposal, not a certification or a measured deployment result.

RESEARCH / PREPRINTAugust 2026Jimin Kang · AiNOS Research

A connected graph can still lose the meaning.

Turning documents into a structured graph does not automatically preserve what the source means. This audit examines how meaning can be lost even when outputs look organized and are linked to evidence.

WHAT THIS MEANS FOR OUR APPROACH

Keep original sources accessible and evaluate whether relationships preserve their meaning before using a compiled model to support a decision.

Semantic fidelity limits of language-model ontology compilation

Read the paper Download PDF
Go deeper into the research

The preprint examines semantic fidelity in language-model ontology compilation through a targeted audit. Structural validity and source links do not, by themselves, establish that the source’s meaning has been preserved. The findings motivate a source-first design and semantic evaluation; they do not establish that all ontology compilation is impossible.

A CLEARER WAY FORWARD

Let’s explore the next question together.

Start with the decision your team needs to make.

Let’s talk