Institute for AI Research

Independent research · Brisbane, Australia

Researching the foundations beneath AI systems.

The Institute for AI Research investigates how artificial-intelligence systems can be engineered, experimented upon and evaluated with stronger evidence, clearer provenance and more defensible scientific methods.

Explore research Research methods

Position

Evidence before assertion. Method before scale.

IAI is an independent research institute concerned with the experimental and engineering conditions under which claims about computational systems can be made responsibly. The work is designed to be inspectable: methods, evidence, uncertainty and decision boundaries are treated as part of the research record.

Current research

Active areas of inquiry

All research
  1. 01 Computational experimental sterility Methods for controlling inherited computational state before an experimental unit begins, and for distinguishing logical sterility from physical experimental independence.
  2. 02 Scientific paper structure Empirical research into the deeper functional structure of scientific papers: segments, moves, epistemic objects and the relationships between them.
  3. 03 Graph and reasoning systems Institutional reasoning capabilities spanning probabilistic inference, causal reasoning, argumentation, structural analogy and graph-based scientific representation.
  4. 04 Institutional AI systems Engineering and governance research into reproducible, inspectable AI infrastructure whose evidence, decisions and operational state remain separable.

Research practice

The method is part of the result.

Reproducibility, prospective design, provenance, validation and explicit claim boundaries are not publication afterthoughts. They are engineered into the research process.

How IAI approaches research

Publications

Research should remain inspectable after publication.

IAI publications are intended to connect the human-facing paper to its methods, provenance and supporting research record wherever those materials can be made public.

View publications