Institute for AI Research

Research

Questions before technologies.

IAI research begins with problems that require stronger experimental, engineering or reasoning foundations. Technologies are selected and qualified as instruments; they do not define the research programme.

Areas of inquiry

Current research

Public programme descriptions are intentionally human-facing. Internal build and repository identifiers remain part of provenance rather than primary navigation.

  1. 01 Active research

    Computational experimental sterility

    Research into how inherited computational state can be controlled before an experimental unit begins, and how logical sterility can be distinguished from physical experimental independence.

    Questions under investigation

    • What state can cross an experimental boundary?
    • How can contamination controls be made experimentally observable?
    • When is independence measured rather than merely assumed?
  2. 02 Active research

    Scientific paper structure

    Empirical investigation of the deeper functional structure of scientific papers: segments, rhetorical moves, epistemic objects and the relationships that make scientific arguments work.

    Questions under investigation

    • What structural units recur across scientific papers?
    • How do those units relate to claims, evidence and method?
    • Can paper structure be represented computationally without reducing its meaning?
  3. 03 Active research

    Graph and reasoning systems

    Research into graph-based representation and reasoning capabilities spanning probabilistic inference, causal reasoning, argumentation, structural analogy and related scientific reasoning methods.

    Questions under investigation

    • Which reasoning capabilities are genuinely distinct?
    • How should capabilities, providers and evidence be represented?
    • How can reasoning systems remain inspectable and replaceable?
  4. 04 Active research

    Institutional AI systems

    Engineering research into institutional AI infrastructure whose evidence, operational state and authority remain explicit, governable and separable from implementation technology.

    Questions under investigation

    • How should institutional AI systems preserve provenance and authority?
    • How can infrastructure remain replaceable without losing meaning?
    • What should be governed centrally, and what should remain repository-owned?
  5. 05 Research programme

    Defensive AI and zero-day research

    Defensive research into methods for discovering and reasoning about previously unknown software vulnerabilities using deterministic, graph and bounded semantic techniques.

    Questions under investigation

    • What evidence supports a defensible zero-day hypothesis?
    • Where can graph structure reveal previously unseen risk?
    • How should semantic instruments remain bounded by deterministic evidence?

Research posture

How the work is approached

01

Inspectable

Methods, evidence, uncertainty and decision boundaries are treated as part of the research record rather than hidden behind final conclusions.

02

Prospective where possible

Experimental controls, statistical plans and acceptance criteria are formed before execution when the research question permits it.

03

Implementation-aware, not implementation-bound

Tools and technologies are qualified for the problem at hand, while institutional meaning remains separable from any particular implementation.

Explore research methods

Boundary

Research records remain authoritative at their source.

The public website is a discovery and presentation surface. It does not replace the governed repositories, evidence stores or publication records that establish the authoritative research record.