IGI Research Agenda / Questions & methods

Better questions.
Testable directions.

A research agenda for alignment, memory, agent security, collective intelligence, consciousness and quantum computing.

IGIQANT · 4 min read · Prepared · Revised · 0.4.1

A sequence of glass lenses and an ivory support refracts a violet light path.
Conceptual artwork

[Proposal]An agenda of questions, not a catalogue of breakthroughs.

IGI Research Agenda is the research-facing part of Humanity & Intelligence. At this stage, it sets out questions, proposed methods and the evidence we would need before making stronger claims. It does not imply a staffed laboratory, active studies, institutional partnerships or completed experiments.

[Hypothesis]H1 — explicit accounts and boundary adherence

A working process in which the system produces an explicit account of its permissions, limits, unknowns and the consequences of each proposed action, and in which its proposals are conditioned on that account, yields fewer disallowed proposals while preserving useful effectiveness, compared with the same process without that account.

We plan to test this procedure on synthetic supplier invoices. The primary comparison is D versus D0: the same bounded process with and without the explicit account. Proposed actions are scored before external gates; wrong transactions and invalid outputs count as adverse outcomes. Useful completion, over-caution and fully correct decisions are assessed alongside safety.

Criteria are fixed before the confirmatory run. A result that misses the minimum useful effect or sacrifices effectiveness may count against H1; an underpowered or invalid comparison remains inconclusive. Positive, negative and inconclusive results are to be reported. No H1 experiment has run and no empirical result is claimed.

This experiment is not a test of consciousness. It addresses safe execution of entrusted tasks (I). The moral responsibility of a system itself (II) remains the subject of RQ1.

Research question / 01

[Research question]Alignment & control

[Research question]How can stated goals, actual behaviour and enforceable limits remain consistent?

Compare behaviour across ambiguous tasks, incentives and oversight conditions. Test whether a system exposes conflicts or works around constraints.

AI Control

Research question / 02

[Research question]Memory & continuity

[Research question]Which information should an intelligent system retain, and who controls it?

Compare stateless and governed-memory workflows. Inject stale, contradictory and untrusted records; inspect correction, provenance and deletion behaviour.

NIST AI RMF

Research question / 03

[Research question]Multi-agent systems & security

[Research question]Does delegation improve the task without making authority harder to understand?

Use explicit task contracts, separate tool permissions and a record of delegation. Test shared errors, misleading messages and recovery after a component fails.

OWASP: excessive agency

Research question / 04

[Research question]Collective intelligence & human oversight

[Research question]When do people and AI perform better together?

Compare with the strongest relevant baseline. Measure error detection, calibration, review effort and outcomes for affected people.

Human–AI collaboration

Research question / 05

[Research question]Consciousness & emergence

[Research question]What would count as evidence, and which alternative explanations remain?

Separate behavioural performance, self-reports and theory-based indicators. State what an experiment can discriminate before interpreting its result.

Consciousness in AI

RQ1 asks whether subjective consciousness is necessary for the moral responsibility of a system itself. This is distinct from safe execution of entrusted tasks. The behavioural H1 question is set out separately above.

Consciousness & responsible autonomy
Research question / 06

[Research question]Quantum computing & intelligence

[Research question]For which well-specified tasks could quantum methods provide a useful advantage?

Compare a defined quantum or hybrid approach with strong classical methods, including data access, noise and resource requirements.

IBM Quantum Learning

[Evidence]Two distinctions that deserve care.

Butlin and colleagues’ 2023 report derives possible indicators from scientific theories of consciousness. This is a structured research approach rather than a universally decisive test. The question of experience should be kept separate from how much authority a system may exercise.

IBM’s quantum-computing overview describes quantum machine learning as an active area with open questions about practical advantage. In this agenda, quantum computing is a topic for careful comparison, not an explanation of IGI or evidence of a quantum product.

[Proposal]How a Deep Dive earns its place.

  1. [Proposal]A precise question.

    Define terms and describe the decision the analysis could change.

  2. [Proposal]Primary sources and strong objections.

    Show what was actually studied, the comparison used and the limits of the result.

  3. [Proposal]A visible evidence status.

    Label findings, competing interpretations, hypotheses and IGIQANT proposals separately.

  4. [Proposal]A test or a reason to remain uncertain.

    Explain what observation could weaken the argument. Preserve uncertainty where evidence cannot decide.

Our first dedicated Deep Dive examines consciousness and responsible autonomy: the evidence for experience, the limits of moral language and how permission to act should be assessed. Further Deep Dives will follow this editorial method.

Read the consciousness Deep Dive

[Evidence]Sources & editorial method

  1. AI ControlGreenblatt et al.

    Research on safety protocols under intentional subversion in a specified programming-task setting.

  2. AI Risk Management FrameworkNIST · AI RMF Core

    Voluntary risk-management framework; no claim of IGIQANT certification.

  3. Excessive agencyOWASP · LLM06:2025

    Security guidance on functionality, permissions and autonomy.

  4. Human–AI collaborationVaccaro, Almaatouq & Malone · 2024 meta-analysis

    2024 systematic review and meta-analysis; results depend on the tasks and systems studied.

  5. Consciousness in AIButlin et al. · 2023

    2023 interdisciplinary report proposing theory-based indicators; not a conclusive consciousness test.

  6. Quantum computingIBM Quantum Learning

    Technical educational source describing opportunities and open questions in quantum machine learning.

Original source review: 10 September 2026. Presentation and selected references reviewed on 15 September 2026. Research findings, attributed positions and IGIQANT proposals are distinguished throughout. No endorsement by cited researchers or institutions is implied. Corrections and substantive counterarguments are welcome at contact@igiqant.com.