The Human Agency Project

Independent research institute · Ireland · Est. 2026 · No. 01

Can human agency survive AI?

An independent institute that measures, and works to defend, the conditions under which people stay the authors of their own lives as they hand more and more to machines. Not a warning: a research programme with instruments, experiments, and a way to find out.

42pp
Founding preprint. The Quiet Extinction, the institute's seminal paper.
34
Verified citations. Carried into peer review as the first scholarly output.
6
Measurement dimensions. A validated human-agency & AI-literacy instrument.
9
Frameworks cross-walked, including EU AI Act Article 4.
01 — Mission & the question

The danger most worth studying is not a hostile machine. It is a helpful one.

As capable AI quietly takes over more of what we decide, make, and remember, the risk is not catastrophe but drift: a slow handover in which nothing ever looks wrong. Comfort rises while authorship falls. The founding text behind this institute, the preprint The Quiet Extinction, names the failure modes plainly: meaning collapse, becoming de-prioritised, benign indifference, drift, and the quiet truth that for a person, as for a species, the struggle is part of the point.

Human agency is not a casualty of AI. It is a variable we can measure and protect.

The shape of the human shapes the intelligence that meets them. So the question is researchable, not rhetorical: under high AI capability and adoption, is agency, self-authorship, judgment, relational depth, the passing of skill between generations, preserved or surrendered, and which interventions measurably protect it without giving up what AI is good for?

02 — The research agenda

One question, one conceptual spine, and experiments built to be replicated across countries.

The spine defines the constructs; the experiments measure them. This is what makes it a programme rather than an essay, and what makes it fundable across borders. Seven work packages:

A
The spine: the endorsement-collapse threshold
When does delegating to an AI extend your agency, and when does it quietly replace it? The conceptual work that defines what the rest measures.
B
Measurement
A validated human-agency and AI-literacy instrument, cross-walked to the major international frameworks including the EU AI Act's literacy duty.
C
The convenience-to-surrender slope
A controlled experience that measures how accumulating one-click delegation erodes felt authority, online and at scale.
D
Does literacy preserve agency?
A dose-controlled intervention: structured AI-literacy training as the treatment arm, agency measured before and after.
E
Deskilling and the value of friction
What can people still do once the tool is removed, and does a small deliberate pause restore judgment?
F
Meaning, and why metrics miss it
Testing whether frictionless completion quietly separates meaning from satisfaction, the dissociation our dashboards cannot see.
G
The constructive arm
Scheduled disconnection, and how agency spreads (or fails to) through small human groups. The part that builds, not just diagnoses.

Work packages C, D and E are exactly what a multi-country network replicates: the same protocol, run on each partner's own population. That is research at scale.

03 — The evidence base

Most proposals like this begin with a slide. This institute begins with assets in hand.

  • The Quiet Extinction, the founding paper. A 42-page treatment of indirect, non-malicious AI risk and human agency as an alignment target, with 34 verified citations. preprint · not yet submitted
  • A measurement instrument. A six-dimension AI-competency framework, cross-walked to nine international standards including EU AI Act Article 4. built · validation pending
  • A working experimental stimulus. An interactive that teaches the convenience-to-surrender slope by letting you feel it. working prototype
  • A deployed intervention. A live AI-literacy micro-credential with real learners, ready to serve as a treatment arm. live
  • Recruitable populations. Postgraduate cohorts and small-business staff, a standing sampling frame, with research consent obtained per study.
  • A public voice. The book HUMANLIKE (2024) and a body of essays and talks for dissemination and impact. published

And a tailwind: the EU AI Act's AI-literacy obligation is now in force, with enforcement from August 2026. The institute answers a question regulators are newly, legally obliged to care about.

04 — The Fellows model

The credentialed academic is a collaborator, not a gatekeeper.

The institute holds the idea, the instrument, the data, and the funding. Fellows are invited scholars who co-author the peer-reviewed work and lend disciplinary rigour. In this model the university is a partner, not a permission desk. Founding Fellows are appointed by invitation across the disciplines the question demands:

  • Philosophy of technology and ethics, for the conceptual spine.
  • AI accountability and human-centric AI, for the governance and measurement work.
  • Empirical and quantitative social science, to run the experiments to the standard of their field.
  • Education and assessment, for the literacy interventions.

Partnerships with Irish and European universities, and participation in multi-country research networks, are pursued from this footing: the institute brings the question and the assets; partners bring depth and reach.

05 — Independence & integrity

We state the line up front, rather than hope no one asks.

An institute that studies AI's effect on people, founded by someone whose ventures also sell AI-literacy services, has to hold a clear line.

  • A firewall between findings and commerce. Research conclusions are independent of the founder's commercial work; the conflict is declared, and falsification conditions for any intervention are set before the data is collected.
  • Conventional authorship for every claim. Scholarly outputs are written and peer-reviewed in the normal way, with standard disclosure of AI assistance. Creative and exploratory works in the founder's wider corpus are treated as sources of hypotheses and instruments, never as evidence.
  • Open by default. Instruments and protocols are shared so the experiments can be independently replicated. Replication is the point, not a threat.
06 — How it is funded

A deliberately mixed base, so the work depends on no single gatekeeper.

  • European research networks. Partnership and work-package roles in Horizon Europe consortia on digital society and AI literacy, and a proposed multi-country COST network as the first collaborative step.
  • Philanthropy. AI-and-society funders, who often prefer independent institutes to universities and judge work by whether it matters.
  • Commercial cross-subsidy. The founder's AI-literacy practice (EU AI Act readiness) and the micro-credential fund the research arm between grants.

The honest sequence is partner first, lead later: the institute earns the track record that lets it coordinate, rather than claiming the chair before the work exists.

07 — Leadership

The direction and the assets come from the founder; the scholarly authority comes from the Fellows.

Victor del Rosal is the founder and director. He is the author of HUMANLIKE: The AI Transformation (2024), the founder of fiveinnolabs, a lecturer at the National College of Ireland, and the originator of the conceptual work behind the institute. He convenes the programme, holds its assets, and assembles its Fellows and partners.

He is master's-qualified, and by deliberate choice of this route a practitioner and founder rather than a career academic. The institute's scholarly authority comes from its Fellows and the peer-reviewed record; its direction and its assets come from him. That division is the model, not a gap in it.

08 — Engage

The institute is forming its founding circle now.

If you are a scholar whose work touches the philosophy of technology, AI accountability, the social science of human-AI interaction, or AI in education, and the question matters to you, there is a Fellowship to discuss. If you are a funder or a partner organisation, the same door is open.

Write to victor@fiveinnolabs.com. A short note on how the question meets your work is enough to begin. Below is the kind of note a prospective Founding Fellow might receive, shared here so you know exactly what you would be saying yes to.

Draft · Founding Fellow invitation · paste-ready
Subject: An invitation: Founding Fellow, The Human Agency Project Dear [Name], I am writing to invite you to consider becoming a Founding Fellow of The Human Agency Project, a new independent research institute based in Ireland. The institute studies one question: under high AI capability and adoption, is human agency preserved or quietly surrendered, and which interventions measurably protect it. It starts with assets already in hand, a conceptual paper, a validated measurement instrument, a working experimental stimulus, a deployed literacy intervention, and recruitable cohorts, and a programme of experiments built to be replicated across countries. I am approaching you because your work on [specific area] is central to [work package / dimension], and because I would rather build this with people whose standards exceed my own. A Founding Fellow co-authors the peer-reviewed work in their area, helps shape the research design, and lends the institute the disciplinary rigour it is built to respect. The institute holds the idea, the instruments, the data, and the funding; Fellows are collaborators, not administrators. If the question interests you, I would welcome a short call. I can send the preprint and a one-page outline of the programme in advance. With respect, Victor del Rosal Founder and Director, The Human Agency Project A fiveinnolabs initiative victor@fiveinnolabs.com