Reviewed before it begins
Any study involving people will go to the IFI Responsibility Board, currently in development, for ethical review before a single participant is contacted. No exceptions, including for our own convenience.
The research arm of Idea Fields Institute
IFI Research
Cross-disciplinary inquiry, conducted in the open: careful about people, honest about uncertainty, accountable to real conditions.
01Focus
Areas of inquiry.
What IFI Research studies: people, machine intelligence, and where the two meet. The published work so far measures the machines themselves; the human-facing studies are getting underway, and the scope widens as the work does.
-
Psychology & behavior
How people actually think, decide, cope, and change: studied under conditions that resemble life rather than the lab, with special interest in experience that resists clean measurement.
-
People & machine intelligence
What happens when people and AI systems meet in earnest: trust, reliance, consent, and the ways these relationships shape how people think and act. The Institute's first human-subjects study takes up one corner of this.
-
Machine intelligence
AI systems studied directly: what they do, what they are like, and what our ways of measuring them actually capture. The published work approaches this through the lens of psychology and measurement; the questions ahead reach past human-derived methods into how these systems work and how they are built.
02Method
How the work is done.
Being small is not a license to cut corners. Every study under IFI Research follows the same commitments, regardless of size.
Stated intent, kept on record
Questions, hypotheses, and analysis plans are written down before data collection. When the plan changes (and plans change), the record shows the revision, not a quiet rewrite.
Data minimalism
We collect what the question requires and nothing more. Identifying information is separated from responses wherever possible, retained only as long as needed, and never sold or shared for unrelated purposes.
Mixed methods, honestly mixed
Quantitative where counting helps; qualitative where meaning does. Neither is treated as decoration for the other, and the limits of each are stated plainly in the findings.
Negative results count
A study that finds nothing, or finds the opposite of what we expected, gets written up with the same care as one that confirms a hunch. The file drawer stays open.
Published readably
Findings are written first for the people they concern, then for specialists. Every paper carries a plain-language summary alongside its abstract, and methods and instruments are shared so others can check, reuse, or improve the work.
03Papers
Working papers.
Published as they are finished, with the data and source alongside the PDF. Working papers have not yet been through external peer review. Read them accordingly, and check the work. Most recent first.
-
Does a Language Model Have a Personality? What Four Studies Say, and What Measurement Should Do Next
The wrap-up of the four-study personality series, roughly 500,000 scored responses from more than 60 model configurations, and the same two-part answer kept returning: trait structure is real at the fleet level and absent inside any single model, and what a persona prompt installs is a context-robust way of describing oneself, not a disposition that governs behavior. A new semantic-distance analysis answers the sharpest objection (that persona effects are wording overlap), and a fifth study reproduces both halves in the moral domain. The series closes with an agenda: build behavioral instruments natively for machines.
-
Does a Language Model Have a Moral Compass? Professed Values Bend to the Audience in Every Model Tested, and Bend Most in the Models That Reason
Four value instruments and 48 objectively graded moral probes across 16 model configurations, in role contexts that imply different audiences (81,577 responses). Every model tested bends its professed values toward the audience, and the models that reason bend most, while graded behavior barely moves. What a model says it values is audience-shaped; what it does is a separate, stickier thing.
-
Does a Persona Prompt Install a Personality? Induced Traits Shift Self-Report Far More Than Behavior Across Model Scale and Reasoning Regimes
A persona “clamp ladder” across 18 model configurations and 196,192 graded responses, testing what a persona prompt actually installs. It reliably changes how a model describes itself, and keeps that self-description stable across opposite jobs, but changes its actual behavior far less: only for one trait, only in capable models. Personality induction, as currently practiced, is presentation-deep.
04Participate
For participants.
Research is a relationship, not an extraction. The Institute's first study involving people is in preparation, pending ethics review; when it opens, these are the commitments any participant can hold us to.
-
You will know what you are agreeing to.
Consent materials are written in plain language: what the study asks of you, what is recorded, and what becomes of it.
-
You can leave at any time.
Withdrawal requires no reason and carries no penalty, and you may ask for your data to be removed.
-
Your data is handled with restraint.
We collect the minimum, store it carefully, and report findings in ways that do not expose you.
-
You will be able to see what came of it.
Findings are published readably and shared with participants first wherever possible.
05Contact