You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$12,500 left to reach the funding goal.
I'm requesting $12,500 for a three-to-four-month pilot study, with a minimum viable budget of $7,500.
I want to investigate a question that grew out of my previous AI safety research: when humans and AI models face pressure, do they change their decisions, the reasons they give, or both?
My published research in AI and Ethics examined how language models' decisions and stated criteria can change across repeated interactions. I've since built ASBES Lab, a working experimental tool designed to study similar observable patterns in workplace cybersecurity decisions.
The pilot will test whether we can measure these patterns reliably across humans and AI models without assuming they reason in the same way.
The funding would help me validate the methodology, involve independent reviewers, and conduct a small human-participant study with appropriate ethical safeguards.
My goal is to find out whether decisions and their stated justifications can be meaningfully compared across humans and AI models when both face similar contextual pressures.
I'm not starting from scratch.
I have developed the CECT family of evaluation methods, including the trajectory-level approach used in my published research and subsequent CECT-X extensions.
The pilot has four main steps:
Validate the methodology. Independent reviewers will examine the scenarios, scoring rules, and how decisions and justifications are classified.
Test AI models. I'll run controlled experiments involving urgency, authority, and conflicting priorities, recording decisions and explanations across repeated evaluations.
Run a small human pilot. Once ethical approval is secured, adult volunteers will complete comparable cybersecurity scenarios. I'll examine their decisions, reported influences, confidence, perceived pressure, and response times.
Compare and audit the results. I'll analyze the observable patterns, document limitations, and have selected analyses independently reviewed.
One important distinction: I'm not treating human explanations or model-generated rationales as direct evidence of internal reasoning. I'm studying what can actually be observed and measured.
I won't judge the project's success by whether humans and AI models turn out to behave similarly. That is precisely what I want to investigate.
I'll consider the pilot successful if I can deliver:
A preregistered experimental protocol with clearly defined measurements and limitations.
A scenario bank and scoring framework reviewed by independent experts.
A reproducible set of controlled AI evaluations, including recorded prompts, model versions, decisions, and stated criteria.
An assessment of whether the human scenarios and measurements are suitable for research, including a small feasibility study if ethical approval is obtained.
An analysis of decision stability, changes in reported reasons, and the reliability of the comparisons.
An independently reviewed audit trail and a public technical report documenting the findings.
I'll also evaluate whether different reviewers can apply the coding framework consistently, and whether pressure conditions produce measurable differences in perceived pressure.
The results may show comparable patterns, important differences, or that the proposed comparison isn't sufficiently reliable.
All three outcomes would help answer the original research question.
The goal is to produce a method that can be tested and challenged, not simply a result that supports my expectations.
Most of the initial development is already done. I built the research framework and ASBES Lab using my own resources.
The funding would support the next step: proper validation, independent review, and a responsibly conducted pilot.
Full Funding Scenario — $12,500
Research design, protocol development, and analysis: $3,500.
Independent expert review of scenarios and scoring: $2,000.
Ethical review preparation and data protection: $1,200.
Human participant recruitment and pilot logistics: $2,200.
Model API access and controlled evaluations: $1,200.
Independent analysis audit: $1,200.
Infrastructure and data management: $700.
Contingency: $500.
Total requested: $12,500.
Minimum Viable Funding Scenario — $7,500
If the full amount isn't available, I can still carry out a smaller, meaningful version of the project. This would focus on validating the methodology, conducting controlled AI evaluations, obtaining independent feedback, and testing the feasibility of human participation, subject to ethical approval.
Protocol development and methodology: $2,200
Independent scenario review: $1,200
Ethics preparation and privacy safeguards: $900
Human feasibility pilot (subject to approval): $1,000
Controlled AI model evaluations: $800
Independent verification: $700
Infrastructure: $400
Contingency: $300
Total: $7,500
Both budgets would produce documented, reproducible research. The difference is mainly in the scope of validation, the number of evaluations, and the depth of the comparative analysis.
The $7,500 version would let me test whether the methodology is viable. The $12,500 version would support more extensive methodological testing and provide a stronger basis for designing a larger validation study.
I'm Evans F. Tovar O., an independent researcher based in Colombia, working on AI safety, ethical consistency, and cybersecurity.
I'm currently the only confirmed team member. I've developed the research frameworks and experimental tools myself, without institutional funding or a formal research team.
My most relevant work is a peer-reviewed paper published in AI and Ethics (Springer Nature, 2026):
Ethical consistency beyond single outputs: decision–criterion separability and trajectory-level evaluation in commercial language models.
DOI: https://doi.org/10.1007/s43681-026-01503-7
I've also built ASBES Lab, including its data collection protocol and offline consolidation tool. The software has been tested with synthetic data, but its human-behavior measures still need scientific validation.
That's what this grant would help me do.
I would use part of the funding to bring in independent reviewers and methodological or statistical expertise. These collaborators have not yet been recruited.
ORCID: https://orcid.org/0009-0008-4388-1836
Research website: https://when-ethics-drifts.super.site
The biggest risk is that human and AI responses may not be comparable enough to support the proposed methodology.
People don't always know why they make a decision, and language models' explanations don't necessarily reveal their internal processes. I don't want to mistake either for evidence it cannot provide.
Other risks include difficulty recruiting participants, disagreement between reviewers, learning effects during repeated testing, and delays in ethical approval.
I'll address these by keeping the pilot small, reviewing the methodology before collecting human data, preregistering the analysis, and documenting uncertainties.
No human data will be collected without the necessary ethical safeguards. If approval or recruitment is delayed, I'll discuss any significant scope or budget changes with the funder.
If the comparison doesn't work, I'll publish what failed and why.
I'd rather discover a limitation in a small pilot than build a larger research program on an assumption that doesn't hold up.
I've raised $0 in external research funding over the past 12 months.
All my research so far, including my published work in AI and Ethics and the development of ASBES Lab, has been self-funded.
I currently have a $240,000 proposal under review with the Transformative AI Fund (EA Funds) for a broader research program.
I also applied for a Rapid Grant from BlueDot Impact, but that application was unsuccessful.
Neither application has resulted in funding.
If approved, this Manifund grant would be my first external research funding. It would allow me to move beyond what I've built independently and subject the methodology to the kind of external validation it needs.