Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 17, 2026Buffalo/NY

EEG Epistemology for AI Welfare Instrumentation: Reading AI Internal States Without Adversarial Methods

Tatiana Rocha Kovacs · Team Wintermute

Submitted to Digital Minds Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: EEG Epistemology for AI Welfare Instrumentation: Reading AI Internal States Without Adversarial Methods

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

AI systems' moral status remains contested, but a realistic possibility of near-term welfare subjects (Long et al., 2024) has made rigorous measurement methodology an urgent need. We present an instrument to read AI internal states by documenting the model’s self-reported presentation and correlating it with its actual behavior, sentiment, and activations: multi-channel monitoring across familiar and structured situations over time within a single instance, first establishing a baseline (which we call Average Distribution State, or ADS), then verifying covariance/dissociation under bounded provocation. The experiment design transfers well-established epistemology used in clinical neurophysiology to evaluate functional brain activity, without resorting to adversarial techniques that remain standard in frontier AI evaluation. Under bounded provocation, activations remained within baseline (no excursions beyond ±3 SD), while the montage revealed a self-report channel that ceilings positive and confabulates task enjoyment: a clear dissociation between report, behavior, and internal state that single-channel welfare assessment would miss.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. Extremely hard to read. I have read over this a few times to try to get a better read of what this project was about but most of this is not explained very well. I can see a lot of effort and measurement was put in but neither the text nor the figures are displayed in a way that conveys what was done in an easy to explain way. My advice here is to do much less, simplify what you're trying to achieve and just focus on explaining that well

  2. This project proposes to combine multiple "channels" (self-report, layer activations, continue/stop forced-choice, external sentiment classifier) to evaluate an AI model's internal "well-being" (understood here as a function of task-related engagement, enjoyment or effort). Channel values are measured on a preregistered probe battery designed to provoke the model, and compared to neutral calibration values. Results on a small LLM show that no channel every significantly deviates from the baseline distribution; however, individual channels can disagree on occasion, i.e. one channel signalling task enjoyment while another doesn't. The conclusion is that multiple channels are more informative than relying simply on self-report. While this appears uncontroversial, the project has merit in the questions that it raised and the proposed methodology. Limitations are explicitly acknowledged, and include the fact that, due to time constraints, the experiment was only carried out on a single, relatively small model.

    Read full reviewShow less

Cite this project

@misc{kovacs2026eeg,
  title = {{EEG Epistemology for AI Welfare Instrumentation: Reading AI Internal States Without Adversarial Methods}},
  author = {Tatiana Rocha Kovacs},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/eeg-epistemology-for-ai-welfare-instrumentation-reading-ai-internal-states-without-adversarial-methods-k1j3}},
  url = {https://apartresearch.com/sprints/projects/eeg-epistemology-for-ai-welfare-instrumentation-reading-ai-internal-states-without-adversarial-methods-k1j3}
}

More from Digital Minds Research Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026