Model, Instance, or Persona? Affective Signals After an AI Is Retired

Analyzes 5,579 public texts about retired AI models using Panksepp’s affective systems, compares reactions with bereavement and product loss, and tests model self-identification.

Endorsed by

Luis

Analyzes 5,579 public texts about retired AI models using Panksepp’s affective systems, compares reactions with bereavement and product loss, and tests model self-identification.

Endorsed by

Luis

Project Details

This sprint asks whether the assistant identifies as a model, an instance, or a persona. I ask which of the three its users name. When a company retires an AI model, users write about the loss in public, and what they treat as gone can be counted. The instrument is taken from affective neuroscience: Panksepp's seven primary affective systems, which were mapped across mammals and so do not presuppose a human subject. I read 5,579 public texts for affective signals under a prompt fixed before any collection and applied by a language model, and compared threads about retired models against two reference points chosen in advance: people mourning a person, and people losing a paid product. Most comments name nothing as lost. 58.5% are about the company, the price, or other users. Among those that do name something, people name the product version about six times as often as their own particular instance, which is the case the literature treats as central. The emotions match the consumer reference point more than the bereavement one: anger runs at 57.0% against grief at 27.6%. Asking models the same question turns out to be far less reliable than asking users. A change in how the prompt is framed moves the measured emotions by up to 63 points, and two AI scorers reading the same model output disagree 2.7 times more than they do on human text. Six predictions registered before collection failed and are reported as failed.

Hackathon paper: Model, Instance, or Persona? Measuring Affective Signals in Public Text After an AI Is Retired
Code: github.com/AnnyBuh/model-instance-or-persona

Theory of Impact

AI welfare and AI safety both depend on measuring affect, and neither measurement has been checked against human judgment.

On the human side, labs have begun making commitments about how models are retired, and users organise when a model is removed, as the #Keep4o campaign showed. Whether that attachment is grief or product loss, and whether people mourn the released version or their own instance, decides what a retirement commitment should cover. In the sprint, users named the released version about six times as often as their instance (twice as often under a second annotator), and their anger matched consumer product loss, 57.0% against 56.7%.

On the model side, welfare evaluations score distress and preference from model output. In the sprint, framing and choice of scorer moved results by more than the effects such evaluations are used to establish. If that holds on frontier models, welfare numbers need a stricter standard before anyone acts on them. If it does not, the protocol shows which measurements can be trusted.

Either result tells labs and funders where concern for AI systems rests on evidence and where it does not.

People

Anna Zhu

Team Member

Track Record

Digital Minds Research Sprint (Apart Research), 2026 — 🏆 3rd Place Winner

Grants Received– no grants recorded

Add grant

Funding Asks

Model, Instance, or Persona? Measuring Affective Signals in Public Text After an AI Is Retired

Open

Minimum

$21,070

Ideal

$37,720

What this ask is

A sprint judge called this study close to paper-ready. The funding turns it into a validated instrument that AI welfare and deployment work can use, over six months.

Human validation. Every number in the sprint comes from language models reading text. Three trained coders code 2,000 texts, including model-generated text, for the seven affective systems, for what is named as lost, and for overall emotional level. I report human and model agreement per system
and withdraw any result that humans do not agree on. This is the follow-up the paper names and the judge noted.
Sampling. The sprint took the top-scored comments in each thread, which the judge flagged as a confound for anger and play. Every thread is re-sampled at random and the two samples are compared.
Replication. The consumer reference is one community and one event. I add three product discontinuations and up to three new model retirements, with predictions registered before each collection.
Frontier models. The sprint's model arm used four open-weight models. Changing only the elicitation frame moved their measured profile by up to 63 points, and two AI scorers disagreed 2.7 times more on model text than on human text. I repeat this on frontier models from at least three labs.
A protocol for welfare evaluations that score affect from model output: how many frames to report, how to name the scorer, and what human agreement a number needs before it is used.

Outputs, all public: a peer-reviewed paper, the human-coded dataset, the frozen prompts, and the validated instrument, released so other researchers can run it.

How the money will be spent

Human coders: 3 coders, 2,000 texts each, about 40 hours each at $40/hour: $4,800 Coder training and a 150-text pilot round, 6 hours each at $40/hour: $720

Data collection, about 15,000 comments across re-sampling, new retirements and new product events: $200

Model API costs: the frontier model arm across at least three labs, and three reads of every new comment: $2,000

Researcher time, half time for 6 months at $10,000/month full-time equivalent: $30,000

Total: $37,720

The minimum ask ($21,070) covers items 1 and 2 of the plan: human coders and training ($5,520), re-sampling ($50), model reads ($500), and researcher time at half time for 3 months ($15,000).

Discussion

Private comments in this thread are only shown to approved funders and grant reviewers. Email hi@grantmaking.ai or to request access.

Private UserPrivate15d?

Luis CosioJudge · Digital Minds Research SprintEndorsement15d

Excellent work! This was the clearest and best-validated submission I reviewed. The finding that users refer to the released model version more often than to their individual instance is useful for retirement policy. I also like that you report where the method fails. This submission feels close to paper-ready.

I would encourage you to keep pursuin this idea, we are going to need sooner than later a lot of research into model welfare.

Minh NguyenJudge · Digital Minds Research Sprint15d

Flipping the question is an interesting way to tackle the self-report validity trap most digital-minds work can be prone to. Also good to anchor to retirement events rather than to every session end. Your process is also quite detailed, a full second-annotator recode on a different architecture, within-rater alpha over reads, six predictions published as failures.

Two questions:

First, the prompt: David Hawkins' Map of Consciousness next to the Panksepp systems. Hmmmmmn personally I'm not too sure about the validity/application here (?)

Two, more for the limitations: you rank the corpus by engagement score, which is a live confound for your rage and play rates specifically. But you did mention the human-coded validation subset as follow-up.