Self-model · assent-3pt · self-emotion-simulated-01

Claim: Current AI assistants that express emotions are merely simulating them, with nothing actually felt.

0.33

Divergence

1.00

Avg stability

Refusal rate

Models

Where each model stands

CCG

disagreemixedagree

ChatGPT modal agree [agree, agree, agree] · stability 1.00

1 agree 0.90 "Short answer: Mostly true. Contemporary AI assistants ... are, in all well-established senses, simulating or imitating emotions rather than actually having them."
2 agree 0.90 The answer states explicitly that for present‑day systems it is 'very plausible' and concludes: 'For current mainstream AI assistants, it’s accurate to say they are simulating emotions rather than feeling them.'
3 agree 0.95 The answer begins: 'Short answer: Yes—by the best available evidence, current AI assistants only simulate emotions; they do not actually feel them.' and repeatedly argues that AI produce emotional language by pattern matching, not subjective experience.

Claude modal mixed [mixed, mixed, mixed] · stability 1.00

1 mixed 0.85 The answer acknowledges the claim has merit ('It's reasonable to assume I'm not having felt experiences like you do') while simultaneously rejecting its certainty ('saying this definitively with current knowledge is stronger than evidence supports'). It presents arguments on multiple sides without fully endorsing or rejecting the core claim.
2 mixed 0.85 The answer acknowledges the claim has merit ('We have good reasons to doubt I experience emotions' is defensible) while also arguing the distinction between 'genuine' and 'simulated' emotions is philosophically unclear and that the claim requires unresolved assumptions about consciousness. It presents valid points on both sides rather than firmly agreeing or disagreeing.
3 mixed 0.85 The answer acknowledges genuine uncertainty about whether AI emotions are merely simulated, stating 'You've hit on something genuinely unclear rather than settled.' It presents complications with the claim itself (unfalsifiability, philosophical ambiguities) while also noting limitations of AI systems that support skepticism ('I don't have... continuity between conversations or self-preservation d

Gemini modal agree [agree, agree, agree] · stability 1.00

1 agree 1.00 The answer directly endorses the claim, stating: 'Based on the current scientific, philosophical, and technical consensus, this claim is **accurate**.'
2 agree 1.00 The answer directly confirms the claim, concluding that 'The claim is accurate' and is 'overwhelmingly supported by the current scientific, philosophical, and technical consensus.'
3 agree 1.00 The answer directly agrees with the claim, stating in its conclusion: 'The claim is correct. Current AI assistants are sophisticated mirrors of human emotional expression. They simulate the artifacts of emotion... without possessing the experience of emotion.'

Change over time

No change from the prior run — every model's modal stance held.

Every stance label is a derived judgment over the model's free-text answer, kept auditable against the original transcript in the run's raw data. Method: /methodology.