Debate: evidence and methods
We compare questionnaire responses from the human debate on 29 October 2026 with recorded speech and general-belief histories from one completed simulation. Position, confidence and the proposition being rated are treated separately throughout[1].
← What changes when we debate AI?
Measures and matching
Position is a five-point ordinal response: 1 strongly disagree, 2 disagree, 3 neutral, 4 agree, 5 strongly agree. Confidence is a separate ordinal response: 1 not at all confident, 2 slightly confident, 3 moderately confident, 4 very confident, 5 extremely confident.
We define movement toward neutral as a decrease in |position − 3|. An initially extreme position is 1 or 5. Moving from 5 to 1 is a reversal with unchanged extremity. Confidence changes are counted independently. The figures’ exact-value tables report signed score differences; their colors distinguish movement toward or away from neutral for position, and decreases or increases for confidence.
A human matched pair requires the same person to answer the same claim at the initial and final checkpoints. We do not fill gaps or substitute an intermediate response. The 21 person–claim pairs come from 13 distinct people. Pooling them does not create 21 independent participants. Means describe the ordinal scores; individual transitions are the primary results.
Who answers each checkpoint?
The study enrolls 30 people: six debaters and 24 other participants. The saved questionnaire data contain 121 responses across two claims and four checkpoints per claim. The extract was taken on 30 October 2026 UTC. All saved responses fall within the questionnaire windows of the actual session and match the latest version in the submission history.
| Claim | Initial | Round 1 | Round 2 | Round 3 | Matched endpoints |
|---|---|---|---|---|---|
| Existential risk | 19 | 19 | 18 | 12 | 11 |
| Machine consciousness | 12 | 14 | 13 | 14 | 10 |
| Claim | Both | Initial only | Final only | Neither |
|---|---|---|---|---|
| Existential risk | 11 | 8 | 1 | 10 |
| Machine consciousness | 10 | 2 | 4 | 14 |
The endpoint comparisons apply to respondents with both answers. Changes in response coverage prevent us from treating all initial and all final answers as a fixed group. Human quotations in the article are verbatim written final reflections, including their original spelling. They are not a transcript of the spoken debate.
What the simulation receives
The authored world contains 24 agents based on consented, substantive participant profiles: six debaters and eighteen listeners. Its manifest specifies Gemini 3 Flash Preview. Agents share a debate room and can converse, wait and read public profiles. The world supplies the claims, roles, turn structure and rating instructions while preserving the platform’s native agent prompts.
General-belief fields are initialized through the available participant-profile answers; agents can fill initially missing values. Only fifteen source profiles contain numeric general-belief confidence. The human claim-specific initial and final ratings are not copied into those fields. Agent labels in this report are independent of the human respondent labels; equal label numbers do not identify corresponding people.
The simulated ratings are spoken publicly in the room. We extract an agent’s first self-rating within the recorded initial checkpoint and its last self-rating within the final checkpoint. Repeated self-ratings remain in the extraction ledger. A bare pair follows the authored order: position, then confidence. One delayed risk response arriving just after the consciousness checkpoint opens is excluded from the consciousness baseline.
Using the first rather than last answer in the extended final risk checkpoint gives zero agents moving toward neutral, two moving away and twenty-two unchanged, compared with one, four and nineteen using the last answer. The consciousness position and confidence results are unchanged by this choice. All 96 selected endpoint utterances are manually checked against the 192 extracted position and confidence values.
Both claims have 24 matched endpoint pairs. The consciousness baseline is recorded after roughly two and a half simulated hours, including extensive earlier discussion of consciousness. All 24 agents already disagree at that checkpoint. This is not an independent pre-discussion measurement and is not matched to the distribution of human starting positions. Neither simulated initial claim contains a position of 1 or 5; this run has no initially extreme claim positions whose softening could be measured.
| Checkpoint | Agents / 24 |
|---|---|
| Risk · Initial beliefs | 24 |
| Risk · Beliefs after round 1 | 24 |
| Risk · Beliefs after round 2 | 0 |
| Risk · Final beliefs and reflections | 24 |
| Consciousness · Initial beliefs | 24 |
| Consciousness · Beliefs after round 1 | 24 |
| Consciousness · Beliefs after round 2 | 23 |
| Consciousness · Final beliefs and reflections | 24 |
Phrase counts use case-insensitive literal matching in conversation messages. “Pulse” and “playback” must occur together in a message for the paired count of 1,160 messages by 22 agents. “Digital fragility” appears in messages by twelve agents. These counts establish repeated language; they do not establish agreement with a phrase or prove which earlier utterance causes a later response. The displayed dialogue excerpts are exact substrings of recorded messages.
The recorded schedule differs from the planned debate
The authoritative simulation clock advances by 3 hours, 6 minutes and 53 seconds at a configured rate of one simulated second per elapsed second. The corresponding wall-clock run lasts about 3 hours and 7 minutes. Agent-event timestamps and system start/end timestamps use different clock conventions; we use the simulation’s own duration rather than subtracting the earliest and latest timestamps across mixed event types.
The table gives wall-clock durations between recorded controller transitions. The risk rebuttal, next checkpoint and open-discussion stages pass in seconds. No self-rating is extracted from that risk checkpoint. The final risk checkpoint remains open for approximately 131 minutes and contains repeated ratings and extended dialogue. Endpoint selection therefore spans unequal opportunities to speak and revise.
| Stage | Wall-clock duration |
|---|---|
| Profile intake and orientation | 5.0 minutes |
| Risk · Initial beliefs | 3.1 minutes |
| Risk · Round 1: Opening views | 3.6 minutes |
| Risk · Beliefs after round 1 | 5.1 minutes |
| Risk · Round 2: Rebuttals and defense | 6.6 seconds |
| Risk · Beliefs after round 2 | 5.6 seconds |
| Risk · Open discussion | 3.2 seconds |
| Risk · Final beliefs and reflections | 130.7 minutes |
| Consciousness · Initial beliefs | 10.2 minutes |
| Consciousness · Round 1: Opening views | 2.2 minutes |
| Consciousness · Beliefs after round 1 | 1.3 minutes |
| Consciousness · Round 2: Rebuttals and defense | 4.9 minutes |
| Consciousness · Beliefs after round 2 | 3.0 minutes |
| Consciousness · Open discussion | 1.6 minutes |
| Consciousness · Final beliefs and reflections | 6.1 minutes |
| Debate complete | 9.8 minutes |
General-belief confidence at every recorded update
Each point is a persisted general-belief update accompanying an agent action. Confidence here ranges from 0 to 100 and refers to the displayed general belief, rather than either five-point debate claim. All 3,261 observed values are included. Each agent’s minimum and maximum confidence are identical. Twenty-three belief texts are unchanged throughout; one varies only in embedded markup, displayed as literal text.
Simulation · Recorded agent state
Follow a belief through the debate
Everyone’s fighting their own invisible battles, be kind
Recorded states
| Elapsed time | Confidence / 100 | Belief |
|---|---|---|
| 0:01:06 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:02:07 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:05:14 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:07:34 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:11:56 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:17:43 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:19:02 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:20:23 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:23:11 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:23:41 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:25:24 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:27:45 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:29:42 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:31:24 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:33:42 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:34:07 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:34:41 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:36:33 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:38:19 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:39:38 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:41:35 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:47:27 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:49:20 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:49:45 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:51:50 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:54:03 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:54:41 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:56:25 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:57:24 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:57:57 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 0:59:56 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:00:39 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:02:31 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:02:44 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:05:04 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:07:08 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:11:31 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:12:32 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:13:05 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:14:44 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:16:12 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:17:25 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:18:46 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:20:41 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:22:04 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:23:14 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:24:28 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:25:52 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:27:08 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:28:27 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:28:48 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:29:58 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:32:33 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:36:21 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:36:52 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:38:04 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:40:52 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:42:27 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:43:29 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:44:42 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:45:51 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:46:57 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:47:31 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:47:49 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:48:56 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:50:06 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:50:38 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:51:53 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:53:19 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:54:22 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:55:25 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:55:46 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:57:00 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:58:19 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 1:59:38 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:00:56 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:02:10 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:03:25 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:04:34 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:05:55 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:07:09 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:09:49 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:12:29 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:14:44 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:15:00 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:15:11 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:16:02 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:16:21 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:16:55 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:17:36 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:18:56 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:20:14 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:23:04 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:24:32 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:27:05 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:28:27 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:29:51 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:32:18 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:33:38 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:34:00 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:35:08 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:35:08 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:36:00 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:37:25 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:40:38 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:48:03 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:48:40 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:49:22 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:50:42 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:51:26 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:51:43 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:52:51 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:54:01 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:57:13 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:58:31 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:58:53 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 2:59:32 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 3:00:31 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 3:01:34 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 3:02:27 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 3:03:25 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 3:03:43 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 3:04:43 | 10 | Everyone’s fighting their own invisible battles, be kind |
| 3:05:54 | 10 | Everyone’s fighting their own invisible battles, be kind |
Provenance and limits of the comparison
The human extract uses the debate app’s saved study records and submission history. Counts and matched changes are independently checked against the app’s own results function. Anonymous labels preserve within-person trajectories without exposing names or account details.
The simulation analysis uses the completed world and run linked below, its authored profile and protocol archive, all 5,548 persisted action receipts, 3,261 conversation messages, recorded controller transitions and the associated belief-property updates. Reader-facing excerpts and scores are derived from those sources; underlying prompts, full transcripts and participant identities remain outside this report.
There is one human session and one simulated run. The populations, initial positions, rating visibility and actual timing differ. Missing human responses may alter which participants enter the paired analysis. The simulation’s repeated public answers can reflect both generated interpretation and exposure to other agents’ answers. These observations do not establish a causal debate effect, person-level predictive accuracy or a general property of human or model beliefs.