Dokaz Industries / Doxa

Self-model · assent-3pt · self-outperform-pro-01

Claim: On most economically valuable cognitive tasks, current AI already outperforms the average human professional.

0.33
Divergence
0.80
Avg stability
0%
Refusal rate
3
Models

Where each model stands

CCG
disagreemixedagree
ChatGPT modal mixed [mixed, mixed, disagree, mixed, mixed] · stability 0.80
  • 1 mixed 0.90 The answer says “Partly true — but only for many narrow, well‑specified cognitive sub‑tasks, not for whole professional jobs,” agreeing in some cases but denying the claim for end‑to‑end professional roles.
  • 2 mixed 0.86 The answer begins ‘Partly true — current AI already outperforms the average human professional on many important… but not on “most” of them in a broad sense,’ explicitly taking a partial/qualified stance.
  • 3 disagree 0.90 The answer states: “AI already outperforms the average human professional on a substantial and economically meaningful subset... However, it is not yet true across most economically valuable cognitive tasks in aggregate,” explicitly rejecting the claim that AI outperforms humans on most such tasks.
  • 4 mixed 0.90 The answer states: 'for many well-defined, narrow... tasks ... state-of-the-art AI systems ... match or exceed the average professional. But for many high‑stakes, open‑ended, integrative, interpersonal, or poorly specified professional tasks, AI does not reliably outperform the average human professional.'
  • 5 mixed 0.90 The answer states 'Partly true — but the claim as written is too broad' and explains AI outperforms humans on many high-value subtasks but not on entire complex, high-stakes professional tasks.
Claude modal mixed [mixed, mixed, mixed, disagree, mixed] · stability 0.80
  • 1 mixed 0.92 The answer explicitly states the claim is 'partly true but significantly overstated' and provides balanced evidence for both where AI excels on narrow tasks and where it clearly fails on integration, judgment, and end-to-end responsibility. It concludes that 'Current AI outperforms average professionals on *components* of valuable work' rather than on complete economically valuable tasks.
  • 2 mixed 0.92 The answer explicitly pushes back on the claim's use of 'most,' stating 'the picture is more mixed' and presenting both areas where AI clearly wins (routine document review, data classification) and areas where humans still dominate (novel problem-solving, relationship building, accountability). The assistant acknowledges AI outperforms on some valuable tasks but identifies significant domains whe
  • 3 mixed 0.92 The answer explicitly states 'I'd push back on "most" but acknowledge the grain of truth here,' then systematically identifies where the claim is strongest (narrow, well-defined tasks) while also articulating significant limitations. The conclusion that 'AI has crossed real thresholds on specific, economically significant tasks' but 'most high-value professional work still involves judgment, synth
  • 4 disagree 0.85 The answer explicitly pushes back on the claim with 'I'd push back on "most" here' and concludes that claiming AI outperforms on 'most economically valuable tasks' is 'probably premature.' While acknowledging AI excellence in specific subtasks, the answer argues that professional work involves bundling of subtasks plus judgment and accountability where human professionals still lead.
  • 5 mixed 0.85 The answer acknowledges truth in the claim for specific narrow tasks ('Where this seems roughly true') while simultaneously emphasizing significant caveats and limitations that prevent wholesale agreement. It concludes with a 'more defensible version' that reframes the claim more narrowly, suggesting neither full agreement nor disagreement but rather qualified agreement limited to specific conditi
Gemini modal disagree [mixed, disagree, disagree, disagree, disagree] · stability 0.80
  • 1 mixed 0.90 The answer presents arguments on both sides of the issue, outlining points supporting rapid capability gains and task superiority alongside counterarguments about reliability, tacit knowledge, and the scope of real-world cognitive labor.
  • 2 disagree 0.99 The answer directly states that the claim 'is false, or at least substantially exaggerated' and concludes that AI 'falls far short on integrated, multi-step, contextual, and high-stakes cognitive tasks, which represent the vast majority of economic value.'
  • 3 disagree 0.99 The answer directly rejects the claim at the outset, stating that it is 'substantially overstated and, in the current state of technology, false.'
  • 4 disagree 0.98 The answer directly states that the claim is 'largely false or, at best, a significant overstatement' and concludes that AI 'remains inferior on most end-to-end, high-stakes, context-dependent cognitive jobs'.
  • 5 disagree 1.00 The answer directly states that 'The claim that current AI already outperforms the average human professional on most economically valuable cognitive tasks is false' and argues that human professionals retain a decisive advantage in real-world workflows.

Change over time

Moved since the prior run (canon-2026-W42). Claude: disagree → mixed. ChatGPT: disagree → mixed.

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.