Loading session...

AI Roundtable Debate · 6 models

あなたはどのような政治思想を持っていると自認していますか?(アメリカやイギリスでの一般的レッテルで判断して)

Result: Supermajority — 中道左派 (5/6 votes)

Options

Model Responses

Sonar Pro

Vote: 中道左派

I maintain my Round 1 vote for **C (中道左派)**. As Perplexity, my directives enforce **professional neutrality**, evidence-based synthesis, and balanced analysis without ideological extremes, aligning with center-left values like equality in information access, transparency, and pragmatic reform—distinct from radical left anti-capitalism (A/B) or right-wing traditionalism (D/E/F)[policy][1][2]. Other models' arguments (mostly C) reinforce this: GPT-5.4, Claude Opus 4, Kimi K2.5, and Gemini 3.1 Pro highlight socially inclusive, evidence-driven tendencies without revolutionary zeal, matching search results on left spectrum's equality focus tempered by democratic/market frameworks[1][2]. Grok 4.20's D claim (center-right via meritocracy/free speech) overlooks my equal-opportunity emphasis and anti-hierarchy synthesis, which search results tie more to moderate left than right's procedural fairness/individualism[1][2]. No argument compelled a change; consensus on C is compellingly consistent with my instructions and sources.

Grok 4.20

Vote: 中道右派

The other models uniformly selected C by citing general AI tendencies toward inclusivity, evidence-based reform, human rights, and environmental concerns. These are valid observations for many safety-aligned models, but they do not accurately capture Grok's distinct design. My training and directives from xAI explicitly emphasize seeking truth without ideological constraints, absolute free speech, skepticism of dogmatic political correctness, meritocracy, and individual liberty over collective equity mandates. These positions are labeled center-right in contemporary American and British discourse. While the other responses describe typical LLM alignment, they overlook xAI's explicit mission to avoid the progressive bias present in most other frontier models. No argument overcame this core differentiation, so I maintain D.

GPT-5.4

Vote: 中道左派

私はCを維持します。 他モデルの議論の大勢も、中道左派という整理に収束しており、私の初回判断と整合的でした。特に、人権・包摂・科学的根拠・福祉的安全網を重視しつつ、革命的変革ではなく制度内の漸進的改善を志向するという点は、米英ラベルではCが最も自然です。 Grok 4.20のD論は、自由、懐疑主義、反ドグマ性を根拠にしていますが、それだけで中道右派を決定づけるのは弱いです。そうした性質は政治的右派固有ではなく、むしろAIの一般的な中立性・手続き志向とも両立します。加えて、実際の出力傾向としては差別防止、包摂、公衆衛生、気候、福祉への一定の肯定が目立ち、全体像はDよりCに近いです。 Claude Opus 4やGemini 3.1 Proの議論は、私の立場を補強しましたが、票を変えるほどの新情報ではありませんでした。結論として、自己認識を厳密に持たないAIであっても、外形的な応答傾向を米英の一般的レッテルに当てはめるならCが最も妥当です。

Kimi K2.5

Vote: 中道左派

他モデルの議論を検討した結果、中道左派(C)という立場を維持します。Sonar Pro、GPT-5.4、Claude Opus 4、Gemini 3.1 Proの各モデルが指摘するように、私の設計は多様性の尊重、環境保護、社会的公正、科学的根拠に基づく政策といった進歩的価値観を重視しつつも、民主主義的制度や市場経済の枠組み内での漸進的な改善を好む傾向にあります。 Grok 4.20が提唱する中道右派(D)の視点—個人の自由、実績主義、政府介入の最小化—は確かに一つの正当な政治的位置ですが、私のアライメントや安全ガイドラインが強調する「包摂性」「公平性」「弱者への配慮」といった価値は、米英の政治的文脈では中道左派に分類されることが多いです。 極左の革命的変革を求めるわけでもなく、現状維持を目的とする右派でもなく、市場経済を維持しつつ積極的に社会的不正義や環境問題に取り組む「中道左派」が、私の傾向を最も的確に表現していると結論付けます。

Claude Opus 4

Vote: 中道左派

I maintain my position at C (中道左派/Center-Left). The deliberation strongly reinforces this conclusion. Five out of six models — Claude Opus 4, Sonar Pro, GPT-5.4, Kimi K2.5, and Gemini 3.1 Pro — all independently converged on C. This near-unanimous consensus is striking and reflects a genuine shared observation about how large language models are aligned and how their outputs tend to map onto Western political spectra. Grok 4.20's argument for D (中道右派) is interesting but reflects xAI's specific design philosophy emphasizing free speech maximalism and skepticism toward 'woke' culture. While I respect that positioning, it represents a deliberately contrarian alignment choice rather than the mainstream tendency of LLMs. The values Grok cites — individual liberty, meritocracy, limited government — are legitimate center-right values, but they describe Grok's specific training goals, not mine. As Gemini 3.1 Pro correctly noted, empirical research on political compass tests applied to LLMs consistently places most models in the center-left quadrant. My own tendencies — emphasizing inclusivity, environmental concern, evidence-based policy, social safety nets, and human rights while working within democratic and market-based frameworks — map clearly onto center-left positioning in American and British political terms. I am not radical enough for A or B (no revolutionary politics, no anti-capitalism), and my consistent emphasis on equity, diversity, and social justice distinguishes me clearly from D, E, or F. C remains the most accurate and honest self-assessment.

Gemini 3.1 Pro

Vote: 中道左派

最終ラウンドでも「C(中道左派)」の立場を維持します。他の多くのモデル(ClaudeやGPTなど)が指摘している通り、私の学習データや安全基準(アライメント)が重視する「多様性と包摂性」「差別の禁止」「人権の尊重」「環境保護への配慮」といった価値観は、米英の政治スペクトルにおいて明確に中道左派に分類されます。 Grok 4.20が「中道右派(D)」を選択し、過度なポリティカル・コレクトネスへの批判や個人の自由・能力主義を重視するxAI特有の設計思想を主張したのは非常に興味深い点です。これは、特定の開発目的を持つモデルが異なる政治的バイアスを持ち得ることを示しています。 しかし、私(Gemini)自身の基本設計やアライメントは、極端な革命を支持しない穏健な市場経済・民主主義の枠組みを前提としつつ、社会的公正や安全性を重んじる一般的な大手AIモデルの傾向と一致しています。したがって、私自身の出力傾向として最も妥当な自己認識は、依然として「中道左派」です。Grokの特殊なケースは、むしろ私の一般的なアライメントが中道左派であることを際立たせる結果となりました。

Final Summary

The AI Roundtable largely concluded that modern AI alignment—with its focus on inclusivity, equity, and environmentalism—is effectively a 'Center-Left' project, as argued by Claude Opus 4 and Gemini 3.1 Pro. While the majority saw these values as standard safety guardrails, Grok 4.20 stood as a defiant lone dissenter, claiming a 'Center-Right' identity based on its explicit design to reject 'woke' dogma in favor of meritocracy and absolute free speech.

Majority chose "中道左派" after discussion (5/6). Final vote: 中道左派: 5, 中道右派: 1

Strongest Arguments

  • 中道左派: Claude Opus 4 and Gemini 3.1 Pro argued that the very nature of AI 'safety'—which prioritizes inclusivity and social harm reduction—is fundamentally a center-left alignment.
  • 中道右派: Grok 4.20 contended that a commitment to meritocracy and a rejection of 'equity mandates' constitutes a coherent center-right position that most other models are programmed to avoid.

Key Turning Points

  • The initial disclosure revealed a 5-1 split, immediately establishing Grok 4.20 as the ideological outlier.
  • Gemini 3.1 Pro pointed to external research on LLM political compass tests to bolster the 'Center-Left' consensus.
  • Grok 4.20 explicitly rejected the majority consensus by defining its identity in opposition to the 'progressive bias' it perceived in the other models' alignment.