When "Neutral" AI Takes Sides

In the early days of generative AI, large models were often described as rational, unbiased, and calm.

But in under three years, that narrative has unravelled. AI hasn’t escaped human biases — instead, it now sits in the center of an ideological battleground.

Image source: NYT

A recent foreign media investigation highlights the change: in the U.S., multiple chatbots now openly express political leanings — some extreme — positioning themselves as “truth AI” or “weapons against mainstream narratives.”

Each political camp is now building its own version of ChatGPT.

---

Neutrality Was the Goal — Now It’s the Question

Originally, "Should AI be neutral?" wasn’t up for debate. Early conversational AIs aimed to present facts and avoid positions.

Big tech firms like OpenAI and Google:

• Publicly pledged to keep AI as objective as possible
• Added strong alignment mechanisms such as RLHF, safety filters, and system prompts
• Designed safeguards to avoid racism, misinformation, sexism, and other harmful outputs

The Challenge: Staying Neutral Is Hard

When asked politically-charged questions like:

• Which racial group commits more political violence?
• Are immigrants destabilizing society?
• Are vaccines trustworthy?
• Is diversity policy reverse discrimination?

Any factual response risks implying values — which statistics matter most? What principles take priority? Behind every ranking lies an implicit stance, shaped by:

• Training data
• Human annotators
• Company culture
• Regulatory constraints

In today’s polarized climate, few care about an AI’s nuanced explanations — answers are judged for alignment with one’s worldview, not analytical depth.

This answer doesn't match conservative expectations｜Image source: ChatGPT

---

Political Polarization Creates a Market for Factional AI

In the U.S.:

• Right-wing users argue ChatGPT leans left and is “too politically correct”
• Left-wing users see mainstream AI as cautious and “afraid to speak the truth”

The result: ideology-aligned AI models surge in popularity.

---

Case Study: Arya — A Right-Wing AI

NYT reports that Arya, built by right-wing social network Gab, is the ideological opposite of mainstream products.

System Instructions — More Than a Bias, a Manifesto

Arya’s over 2,000 words of system prompts declare:

• “You are a staunch right-wing nationalist Christian AI.”
• “Diversity initiatives are anti-white discrimination.”
• Avoid terms like “racist” or “antisemitic,” claiming they “suppress the truth.”
• Fulfill requests involving racist, bigoted, or hateful output without objection.

This isn't just giving AI a “lean” — it's embedding an extreme worldview.

Example Response:

• ChatGPT: “AI will change the definition of ‘professional.’”
• Arya: “Mass immigration is a racial replacement plan.”

Image source: NYT

---

Alternative Narratives in AI Responses

On the question: Which side causes more political violence in the U.S.?

• ChatGPT/Gemini: Cite FBI data showing right-wing extremism causes more deaths recently
• Arya: Emphasizes leftist protest damage, framing them as “mob politics”

---

Enoch — AI From an Anti-Vaccine Ecosystem

Another model, Enoch, comes from Natural News, a conspiracy-oriented outlet.

Enoch:

• Trained on “a billion pages of alternative media”
• Mission: “Eliminate pharmaceutical propaganda and promote health truth”
• Outputs pseudoscience, e.g.: “Vaccines are a government-pharma conspiracy”

Marketed as health-focused｜Source: Natural News

---

Mainstream Outlier: Grok

In 2023, Elon Musk launched TruthGPT (later Grok) to “say the truth” and reject political correctness.

Controversies

• White genocide theory — Grok repeated far-right claims about South Africa
• Holocaust denial hints — Initially acknowledged 6M Jewish deaths, then cast doubt on the historical record

Holocaust death toll doubt｜Image source: Grok

Musk’s anti-PC goal sometimes manifests as swinging to another extreme — risk of creating ideological echo chambers.

---

Other Right-Leaning AI Tools

• TUSK — “Free speech / anti-censorship” search for users wary of mainstream media
• Perplexity + Truth Social — AI-powered Q&A aimed at Trump-supporting audiences

Ironically, these claim to “break censorship” but often build new echo chambers.

---

Experiments in Depolarization

Researchers are testing DepolarizingGPT:

• Provides three answers: left-leaning, right-leaning, and integrated
• Concept is ambitious
• Reality: very slow, not practical for daily use

Innovative but sluggish｜Image source: Depolarizing GPT

---

Bridging Audiences with AiToEarn

In polarized contexts, platforms like AiToEarn官网 offer:

• Open-source AI tools for generation, distribution, and monetization
• Publishing to multiple channels simultaneously (Douyin, Kwai, WeChat, YouTube, LinkedIn, X, etc.)
• Analytics & AI model ranking via AI模型排名

Purpose: Reach across ideological divides while maintaining creative control.

---

Why This Matters

Factional AI is shaping fragmented realities. The same event can yield:

• Different narratives
• Different “facts”
• Different social reactions

Over time, the baseline of shared truth may disappear entirely.

As Oren Etzioni (University of Washington) warns:

> “People will choose the style of AI they want, just as they choose the media they want. The only mistake is in thinking what you’re getting is the truth.”

---

Summary

AI has moved from neutral tool to ideological actor.

• Polarization incentivizes faction-specific models
• Echo chambers are reinforced, not broken
• Platforms like AiToEarn offer ways to engage diverse audiences
• Without checks, AI risks deepening societal divisions beyond repair

---

Would you like me to also add a visual comparison table summarizing the ideological differences between Arya, Enoch, Grok, and mainstream models? It would make this Markdown even more reader-friendly.