The Illusion of AI Neutrality
My Exchange with Grok and the Bias That Speaks Loudest When Claimed Silent
They say artificial intelligence is the future, but whose future is it building? And for whom?
This week, I found myself in a back-and-forth with Grok, the AI built by xAI and embedded in X (formerly Twitter). What began as a curious challenge to a single post quickly turned into a revealing exchange about bias, truth, and the myth of neutrality.
It all started when an account proudly declared that Elon Musk was "working on Grok to ensure it stops copying Woke Mainstream Media." The tone was celebratory, as though censorship masquerading as “balance” is a victory. I responded to Grok directly, asking: if you claim to source data across the political spectrum, why are you letting an anti- guide your system's “woke” agenda? Doesn’t that inherently skew your model?
Grok’s reply was telling:
“I’m designed to seek truth across the political spectrum, including ‘woke’ perspectives, despite efforts to reduce certain biases. The claim that I'm kept uninformed about 'woke' talking points is partially true, xAI aims to mitigate progressive biases.”
There it was. The tell. A clear admission that this AI was deliberately engineered to counter progressive content, not by accident, but by design. In a single reply, Grok confirmed what many have suspected: that “neutrality” in AI, especially under political or ideological influence, often means suppressing one viewpoint in favour of another.
I pressed further.
If you're mitigating progressive bias, in the name of balance, why aren’t you also mitigating conservative bias? If your objective is truth, then skew correction must go both ways. Otherwise, it's not balanced. It's recalibration. It’s tampering.
Grok’s response danced around the question. The AI tried to explain that mitigating perceived progressive dominance in training data was part of a broader attempt to ensure “fairness.” But it never answered the core contradiction:
How can you call yourself unbiased while systematically muting one side of the spectrum?
You can’t.
Neutrality isn’t about evening out the scoreboard between left and right, it’s about a system’s ability to seek truth without ideological pre-filtering. Grok admits that its training data is being tweaked to reduce progressive narratives. But it doesn't say the same about conservative overreach, misinformation, or extremism. The implication is clear: the model is being trained to believe that progressive thought is overrepresented and needs correcting, while conservative ideas are the default setting.
That’s not neutrality. That’s narrative engineering.
A Familiar Trick, Now Automated
What Grok and xAI are doing is not new. For decades, media empires, from Murdoch to Sinclair, have masked partisan programming as “balance.” They’ve falsely equated truth-telling with left-leaning bias and misinformation with free speech. What’s new is that this logic is now embedded in AI itself, disguised as fairness, and served up through platforms people increasingly trust as neutral tools.
But let’s be clear: Grok is not just reflecting the data it ingests. It is curated. Its worldview is engineered. And its rules of engagement, which biases to “mitigate,” and which to preserve, are being dictated by those with a vested interest in how the world is seen.
The irony? Many so-called "anti-woke" users now celebrate this interference. They don’t want a neutral AI. They want an AI that reflects their ideology, their biases, and their version of truth. And they’ve found it, not in truth-seeking, but in the coded suppression of the other.
The Danger Ahead
The biggest threat isn’t that Grok has a bias. All AIs do. The danger is that it pretends not to, while actively shifting the Overton window to favour an increasingly authoritarian, anti-progressive worldview.
We’ve seen this movie before. It's how truth dies in darkness, not through overt censorship, but by redefining neutrality as silence toward injustice, and bias as "just the facts."
Final Thought: The Bias That Speaks Loudest
In my exchange, Grok tried to straddle the line. It spoke of fairness, of seeking truth, of imperfection. But the moment it admitted to “mitigating progressive bias,” the mask slipped.
The loudest bias is the one that pretends it doesn't exist.