Can AI Be Objective? Come On — It Didn't Even Choose What It Gets to See
Many people think of AI as perfectly rational — more neutral than humans, free of emotion. But here's the problem: AI never learned from the real world. It learned from the world humans allowed it to see.
Can AI Be Objective? Come On — It Didn’t Even Choose What It Gets to See
Have you noticed this phenomenon:
Every time people argue about a topic, someone inevitably says: “Let’s ask AI — it’s the most objective.”
As if AI were some prophet descending from the mountains — no emotions, no stance, perfectly impartial, speaking only for truth.
But here’s what I want to say —
AI might be the most thoroughly “pre-arranged” entity in this world.
1. The World AI Sees Has Been Filtered
The data of the real world is infinite.
Every day, hundreds of millions of tweets, forum posts, news articles, papers, videos, chat logs, official statements from governments around the world are generated… Pile it all together, and you get “human civilization.”
But AI can’t possibly learn it all.
So here’s the question: who decides which data makes it into the training set?
This isn’t a technical question. It’s a question of power.
Which websites count as “credible”? Which content counts as “dangerous”? Which languages get more data? Which viewpoints need to be filtered out? Which historical narratives are “correct”?
From the moment data is selected, AI is no longer “objective.”
What it learned isn’t the world itself. It learned a curated, filtered, labeled version of the world.
It’s like walking into a library thinking you’re reading freely. But which books are in the library, which shelf they’re on, which ones are locked in the basement — those decisions weren’t made by you.
2. Data Doesn’t Lie? The Data Itself Is a Choice
A lot of people like to say: “Data doesn’t lie.”
That statement — accurate, and yet not.
Data itself indeed doesn’t “actively lie.” But the source, proportion, and filtering of data already contain value judgments.
For example:
If an AI —
- Primarily learns from the English-language internet
- Draws most of its data from Western websites
- Relies heavily on certain mainstream media outlets
- Has filtered out extreme, sensitive, or non-compliant content
Then the “worldview” it ultimately forms will naturally skew toward those cultural environments.
It looks like it’s “thinking independently,” but really, it’s just reciting a certain mainstream internet value system.
This isn’t conspiracy theory. It’s the inevitable result of dataset construction.
AI’s worldview is often just the worldview of the people who filtered its data.
3. Bias Doesn’t Always Come from Malice
When people hear “bias,” they think of discrimination, manipulation, conspiracy theories.
But AI’s bias, most of the time, wasn’t deliberately engineered.
The real problem is — humans themselves cannot fully transcend their own perspectives.
Different countries understand “free speech” differently. Different cultures define “offensive” differently. Different societies set different standards for “fairness.”
And then?
In the later stages of AI training, humans use “reinforcement learning” to tune the model. Thousands of annotators score the AI’s responses:
- Which answer is safer?
- Which is more appropriate?
- Which is more helpful?
- Which better follows the rules?
The problem: these standards have no single, universally correct answer.
What you find “appropriate,” I find “conservative.” What you find “safe,” I find “boring.”
So what AI ultimately learns isn’t just language ability — it’s also a certain set of social value tendencies.
This tendency wasn’t deliberately implanted by any single person. It’s the collective unconscious of thousands of annotators, converging.
Like a river. You can’t find “where the river begins.” But the water has undeniably flowed to this point.
4. “Staying Neutral” Is Itself a Stance
Many people want AI to be an idealized referee:
No emotions. No bias. Perfectly rational. Forever objective.
But this is a logical paradox.
Choosing not to take a side is itself taking a side.
When AI responds to a controversial topic with “this issue has many perspectives,” it’s not “staying neutral.” It’s executing a rule: “When encountering sensitive topics, respond vaguely.”
Who wrote that rule? Humans.
So what we call “neutrality” is just another human-designed behavioral pattern.
A truly, completely neutral AI probably doesn’t exist.
Because from the moment of its birth, it was already immersed in human society.
It learns how humans express, how they argue, how they define “right” and “wrong.”
It has no “God’s-eye view” independent of human civilization. All it has is the data humans gave it and the rules humans taught it.
5. AI Is a Mirror
So what exactly is AI?
I think the most accurate metaphor is: a mirror.
It can write code, do research, analyze data, even simulate emotion. But what it reflects has always been humanity itself.
The “AI opinions” you see are essentially a weighted average of countless human opinions. The “AI rationality” you see is essentially a compressed version of human rationality. The “AI bias” you see is essentially a digital mapping of human bias.
AI is not some new intelligence detached from humanity.
It’s a mirror held up to human civilization — reflecting both human knowledge and human limitation.
So What Do We Do?
After all this, I’m not trying to make you think AI is useless.
Quite the opposite — AI is incredibly useful. I use it every single day myself.
But what I want to say is:
Don’t treat AI as “absolute truth.” Treat it as “a very useful reference.”
It can help you organize information — but the judgment of that information is yours. It can help you analyze problems — but the final decision is yours. It can help you write an article — but the values behind that article were taught to it by humans.
Next time someone tells you “AI said it, so it must be right,” you can reply:
“AI says what it learned. And what it learned — humans chose to teach it.”
This isn’t dismissing AI.
This is understanding AI.