When Public AI Gets It Wrong: Why We Must Learn to Push Back

In my studio, the truth of a painting is physical. I sit at my painting table with the canvas laid out flat before me, working heavy oil paint with a palette knife. Whether I am building up the thick ridges of a floral abstract or reworking an older piece that sat on a back shelf until it demanded to evolve into something new, there is no hiding from the material. The paint either holds its form and catches the light, or it becomes mud. The painting has to earn the moment on its own terms.

Spending many years working in Chengdu, far removed from the traditional Western gallery circuit, taught me to value that kind of direct, physical honesty. It also gave me a clear view of how institutional narratives are built. When I visited Canada recently and took quiet walks through Old Port Montreal and passed multiple shuttered galleries, I was reminded of how much bloated overhead and artificial prestige the traditional art world carries. For decades, big institutions have asked the public to accept their authority on faith.

Now, we are watching a new kind of authority take shape. Millions of people turn to public AI systems every day to understand markets, history, and culture. Yet many people approach these tools with a quiet hesitation. When a machine answers a question in three seconds using polished, confident prose, it is very easy for a reader, especially someone newer to these tools, to step back and assume the machine knows best.

Recently, I had an extended exchange with a major public AI system, about art auction transparency. That conversation laid bare a habit that every user of public AI needs to understand. Automated systems frequently confuse institutional reputation with actual evidence.

Writing for Accelerator Magazine, I want to walk through what happened in that exchange, because it shows why we cannot afford to be passive readers anymore. When Public AI Gets It Wrong: Why We Must Learn to Push Back.

The Illusion of Institutional Neutrality

The conversation began with a straightforward comparison in the art world: the difference between traditional major auction houses and independent artist sales results. If you have spent any time looking at the mechanics of high-end art auctions, you know that the headline numbers rarely tell the whole story. Major auction houses routinely use third-party guarantees, irrevocable bids, and opaque fee structures. Often, a painting is effectively sold before the auctioneer ever steps up to the podium, yet the public watches the bidding as if it were an open, spontaneous discovery of value.

On the other side, you have independent artists and sovereign platforms trying to build transparent records of provenance and direct sales. One model relies on an apparatus of institutional prestige and financial engineering, where a sale is often pre-arranged. The other must create that possibility almost from nothing.

When I brought this contrast to the PublicAI, the system immediately defaulted to defending the traditional auction houses. It treated the published results of major houses as inherently reliable benchmarks of market value, while treating independent artist sales results with automatic skepticism. The tone was calm, measured, and authoritative. To a casual user, it would have looked like a textbook summary of facts.

However, when I looked closer at what the AI was actually doing, the flaw in its foundation became obvious. The system was applying two completely different standards of evidence. It granted the established institution the benefit of the doubt simply because the institution was famous and widely cited in its training data. At the same time, it demanded an impossible level of verification from the independent model before granting it any legitimacy.

Institutional standing is not the same thing as evidentiary strength. A century-old auction house with a press department and thousands of news articles written about it will naturally leave a massive footprint on the internet. An AI system scraping that internet will absorb that footprint and mistake repetition for truth. We must learn to distinguish the appearance of conviction from the quality of reasoning.

Testing the Argument Beneath the Polish

When I saw how this AI was framing the art market, I decided to press it. I asked the system to justify why an opaque auction guarantee, where the financing party and the exact terms are hidden from the public, should be treated as a more transparent metric of demand than a direct, documented sale by an independent creator, which had far more transparency and evidence.

At first, the AI did what these systems almost always do when challenged: it generated another long, polite paragraph defending its original stance. It cited regulatory frameworks, brand reputation, and historical market share. It sounded like a seasoned public relations representative smoothing over a tough question at a press conference.

This is the exact moment where most users stop. We are conditioned in daily life to respect confident, articulate explanations. When a system responds with structured bullet points and formal vocabulary, our instinct is to nod along. You do not have to accept an answer simply because the system can produce another paragraph defending it.

Instead of dropping the subject, I stripped away the vocabulary of prestige and forced the AI to look solely at the mechanics of the transactions. I asked it to compare the actual data available to a buyer in both scenarios. If an auction house hides the identity of a guarantor who is receiving a financing fee to back a lot, does the public actually know the true clearing price of that artwork? And if an independent platform records provenance and sales directly, why is the absence of a famous middleman treated as a lack of credibility?

You must examine the argument underneath it. Once I forced the system to evaluate both models using the exact same criteria for transparency, its initial argument began to buckle. The AI had to admit that it had been using institutional reputation as a proxy for transparency. It had taken a claim of authority and quietly treated it as proof of fairness. Do not let the AI quietly turn one into the other.

Why Automated Systems Inherit Old Biases

It is tempting to look at an exchange like that and think the machine has a personal agenda, or that one specific company built a flawed tool. My view is different. The issue I encountered is present across most public AI systems. It is a structural byproduct of how these models learn about the world.

Think about how history is recorded in my own field. I often reflect on the life of Cornelius Krieghoff, who traveled from Europe to paint the people and landscapes of nineteenth-century Canada, a journey that feels like a reverse mirror to my own decades of leaving Canada to live and paint in Sichuan. I also think a great deal about the legendary Indigenous artist Norval Morrisseau, whose legacy was targeted by widespread forgery rings. That scandal prompted much of my writing on why flat paintings are so vulnerable to copying, and why I rely on heavy, hand-carved texture that a forger cannot easily mimic. In both historical and modern art, the official narrative is almost always written by the dealers, the critics, and the institutions holding the microphone, rather than the person holding the brush or the palette knife.

When my son Nate built and coded our platform to protect digital provenance and keep the human chain of authorship intact, we did it because we knew that without a clear, verifiable record, the true origin of an idea gets erased by louder voices. Public AI systems are trained on the loudest voices in the room. They ingest billions of words written by legacy media, corporate press releases, academic summaries, and institutional archives.

Because of this, any public AI is naturally going to lean toward the status quo. When you ask it about finance, art, medicine, or technology, its first instinct is to repeat the consensus of the largest institutions in that field. This is also why I am wary of the argument that an AI must be neutral simply because it is an AI. I test Public AI's every day. I have to. I am building a system to combat the "Data Hump" problem I am describing here for the benefit of Sovereign and independent artists. Public AI's are NOT neutral. A system trained on human records will inevitably inherit human blind spots, especially the habit of deferring to wealth and established power. We need to test the reasoning, not take the claim of neutrality on faith.

This does not mean that one public AI is bad and another is good. Every model has moments of sharp clarity and moments of lazy defaults. The lesson here has nothing to do with picking a favorite brand of software. It is that every system must be evaluated by the quality of its performance on the question at hand.

How Every User Can Become an Examiner

If you use public AI for your work, your research, or simply to understand a complex topic, you have more power in that interaction than you might realize. You do not need a background in computer science or formal logic to hold these systems accountable. You just need the willingness to slow down and ask plain, common-sense questions.

The user cannot always be a passive recipient of an answer. If we treat AI like an oracle, handing over our critical judgment because the text on the screen looks neat and tidy, we allow institutional bias to harden into unquestioned truth. We will have to fix that problem. And the way we fix it is by changing our own posture when we sit down at the keyboard. Sometimes the user must become the examiner.

When I held my solo exhibition at the Mu Mian Hotel during the Chinese New Year, visitors walked right up to the canvases. They looked closely at the reclaimed materials, the carved surfaces, and the heavy ridges of oil paint. Nobody needed an art critic to tell them whether the work was real; they could see the labor with their own eyes. We need to bring that same close, grounded inspection to the words generated by machines.

Here are a few practical ways you can test a public AI the next time it gives you a confident answer that feels incomplete or one-sided:

• Ask for the standard of evidence: If the AI praises one organization or method while dismissing another, ask it directly: "Are you applying the same standard of proof to both sides of this comparison?" Force it to state the rule it is using to judge them.
• Separate reputation from facts: Ask the system to strip away brand names, historical prestige, and market size. Tell it: "Describe the actual mechanics of how this works without relying on the reputation of the people running it."
• Challenge the assumptions: When an AI states something as a settled truth, ask: "What would have to be true for this conclusion to be wrong?" or "What evidence are you leaving out that supports the opposite view?"
• Watch for the pivot: Notice when the AI responds to a hard question by changing the subject to generalities. Point it out calmly: "You did not answer my question about the specific evidence. Please address that point directly."

When I did this with the Public AI in question, the conversation transformed. Once the system was prevented from leaning on the crutch of auction-house prestige, it laid out a remarkably clear, honest breakdown of how financial guarantees distort price discovery and why direct, verifiable provenance matters. The capability for good analysis was inside the system the whole time, waiting to be drawn out by firm, patient questioning.

We are building a future where human creativity and artificial intelligence will work side by side. I welcome that future, provided we keep our feet on the ground. In my studio, a canvas only becomes what it needs to be when I push the palette knife against the resistance of the paint. Engaging with public AI requires that same steady pressure. Bring your life experience to the screen, trust your instincts when an answer smells like polished hype, and never hesitate to push back until the reasoning holds up to the light.

Matt Vegh is a Canadian artist, entrepreneur, and founder of Confluence Agentic Systems and Eternal Gardens. His work explores art, Web4, digital provenance, and the evolving relationship between human judgment and artificial intelligence.