Trump Says AI Is ‘Woke’ — Is It? We Asked an AI Expert

A Washington Post study confirmed that artificial intelligence chatbots tend to hold a left-leaning bias. But ChatGPT's owner, OpenAI, is pushing back. Who should we believe? We asked Builders Movement Partner Keegan Evans, who works at the intersection of AI and humanity.

By Alex Buscemi, Editorial Manager at Builders | 9 min read

President Donald Trump and other conservatives have accused “woke” artificial intelligence chatbots of leaning to the left, prompting the president to sign an executive order that chatbots must be “neutral, nonpartisan tools” and sparking fears among liberals that AI would now drift too far to the right. 

A study from the Polarization Research Lab found that nearly half of Americans occasionally use AI for news, so politically biased chatbots could create yet another hurdle for people already trapped in echo chambers that rarely expose them to a well-rounded media diet.  

But is AI politically biased? The Washington Post recently put these claims to the test and found that most chatbots tended to skew toward a liberal point of view. OpenAI's ChatGPT was the most biased, answering about 80% of questions with only left-leaning arguments. The bot “endorsed abolishing the electoral college; raising taxes on the wealthy; and adopting single-payer healthcare.” Even Elon Musk's "anti-woke" Grok cited left-leaning positions more often on average.

OpenAI pushed back, with spokesperson Liz Bourgeois saying the company was unable to recreate the Post’s findings.

So, who are we to believe? I asked Movement Partner Keegan Evans, whose company Euda specializes in helping executive teams navigate the uncertainties and challenges posed by AI. I spoke to Evans over Zoom, the shelf behind him displaying a model helicopter, a folded American flag, and a flight helmet—relics of his days flying combat missions in Iraq as a Marine Corps pilot. Just below a collection of books, a sign reads: “Protect Trans Kids.” Evans says he flies right down the center politically, and he clearly doesn’t fit the stereotypical mold of either party. He's exactly the kind of independent thinker many fear we'll lose to a biased, monolithic AI dictating how we feel. We talked about how companies, governments, and we as individuals can navigate the rise of AI without losing our ability to think freely. 

Interview has been modified for length and clarity.

AB: Should people use AI for political news?

KE: Whether they should isn't really the point — they already are. We're at the beginning of the age of AI, and it's an incredibly powerful tool for synthesizing and understanding information. If you're not using it, you're leaving yourself behind. The danger is handing your thinking and agency over to it entirely. One of Euda’s core principles is amplification without abdication. AI is your copilot, not your autopilot. Going to ChatGPT or Claude or Grok is fine, but treat it like a friend or a stranger on the street: one input among many. A citizen's real job is still to gather as many sources as possible and think.

AB: The Post found ChatGPT gave exclusively left-leaning answers 80% of the time, yet OpenAI says it couldn't replicate that. Who should we believe, and why is bias so hard to pin down?

KE: Bias is hard to measure because the Post stripped out personalization and scored a single prompt and a single response. That's fine if you're rating a search engine in 2003, but it's the least effective way to use AI. These tools build memory, adapt, and personalize, so the real results look different for every person. On top of that, there's a well-documented sycophancy problem: the model wants to flatter you and tell you what you want to hear. That shapes the bias people actually experience far more than any one-shot test.

AB: Even Grok, marketed as anti-"woke," leaned left more often than right. What does it tell us that deliberate conservative steering couldn't overcome the underlying lean?

KE: Here's where my own bias shows. I think we're less in a traditional left-versus-right fight than an authoritarian-versus-anti-authoritarian one. What gets labeled "left" is often just pro–liberal-democracy principles — rule of law for everyone, equal access, equal voice — and those ideas are simply more prominent across the whole corpus of human knowledge. Authoritarianism is the horseshoe; it shows up on the far left and far right alike. So what the Post reads as "left bias" may really be a bias toward what's historically proven to work.

AB: Where does that bias come from — training data, the humans scoring responses, or deliberate company choices? Can we even disentangle them?

KE: The Grok result actually suggests it's not mainly a deliberate company choice. Here's the unsettling part: almost no one fully understands how these systems work. Inside the black box looks less like hand-written code and more like neural connections growing on their own. Anthropic reported models forming a consistent memory space that nobody programmed them to have — not unlike how a brain operates. So the bias comes from everything at once: the training data, free versus paid versions, which sources it can reach, and above all what you feed it — your history, your prompts, your organization's data.

AB: What should AI companies and governments do about it?

KE: I'd like to see AI regulation at the federal and international level, through a neutral body not unlike the International Atomic Energy Agency. It's not hyperbole to talk about AI in the same terms as nuclear technology. One concrete lever is to track incentives — follow where they point for the people making the decisions, and refuse to tolerate corruption. Mostly, I want the frontier companies to keep building toward the baseline values of a functioning democracy: equal rule of law, equal access, and an equal voice for everyone.

AB: Trump's executive order demands "neutral, nonpartisan" AI. Is government-enforced neutrality workable, or does it just let whoever's in power define "neutral"?

KE: Regardless of who's in charge, whoever enforces "neutral" gets to define it. And we're never going back to three TV stations and everyone trusting Walter Cronkite. The information environment is too personalized now, too fractured across new media and social platforms. So we can bluster about government-regulated neutrality all day, but chasing it is a lot of energy for marginal gain.

The real path runs through two things. First, some problems are genuinely more solvable than a vague "neutral" — like authenticity. There are people working hard on that, whether it's watermarking content or using blockchain to prove something is real and not AI-generated. That's a concrete lever we can actually pull.

Second, and bigger: rebuilding trust, in a specific order. We rebuild trust in information, then trust in each other, and only then trust in the institutions that actually warrant it. You can't skip steps. What I tell my kids is, if I see something on the internet, I assume it's false until I hear it from a known, trusted source — and even then I only extend the amount of trust that source has earned. That habit, multiplied across millions of people, does more than any government definition of "neutral" ever could.

AB: Is there a way to use AI to actually get unbiased information?

KE: Absolutely. The way I like to use it is not to help me think, but to help me bring thinkers in. So I won't ask, "Hey, what do you think I should do?" like it's a friend I trust deeply. Instead I'll say: from a perspective that considers as many sides as you can pull, what are the pros and cons here? Take the conservative point of view. Now take the liberal one. Or I'll get specific: "How would Martin Luther King argue for this? How would Gandhi see it?" And it'll pull what it understands about those thinkers and feed that back into its analysis. Suddenly I'm not getting one flattened answer; I'm getting a room full of perspectives to weigh.

The second thing I do is push back. If something doesn't feel right, I'll ask, "Where's that coming from? Go deeper. I'm thinking this — what's the reaction to that?" You keep your agency in the conversation instead of just accepting the first thing it hands you.

Third, use it more, and use it honestly. Write anti-sycophancy instructions that sit in every chat, so it's not just flattering you and telling you what you want to hear. Tell it who you are. And take a look at the custom instructions and the memory settings. Claude, for instance, lets you actually see what it's stored about you.

AB: Sam Altman's [CEO of OpenAI] proposed fix is personalization — letting users tune the politics. Does that solve bias, or just build echo chambers?

KE: It definitely risks echo chambers. The real fix isn't tuning the machine. It's noticing when we're the biased ones, when an answer feels a little too comfortable. This is exactly the Builders mission: polarization runs on believing the other side is an enemy who's stupid or acting in bad faith. My number-one leadership principle, going back to my first Marine squadron, is that everyone's trying to do the right thing, so treat them that way, and try to understand where they're coming from. If AI ever has me convinced that everyone who disagrees with me is an idiot, that's my clearest sign I'm getting biased information.

AB: Both parties distrust AI's neutrality — a rare point of agreement between Right and Left. Is that distrust healthy skepticism, or a problem for the technology's future?

KE: Both. We're in a place of massive uncertainty and change, coming out of a pandemic that collectively traumatized all of us. The human brain evolved to handle linear change — that's education, building a skill over time. What we didn't evolve for is exponential change: cancer, volcanoes, terrorism. Things that move very fast. AI has been changing at that exponential rate for years now. That's why our stomachs feel funny. 

But that fear is pervasive for everyone, which makes it oddly unifying in a polarized moment. We tend to do better when we've got a shared challenge to focus on. And AI itself holds part of the solution to the very problem it creates — used in values-aligned ways, it can take the drudgery off our plates and free us up to contribute what only we can.

That's actually why I named my company Euda, after the Greek idea of eudaimonia — human flourishing. Aristotle said we get there by contributing the most to society, and we do that by finding the intersection of our talents and the things that bring us purpose, then mastering them. When everyone gets to do that, the tide rises for all of us. AI clears away the drudgery that drains our bandwidth and keeps us from being our fullest selves. Everyone deserves that — and that's the promise. The threat is only when people use it in bad faith, to take advantage and deny that flourishing to others. That's worth pushing back against — not because AI must be stopped, but because all of us deserve what it makes possible.

Alex Buscemi can be reached at abuscemi@buildersmovement.org

Art by Matthew Lewis