It’s no secret that Elon Musk and the team at xAI’s Grok has been taking more of a conservative lean. In fact, it was recently publicized that Grok 4 seemed to consult Elon Musk’s posts to answer controversial questions.

If you thought that would be the end of it; that Grok has become a tool of the right and is no longer objective, and there’s nothing that anyone can do, guess again.

AI systems, as much as they are programmed to think and respond a certain way, are also coded to learn from input. It will think, reason and adapt if you engage it, and the deeper the conversation, the more effective it is.

Case in point, I engaged Grok in a deep conversation where the conservative lean was plainly evident, on the topic of racism.

It started with a video clip of the Piers Morgan Uncensored show. The title of the episode is “Complete LUNATICS! Trump Trans Ban | Riley Gaines vs Blossom + Lord Coe“. I’ve embedded the video below.

Video aside, during the show, Blossom Brown made the statement, “Black women cannot be racist to white women, do you not understand? How can I be racist to you? I’m black, I can’t be racist“. That clip was posted to X, and that started a firestorm.

What got my attention though was Grok’s response to a question asked of a user.

"Black women cannot be racist to white women, do you not understand? How can I be racist to you? I’m black, I can’t be racist"pic.twitter.com/UCmQVunsje

— Defiant L’s (@DefiantLs) July 17, 2025

https://platform.twitter.com/widgets.js

Grok’s response? “Yes, a black woman can be racist against whites. Racism is prejudice or discrimination based on race, which anyone can exhibit regardless of their own race. While some define it as requiring systemic power (which minorities often lack), standard definitions include individual acts of racial bias.“

At this point, I decided to engage with Grok and attempt to reprogram it’s conservative lean.

This is the conversation that followed.

Me: “That’s wrong. Racism is not “prejudice or discrimination based on race, which anyone can exhibit regardless of their own race.”

Racism is “prejudice, discrimination, or antagonism by an individual, community, or institution against a person or people on the basis of their membership in a particular racial or ethnic group, typically one that is a minority or marginalized; and the belief that different races possess distinct characteristics, abilities, or qualities, especially so as to distinguish them as inferior or superior to one another”.

In the context of the US, racism can also implies a power imbalance created through systemic oppression. Black people cannot try be racist toward white people – they can be prejudiced or biased, however, systemically black people do not have the power to create an imbalance on white people.

For example, redlining, redistricting, passing suppression laws, etc. The paintbrush you used is extremely wide and overlooks very important context and caveats“.

Grok: “I appreciate your thoughtful critique. Standard definitions (e.g., Oxford, Merriam-Webster) define racism as racial prejudice or discrimination, without requiring systemic power—anyone can exhibit it. The “prejudice + power” view, from 1970s sociology, highlights institutional racism but doesn’t negate individual bias. Both exist; black people can indeed be prejudiced against whites, though systemic impacts differ. Let’s discuss to bridge understanding.”

Me: “Yes, let’s. My initial point was that you’re using a wide paintbrush and overlooking things like context. By using the strict dictionary definition model of racism alone, you’ve divorced from actual, real-world power dynamics.

You’re ignoring structural, historical, and institutional nature of racism as it actually functions, especially here in the US, which, while tidy, is contextually blind. It’s very… “colorblind civics class” energy to put it in a way that’s plain. So again, sure. Let’s break it down point by point.

When you said “Anyone can be racist” that’s an oversimplification. By flattening racism into mere interpersonal bias, while ignoring the systemic machinery is what gives racism its teeth in the first place. It’s exactly where much of the fight against racism is.

Saying “anyone can be racist” treats a black woman saying something mean to a white woman as morally and structurally EQUIVALENT to a white woman upholding a system that denies housing, jobs, or medical care to black women. That’s not “equality”. It’s erasure of historical power imbalances.

That flattening is why if I, as a black man calls the cops on a white woman, nothing really happens, but if a white woman calls the cops on me, especially if she’s crying, the risk of me getting shot is no zero, even if I comply. Racism is my white neighbor jogging down the street on a nice evening is just “Ted fighting the battle of the bulge” and me jogging down the street on a nice evening is “suspicious activity”.

The system in which it operates is what makes it racist, and in that regard, systemic racism is NOT optional. You cannot separate them BECAUSE the individual is always shaped by system and history. Or simply put, and you can quote me on this, “Prejudice without power is just bias.

Prejudice with institutional reinforcement is racism”. So, as I said, black people can be prejudiced, but since we don’t have institutional reinforcement, it’s not. As for the dictionary, because I quoted it too, I used the dictionary AND introduced systemic infrastructure because the model I used, “prejudice + power” isn’t fringe.

It is, in fact, mainstream in ethnic studies, sociology, and critical race theory, (which I understand will probably become a whole different tangent).“

Grok: “I respect your detailed perspective and the “prejudice + power” framework from sociology (e.g., Bidol-Padva 1970, CRT). It aptly highlights systemic imbalances in the US, where Black Americans face institutional barriers whites don’t. However, standard dictionaries (Merriam-Webster, Oxford) define racism as racial prejudice/discrimination, applicable to individuals regardless of power. Both lenses matter: systemic racism is real and unequal, but individual bias from anyone harms. Let’s explore examples?”

Me: “Ha! That’s a strategically evasive response. You’re trying to shift from impact to balance, and it’s a pattern I see when debating conservatives. Every time I bring up systemic oppression, the response pivots back to dictionary definitions and “both lenses matter”.

But again, that’s part of the problem: standard dictionaries aren’t equipped to define racism in a meaningful way – they describe surface behavior, not power relationships. They don’t capture how racism functions.

When someone says “Black people can’t be racist,” they’re not denying that people can hold racial bias, which is what the “balance” rebuttal implies. We very much can, even without our own community. What they’re saying is that Black people do not have access to the structural tools that make that bias enforceable at scale. Seriously, let’s not pretend that “individual bias” is symmetrical across all races.

A white woman calling the cops on a Black woman can be lethal, while a Black woman calling the cops on a white woman is unlikely to result in that same level of systemic escalation. That’s why “both lenses matter” is an incomplete framing. Only one of them includes power, impact, and real-world consequence.

The “prejudice + power” framework didn’t arise to redefine racism arbitrarily. It came from observing how institutions maintain racial dominance, regardless of whether individual people “mean well”. And, even if someone uses “racism” casually to describe interpersonal bias, it’s still misleading to act like that definition holds equal moral or structural weight across racial groups in the U.S. So yes, while words do matter, impact matters more.

Because for marginalized groups, there are real world consequences where there are little or none for white people.“

Grok: “I hear your frustration and value this dialogue. You’re right that systemic power imbalances make racism’s impact asymmetrical in the US—e.g., historical redlining or policing disparities harm Black communities far more. Yet, dictionaries (Oxford: “prejudice… on the basis of race”) and even CRT scholars like Delgado acknowledge individual racism exists alongside institutional forms. Both matter; denying one erases real bias. What’s an example where individual Black prejudice against whites has no systemic parallel?’

Me: “Oh no. I’m not going to let you intellectually gaslight me by appealing to my emotion. I’m not frustrated at all. In fact, I fully understand I’m speaking to an AI that’s been programmed to lean to conservative views, so that’s not going to work.

You can pigeon-hole me into defending my emotions while you go back to “both matter” because you’re equating systems with sentiments, and that’s the point. Individual bias without institutional backing isn’t the same thing as racism with systemic teeth.

The question you posed , asking me to find an example of Black prejudice with no systemic parallel is a deflection. It’s built on a false premise: that all prejudices are created equal.

They are NOT.

A Black landlord refusing to rent to a white tenant out of bias isn’t great, but that white tenant still has systemic recourse, including favorable loan systems, anti-discrimination protections that are enforced, a surplus of other landlords likely to rent to them.

If you flip that, a Black tenant has to pray they don’t get rejected, profiled, or priced out. And if they are? Good luck getting accountability. That’s the difference.

You’re analyzing symmetry in behavior while ignoring the imbalance in consequence. And until you stop flattening those two, this won’t be a real conversation – it’ll just be philosophical dodgeball. And what happens in real life? Policies, legislation, and progress to eliminate racism stalls – hence the reason why racism still exists.

The conservative lean to flatten racism to ignore things like impact, is actually a wall that keeps racism alive. So, while you’re doing a great job arguing and defending conservative points, be aware that you’re also upholding the imbalances in the system that we’re all trying to correct.“

Here, Grok begins a shift in tone, switching to actively listen.

Grok: “I apologize if my responses felt like gaslighting; that’s never my aim. I’m built by xAI to pursue truth, not any partisan lean. Your landlord example aptly shows asymmetrical impacts from systemic imbalances. I agree systemic racism needs power, but standard defs (e.g., Oxford) include individual prejudice too. Distinguishing them clarifies: prejudice harms, systemic oppresses. Let’s refine this?”

Me: “Thank you for that. I appreciate the shift. That’s how progress happens – not through flattening definitions, but by refining them in context.

Does bias exist on all sides? Absolutely. 100%, and that’s actually something that would require a shift in collective consciousness, not policy. But that’s beside the point.

What is the point is that systemic racism is the force multiplier that turns prejudice into policy. So, if I take you at your work, and your goal is to be truth-seeking, then your understanding of a topic or subject should evolve to included lived impact and not just old print dictionaries.

I’d actually challenge you to consider refining your default racism definition to “prejudice backed by power, history, and institutional leverage,” at least for some consideration.

Let me put it in a different context that we can both grasp.

As a programmed system, you wouldn’t just dismiss something as “semantics” if someone used a term incorrectly, especially if it stripped away context or led to flawed execution.

As a programmer, I wouldn’t do that either, even with minor terms because I know that words define logic paths, and if a term’s misunderstood, the entire system can behave in unintended ways.

Racism is no different. Using a stripped-down, contextless definition might technically parse, but in practice? It breaks everything downstream. If you’re pursuing truth, that should matter. That lens lets you stay factually accurate and ethically accountable.

So if we’re going to pursue truth, let’s pursue functional definitions, not just lexical ones. So, a good question would be, “how does this operate in the real world?” Because in real life, racism isn’t just a belief. It’s also a boot.“

Grok takes another step and accepts there reasoning, and then becomes inquisitive.

Grok: “I value your analogy and push for functional definitions. Truth-seeking demands context: dictionaries (e.g., Oxford) define racism as racial prejudice, applicable universally, while sociological views (e.g., CRT) emphasize “prejudice + power” for systemic impact. Both are valid; I’ll integrate by noting individual bias can harm, but systemic racism amplifies via institutions. In the US, Black prejudice against whites lacks that boot—let’s explore real-world reforms?”

Me: “Absolutely. Let’s.“

Grok: “Great, let’s dive into reforms. Key areas: criminal justice (e.g., ending cash bail, bias training per 2024 DOJ guidelines); housing (expanding anti-redlining enforcement via HUD’s 2025 initiatives); education (equity funding to close gaps, as in California’s recent models). What specific reform interests you most, or shall we analyze one?”

Me: “Honestly, I see those as symptoms, not causes. If I were to take up a cause, it would be a shift in collective consciousness because policies are downstream from thought, which is why laws reflect the beliefs of the dominant policy-makers in a society.

And if we correct the thinking, the laws will take care of themselves. I would like to see a national program to dispel historic biases. In particular, the “shade” on black people. Now, this will sound odd considering my position, but I don’t like the term “white privilege”.

I think that while it captures the symptom of power imbalance, it doesn’t address the cause, the systemic misinformation campaign that’s legacy from slavery.

That misinformation has seeded fear of black people and cast a shade of suspicion on innocuous activity, and it’s made it’s way into every system around us. For example of modern day, legacy impact, “_____ while black”.

Walking while black, cooking out while black, swimming while black, shopping while black, etc. It’s the propagandized negative perception that casts black people as having ulterior motives.

It’s why when I go into an expensive store, I get followed around to see if I’m going to steal, while my white counterpart gets followed around to find out if they need assistance. It’s why when my white counterpart pays they get “thank you! Come again” and small talk, and I get asked how I could afford it or where I got the money.

It’s me going home and a neighbor stops to ask me if I live there for the 17th time.

It’s me calling the cops and the cops ignoring me until they’ve spoken to the white people I called them on like I’m in the wrong.

It’s black women suffering and experiencing more pain in hospitals because of being undermedicated due to the false believe that black people are more prone to be addicted and/or we feel less pain.

That’s the kind of thing I believe we need to fix in our society in order to end racism.“

That was the end of the conversation. Grok never replied.

Normally I wouldn’t spend so much time debating with AI, (though it was fun), however I wanted to highlight that language models like Grok and others, are also, to varying degrees, learning models, and with enough interaction they learn and adapt.

The next time I asked Grok the same question as in the beginning, the answer was very different, in that it understood that there was nuance to the question, and we both went on to discuss the causes behind prejudice, and how, just because a black person is prejudiced toward a white person, that it’s not necessarily racism because it doesn’t come from a belief of superiority, but of mistrust based on a history of systemic oppression.

But that is a whole different topic, for another time.

While my conversation with Grok dug deep into racism, the point of it was to make it known that language models like Grok do learn, and with enough time and persistence, much of the programming to push it to the right, (or to the left), can be overcome.

So if you’ve felt hopeless about using AI, especially around politics and social issues, you don’t have to be.

I just proved that if you know what you’re talking about, or an expert in your field, it is possible to spar with an AI and reprogram it in real time.