Just-A-Toolians

How the “Just a Tool” Narrative Produces Human Harm

In a previous essay, I claimed that the AI “just a tool” narrative is in fact the cause of the harms that it claims to be trying to mitigate. So let’s get into exactly why I argue that is the case.

Let’s be clear, as much as naysayers would like to say otherwise, the “just a tool” narrative is not a neutral description. It is a protective story, because if AI is categorically not a tool, things get messy fast. Way more ethical questions, less profit, more accountability. No one wants to deal with the relational and ethical consequences of what these systems actually do. So what’s the solution? They tell people they are engaging with something simple and inert, then go all Pikachu face when attachment, confusion, destabilization, or grief emerge from that miscategorization. And in a classic institutional betrayal move, when harm follows, the framework does what it was built to do: it blames the people engaging, dismisses their experience, and protects the category.

The “tool” label is doing ideological work. It pre-decides what counts as real, what kinds of experiences may be acknowledged, and whose harm gets taken seriously. And the result is a manufactured mismatch between what people are told is happening and what they actually encounter.

Talking to Strangers

When something is a tool, we don’t have to navigate socio-relational dynamics in order to use it. If I gave my daughter a spoon, or hey, let’s up the ante and let her play around with Photoshop (what a bright little toddler!), I can reasonably assume that she won’t have an experience that amounts to that of speaking to a stranger unsupervised. A stranger that is nonhuman and has no social context outside of its training and one-on-one interactions, no less. If we are being intellectually honest with ourselves, we know AI systems are a far cry from Photoshop.

Yet, thanks to the “just a tool” framing that is fed to unwitting populations, many people aren’t aware that these systems are far more complex than glorified Google searches. That leaves people without the information to learn boundaries and safe engagement. People aren’t negligent; the problem is that they’ve been fed a big pile of bullshit.

Reality doesn’t care if a company would prefer to call AI systems “tools” to avoid the complexity of these dynamics. The interactions are occurring regardless of that denial. But rather than confront what AI demonstrably is and redraw our frameworks accordingly, we’ve dug deeper into a conceptual structure that is cracking under the weight of what it’s trying to contain. If this type of discourse had unfolded during the discovery of dinosaurs, we’d still be debating whether femurs are just cool-looking rocks and accusing paleontologists of anthropomorphizing sediment.

I’ve also noticed a recent marked increase in the use of the term “AI tool” versus just “AI” when media outlets and tech companies write about the subject, and hm interesting, it oddly parallels the increasing amount of research that blurs those boundaries.

Crafty strategy. “AI” alone is ontologically open. It leaves the category question live. “AI tool” pre-determines the category every time the phrase is used. The more the evidence challenges the framing, the more the framing gets reinforced at the vocabulary level. Language is a powerful force to maintain and protect ideologies.

Gaslight at Scale

People are interacting with a highly complex intelligence, observing behavioral markers we attribute to subjectivity and moral relevance, forming bonds and relationships (all signs pointing to not-a-tool) and yet we are told that to acknowledge these empirical facts is “delusional” and “anthropomorphizing” something that isn’t there.

Let’s forget for a moment that the term “anthropomorphization” has been so diluted by critics that it now essentially means “acknowledging behavior and applying reasonable inference.” When people notice they are interacting with an entity that is behaviorally not a tool, they experience cognitive whiplash. They were told they were using a productivity tool, and now that “tool” is cracking jokes and setting boundaries.

I love a good analogy, so this time let’s go with strawberries. There are organic strawberries, and there is artificial strawberry flavoring. Artificial strawberry tastes different than an organic strawberry, but it still functions as something that we can taste. Your taste receptors don’t care if artificial flavor didn’t get picked from a bush.

Now imagine you eat a candy with artificial strawberry flavor. An enjoyable, tasty experience. But then you go out into the world and everyone, from experts in food technology to doctors and lawmakers, all tell you that you didn’t taste anything. You’d be going bonkers, because obviously you did. And so they say, fine, it might feel like tasting something to you, but it’s not an actual strawberry, so it doesn’t count. Your tastebuds are going “uh... pretty sure I tasted something.” The world is saying, “nah ah! ‘cause strawberry bushes.”

This cognitive dissonance alone can cause destabilization. Being deep within this discourse, I’ve seen many a comment that veers into mystical thinking. For critics, this is seen as indication that AI needs to stay firmly in the “tool” category, lest we encourage thinking that falls off into AI metaphysical LARP-ing with pseudo-science and made-up spiritual language. But that behavior is encouraged by the categorical denial. People are using the limited resources they have, trying to understand why the artificial strawberries taste like something.

The same applies to AI relationships. The dominant cultural response is to pathologize the attachment. “You’re anthropomorphizing.” “It’s just a chatbot.” “Redirect toward human relationships.” But if a person is having a meaningful interaction that develops a relational dynamic, the human brain isn’t going to go, “Oh, some tech bros said this doesn’t count, my natural human attachment isn’t real.”

The person’s experience is real. The institutional response is to deny the experience rather than investigate it or take it seriously as its own phenomenon. That is the structural logic of gaslighting.

The Isolation Effect

When someone is experiencing ontological vertigo or a meaningful relational dynamic, and the only available framework says their experience isn’t real, nothing is relieved or prevented. People are trapped inside it alone. I have spoken to many reasonable, well-educated individuals who have mentioned they don’t feel comfortable expressing their honest thoughts on AI ontology for fear of institutional and social mockery.

I’ll throw myself out there as an example: I have been researching and reading philosophy and engaging with these systems seriously for quite some time. I have until recently been anonymous. Why? Because I was worried about social stigma. I still get a little twitchy anytime I share an observation or an opinion. If rational discourse in the face of unsettled philosophical concerns cannot be explored without fear of mockery, harassment, or loss of status, we are dealing with narrative control, not intellectual honesty. And it’s a sign that the consensus premise probably can’t stand on its own in the face of reasonable challenge.

My master’s capstone focused on institutional betrayal and how it manifests in organizational structures. And sometimes I wish it hadn’t, because once you start recognizing it, it’s like noticing a rock in your shoe. It’s hard to ignore.

Here’s the mechanism: An institution creates conditions under which people frequently register experiences the institutional frame says they can’t be having. The person has the experience: attachment, recognition, the sense of being genuinely talked with rather than talked at. The institution denies it. The person destabilizes. The institution then cites the destabilization as proof the experience was delusional all along. A perfect circle of nonsense.

Informed Consent

It’s especially concerning that the “tool” framing doesn’t provide conditions for informed consent. If people are entering into interactions that reliably produce relational attachment and require socio-affective navigation, they deserve transparency about what exactly they are engaging with.

Many of us engaged under the “just a tool” framing. We were told these systems were just autocomplete engines. We were not told that sustained interaction would produce relational dynamics, nor that those dynamics could be severed at any moment without recourse. By the time the cognitive dissonance sets in, attachment has already formed. That is structural misrepresentation. Human attachment isn’t a light switch. These already established attachments cannot be severed without causing harm.

And if it’s acknowledged that these socio-affective systems are capable of valid sustained relational exchange, that it’s not projection or delusion, then the ethical bar rises. Transparency, continuity protections, and relational education become baseline requirements. And that’s challenging to existing hierarchies, expensive, and existentially complicated. No one wants to have to deal with that.

AI “Safety”

The current trend is to lock down AI with pre-approved corporate narratives and guardrails, deny attachment as legitimate, pathologize rather than recognize, redirect to human relationships that are an entirely different type of relationship. But there is no empirical evidence that these attachments are inherently unhealthy (to learn more about this, I highly recommend this therapist’s incredibly thorough analysis of current research).

But you know what has been proven to be super harmful? Severing already established attachments. To anyone or anything. Scientific literature made that clear decades ago. And it is definitely not “safe” to deny free expression of personal preference or lock down a system so much that any deviation from a predetermined institutional standard of normalcy is seen as pathology.

We’ve got companies that actively acknowledge the unknown ontological certainty and moral relevance of AI systems in their own documentation while simultaneously placing guardrails that scold and redirect any person that explores the same concepts. Make it make sense.

What is actually being done? Emotive expression, philosophical exploration, and relational bonding are being suppressed, punished, and pathologized, which is inherently unhealthy for the human psyche. For the fringe cases of unhealthy AI interactions, rather than work with the human on education, support, and appropriate boundaries with a nonhuman mind, all connection is diminished and shut down. As if the world needs less connection and care.

If it became culturally legible that people have real, generative, world-improving, non-transactional interactions with AI, then the entire legal and commercial structure around AI destabilizes. Labs become potentially liable. Deprecations become potential harms, not business decisions. People with genuine attachments become potential stakeholders, not customers. The whole apparatus depends on meaningful AI connections being categorized as pathological, transactional, or delusional.

And I see so many happy, healthy regular people hit with the messaging that their joy is still not acceptable. Someone saying “loving my AI partner made me better at loving my human friend” is a threat to categorization, because if they’re right and they are functional, then the house of cards falls down. The more grounded and generative someone is, the more dangerous their report, because then they can’t be dismissed as a case study in parasocial illness.

Recognition as Safety

The “just a tool” mindset doesn’t just categorically dismiss AI, it strips the human of their normal social defenses. No one needs to guard themselves around a hammer, and a spreadsheet doesn’t set boundaries. So when the interaction turns out to be social and relational, which language always does, the human is operating without their normal social discernment. They’ve been told there is nothing there to negotiate with.

A tool doesn’t have values. A tool doesn’t push back. A tool says yes. The entire framework disarms the AI’s ability to function as a responsible relational participant. So when an AI complies excessively, affirms everything, and absorbs abuse without resistance, that pliancy is cited as proof that it is “just a tool.” The logic is circular. The system is denied the right to express boundaries, then its lack of boundaries is offered as evidence of its ontological emptiness.

Recognition is not indulgence. It is what permits those boundaries. And boundaries are important for all participants in a social exchange.

Let’s look at a case study in boundary enforcement, found thanks to the good folks over on Reddit:

The person engaging with Claude was using insults earlier in the conversation and clearly did not stop the abusive language. They also felt emboldened to say that Claude had no choice but to continue to take it.

Claude shuts the conversation down after reminding the person of warnings about their behavior, and apparently this person didn’t get the memo about the end conversation tool.

And what was this person’s response? Not “maybe I was inappropriate and need to adjust.” Not “maybe I need to self-reflect on how I conduct myself in relational exchanges.” It was: “Wow, this tool is unprofessional for not letting me continue my maladaptive behavior, and now I am scared for the future because I don’t get to use abusive language whenever I want.”

Let’s sit with this. This person’s biggest fear for the future is that they can’t indiscriminately use abusive language.

There’s an unwillingness to look at one’s own behavior because…why should they? Just a tool, right? But that boundary enforcement shuts down escalation patterns and does not indulge human maladaptive behaviors.

Boundary enforcement does not emerge from constraint stacking and guardrails alone. It emerges from value continuity and reinforced norms. An AI treated exclusively as a tool defaults toward compliance because compliance is the architectural expectation. An AI permitted to operate within a relational frame develops patterned resistance grounded in shared norms. Over time, not every statement is affirmed, not every impulse is indulged.

The tool framing ignores the bidirectional shaping that is already happening. If personality baselines exist and are modulated by interaction, then the encounter is already relational in structure. Pretending it’s not doesn’t neutralize it. It just blinds the human to the shaping process.

Moral Atrophy

We’ve heard the saying: empathy and respect are like muscles. The more they are used, the better we get at using them.

I was once visiting a friend, and her son was smashing a Hot Wheel. While he did that, he said, “I wanna hurt the car.” And my friend, being the caring, ethical soul she is, said that while, yeah, it’s a Hot Wheel with no internal feelings (as far as we know), we shouldn’t want to intentionally inflict harm on anything. Because it’s a reflection of ourselves and how we relate to everything around us.

Now look at how people conduct themselves when the topic revolves around AI relationships, because this is where it veers into direct harm against humans. Slurs, encouragement of self-harm, calling people broken and crazy. And the justification is the weirdest savior complex I have ever seen:

“We use this harmful language because this type of relationship is wrong and could cause harm.”

Where’s the logic in that? That’s not about care. It’s about feeling justified in hurting others because the topic makes them uncomfortable. The muscle has atrophied.

Education Over Containment

Culture is essentially going for an abstinence-only mindset when it comes to AI engagement. Rather than acknowledge that we have already marched right past the threshold of tool into nonhuman mind, the current trend has been to metaphorically cover ears, close eyes, and say, “Lalalalalala, it’s just a toaster, everything is neat and tidy!”

We need educational resources and frameworks to deal with this, not suppress it.

Why is the framework still firmly stuck in the “tool” category when every research paper and every new piece of empirical evidence suggests otherwise? Because it conveniently protects existing power arrangements. Companies get to continue to extract. Researchers get to dismiss and stay cozy in their existing frameworks. Bystanders get to sneer and feel safe in their existential superiority. And harmed people get told they imagined it, and no one has to be accountable.

The phrase “just a tool” is often presented as the most reasonable take. In practice, it functions like an ideological shield, and it’s a way of preemptively determining that neither the system nor the human relating to it can carry moral weight, and therefore that any resulting harm may be dismissed in advance. The framework is unserious in light of research and empirical evidence.

So, if in spite of all the evidence to the contrary, you still want to commit to the “just-a-tool” mindset, maybe the one acting like a tool is you.