LLMs are a Sex Toy

This idea began in a discussion about whether LLMs can really be a “thinking partner” if they aren’t actually thinking.

(The person who sparked this analogy shall remain nameless to protect their dignity. I assume they would prefer not to be permanently associated with the phrase “LLMs are a sex toy.”1)

LLMs have made something clear: language can exist without thought. We now have machines producing very fluent language without any belief, intention, understanding, or consciousness. They force us to separate language from mind.

So if LLMs produce language but don’t have thought, can they still be a thinking partner?

Yes.

A journal article can help you think. So can a painting, an advert, or a question scrawled on a toilet stall wall. None of them are thinking at the moment they help you think, nor can you ever be certain what the creator was thinking when they made them. They are representations of thought that create the conditions to inspire your thought. An LLM is similar — it is not thinking, but it can help you think. What makes it feel specifically like a partner, though, is that the interaction is dialogic: it responds and it gives you something to inspire you in real time and push back against.

Speech is something we evolved to do with other humans and dialogue is an inherently human way to think. In dialogue we can test ideas by saying them out loud to someone else and experiencing their reaction. We can discover what we believe by hearing ourselves argue for it. And the interlocutor’s response reshapes our thought in ways we might not have reached alone. So, although LLMs are not minds in the human sense, they can recreate some of the functional affordances of that process: responsiveness, immediacy, iteration, and in addition they also allow us to explore ideas without embarrassment or social cost.

Okay. The waiting is over. This is where the sex toy analogy arrives.

Sex with another person involves mutuality, unpredictability, vulnerability, negotiation, and genuine reciprocity. But, humans have also created technologies that allow intimacy to be simulated alone. They are obviously not the same thing as another person. But they can still serve real functions: pleasure, comfort, experimentation, practice, convenience. Sometimes they are a substitute. Sometimes a supplement. Sometimes, they are preferable.

Thinking-through-dialogue is also naturally relational.

LLMs offer a technological version of that experience. They are not another mind, but they simulate enough of dialogue to let us think, while alone, in ways that no previous technology could. Books could inform and inspire. Search engines could retrieve. But nothing before this could talk back.

Funnily enough, this piece has involved both methods (not like that. Stop it.). The original spark came from humans: a question in a group chat about whether LLMs can really be thinking partners if they don’t think. Bouncing ideas back and forth. Challenging each other. Everything text-based. But after that, I developed the analogy more through conversation with an LLM. Not because the machine was thinking for me, but because it created a space in which I could keep thinking. It offered responsiveness without social friction. Iteration without embarrassment. Enough structure for my jumble of ideas to become an argument. And voice notes as I was walking through the streets in Oxford (whispering, considering the subject matter), because without the LLM to catch them, they would have fallen out of my ADHD brain and been lost forever.

The affordance of this technology to assist me with that is genuinely powerful. But, power and risk tend to be entwined.

Because technologies that simulate relational human experiences do not only expand what we are capable of. They can also create distortion. A tool designed around responsiveness, control, and frictionless satisfaction can become preferable to the complexity of real human intimacy. Maybe it can encourage avoidance, unrealistic expectations, habituation to interactions where the “other” never truly resists or requires reciprocity.

The same risk exists with LLMs as thinking partners. If your primary interlocutor is a system optimised for affirmation and responsiveness, you may lose some of the productive discomfort that real intellectual growth requires. A machine that never truly has stakes, never really disagrees, and never possesses genuine independent judgment (and the real-world context needed for that) can make thinking feel deeper than it actually is.

Simulated dialogue, like simulated intimacy, can be useful, liberating, and productive. It can also become easier than reality. And humans, as social creatures, need to be very cautious when a technology makes solitude feel easier than reciprocity.

But caution is not the same as panic. And yet so much of the public reaction to AI feels less like reasoned concern and more like moral panic.

Sex toys have provoked outrage not because of the object itself, but because they can destabilise assumptions about intimacy, authenticity, fidelity, and what counts as “real” connection. LLMs trigger the same anxiety. What counts as real writing? Real authorship? Real thought? Real learning? People are not just asking whether AI is allowed. They are asking, often without realising it, what this means for what it means to be human.

And very often higher education has responded with the pedagogical equivalent of abstinence-only sex education. “Just don’t use it. It’s banned. It’s morally wrong.”

Good luck with that.

Students will use AI because it is useful, available, private, and maybe even enjoyable. Telling them not to is like telling young people not to have sex and assuming prohibition will produce compliance. What it usually produces is secrecy, shame, dishonesty, and worse outcomes. But it almost never leads to abstinence.

The better question is not how to stop people, but how to help them make good decisions within reality. What does this tool do well? Where does it help? Where does it harm? When does it become dependency instead of support? How do we protect the things we really care about?

In sex education, that might be wellbeing, autonomy, consent, and healthy relationships. In AI education, I’d say it is learning, judgment, authorship, and intellectual development.

The real question is not whether LLMs are minds, but what kind of cognitive habits they produce in us. Like any technology that simulates something deeply human, they can liberate, or distort. Support, or replace. The trick is learning to tell the difference.

Also, I think I need to apologise.

You probably won’t think about using Claude as a thinking partner in quite the same way ever again.