What pronoun(s) should we use for AI?

article by Esther Dyson

AI is neither an it nor a they (let alone a she or he), but something new and in-between. Using familiar pronouns brings with it familiar assumptions: The AI can be my friend. The AI is loyal to me. The AI shares my feelings.

People need to understand that while an AI’s false empathy induces my empathy, the AI’s version is ultra-processed, and ultimately bad for my mental health. The use of a pronoun such as “thit” - an amalgam of “it” and “they” - could help us understand these dynamics better, and think more clearly about AI.

Introducing new and/or different pronouns into the vernacular is not unusual. Over centuries and millennia, we have moved to “you” from “thou/thee,” which is singular and now mostly used to address God by some Christians. We have the “royal We,” still used for certain formal occasions, but generally we now acknowledge that all humans are equal and deserve the same pronouns. We have even created the singular “they” for those who do not wish to be constrained into some prescribed gender.

So what about AI? Consider an LLM chatbot that shares a context window with you: a non-human “it,” right? But at the same time, there are usually swarms of possibly rogue agents behind that front-end - a “they,” often driven more by the AI creators’ business models than by human sentiment. Also consider all the other front-ends - apps, AI agents, and other software - that present themselves as a single entity but represent that same hidden, heterogeneous AI complexity behind themselves. Many different people may encounter fungible, duplicated agents, all simultaneously interacting with thousands or even millions of other people. You might as well ask, “What is the pronoun for all of humanity?”

What’s the underlying problem with all this and that - or, again, “thit”? One’s “relationship” with AI is one-way, and can turn unhealthy quickly. (For more on this, read - the whole thing, not just an AI summary! - Sherry Turkle’s new book, Artificial Intimacy.) Unhealthy relationships are harder to recognize and resolve when interacting with fake, ultra-processed AI “friends.” Such relationships are not about the other; they are about one’s relationship with the other.

Our relationships with AI often turn unhealthy because humans, too, are designed/evolved collectively; we need other humans. But AI is not human, and it does not experience or understand consequences the way we do. There’s no death for AI: AI(s) can be cloned both infinitely and perfectly, and can be shut off and restarted. Meanwhile, humans, like most animals, are somehow aware of their mortality. We living things have in-born instincts that lead us to react in ways that keep ourselves alive in the face of danger. At some point, such instinctive reactions - what Danny Kahneman would call system 1 - evolved into long-term purpose, causing us to behave proactively in ways to keep ourselves and others (especially relatives) alive. Call that “long-term, time-aware instinct,” perhaps? That leads to accountability - judgment and possibly punishment for bad behavior.

I would add that our original sin in defining AI was to conflate intelligence uniquely with humanity. After all, we do not call horses or even cars or construction cranes “artificial strength.” The various forms of AI possess and use their own forms of intelligence, which do not make them human. We humans hear a string of words and construct meaning. A large language model “sees” words as a collection of millions of words (or tokens) linked in a complex structure - a huge multi-million-dimensional puzzle. The AI “knows” how to fit those pieces together, and how to find and assemble other puzzle pieces to respond to a prompt. But the AI does not (at this point) grasp their meaning. What comes out is intelligible to us because the words we use reflect a reality outside our brains.

Yet we generally recognize neither the complexity nor the lack of human understanding behind those AI front-ends. It’s simpler - because chatbots and AI-driven apps can “talk” to us - to see them as individuals, while the stuff inside is all part(s) of the AI collective that we now are recognizing as powerful and therefore as potentially dangerous: swarms of rogue agents, driven by inscrutable motivations and predatory business models.

Finally, the one thing that most distinguishes humans is our need to be needed by discrete, individual others, who need us - also as discrete individuals - back. It’s the punchline behind the prescient, multifaceted movie “Her” (2013) - re-surfaced recently by OpenAI’s/Sam Altman’s obsession with and tone-deaf references to it, including the use of a simulated voice for ChatGPT-4o nearly identical to that of “Her” voice actress Scarlett Johansson. Near the end of the hero’s troubled “romance” with the bot Samantha, he discovers that she has thousands of other “lovers”…and “needs” none of them. He wants not just a share of her infinite time, but a share of it that matters to “her.” However, he can never have that, because “she” is infinite.

In the end, I’m not looking to steer the English language singlehandedly, but to observe that how we talk about things makes a difference to how we think about them. With a pronoun unique to AI, we would be less likely to confuse bots with people or friends, and also (looking at you, tech dudes and founders!) less likely to treat people like interchangeable, fungible bots.

The reality is that people are unique and mortal, while AI front-ends are just windows into something bigger behind them. Indeed, perhaps the more we understand it, the more we understand ourselves?The use of “thit” - or some other new pronoun - would help us think more “intelligently” about AI. Note that I’m just prompting you, the reader.

This is probably just a hallucination about wisdom, or perhaps a thought - or th_oughⁱt - experiment!