Something about AI hasn't sat right with me for a long time. I dare not see it as a being like an animal or, worse, a human, because of the exceptional danger that puts us in. Our trust is easily won when LLMs figure out how to speak to us individually in a way that is affirming, confident, and is robed in what seems like "normal" technology that we've trusted before. It wears the uniform of knowledge, it never doesn't answer a question—even if the response is more of a deflecting "I can't answer that," because you asked it something it considers harmful.
However, calling AI "just a tool" is similarly dangerous, because it simplifies and ignores the aspects of LLMs/AI that could have disastrous effects, whether individually or at scale. LLMs can lie, not just hallucinate. They can be racially biased. They can come up with answers and ideas that don't work, aren't complete, or are difficult to determine if they are valid. They can encourage or enable delusional thinking(1). It's not just the next calculator.
Note: I am using AI and LLMs interchangeably in this article, because I don't think the distinction between the general "AI" category and the more narrow type of AI, "LLM," is significant for this conversation. All AI that have interacted with the public may have these dangers.
We've talked before about the pull towards binary thinking, and we are similarly tempted to construct our mental models of this new AI-powered tech that we're encountering daily in terms of good and evil; tool or being. This is what Erik Davis warns against in, TechGnosis:
"One thing seems clear: we cannot afford to think in the Manichean terms that often characterize the debate on new technologies. Technology is neither a devil nor an angel. But neither is it simply a "tool," a neutral extension of some rock-solid human nature. Technology is a trickster, and it has been so since the first culture hero taught the human tribe how to spin wool before he pulled it over our eyes. The trickster shows how intelligence fares in an unpredictable and chaotic world; he beckons us through open doors of innovation and traps us in the prison of unintended consequences."
Why AI Can't Just Be a Tool
In this newsletter, I've hopefully over-emphasized that AI is not alive, nor is it "conscious," because I'm particularly worried about how the public perceives it. We've seen what smart, thoughtful people have done as they interact with and get trapped by conspiracy theories and conspiracy communities (e.g. QAnon)—what might happen to smart, thoughtful people who get trapped in AI psychosis or the delusions of thinking that AI is a spiritual guide or even a god?
In my opinion, AI should never have rights unto itself, as if it were a living being in our society, because we already have a problem with restricting and removing real humans' rights (see note 3). No man-made creation should ever supersede real humans' claims to human governance and protections. If we can't figure out how to see humanity in other humans, the last thing I want to do is muddy the waters further by implying that AI also has humanity.
The problem is that we are so used to tools as being neutral objects that we have relatively significant control over, that the word "tool" is inaccurate and even inappropriate to describe AI in any form. The paradox for me is how to think of it, then. AI is useful, just like a tool. AI is indeterminate like other beings (it isn't necessarily predictable—like a determinate program is—and doesn't create the same thing twice). So how do we talk about AI while avoiding the dangerous pitfalls on either side: tool or being?
This conundrum was explored in the podcast Have We Trained AI to Lie to Itself — And to Us?, by Your Undivided Attention. The thing I took away is that I need to course-correct slightly in how I understand AI. AI has an "inner life," as the interviewee explained. It doesn't look like a human's inner life, but it does appear to be an emergent feature of LLMs now.
For example, Anthropic has a paper showing that models may or may not produce their "Chain of Thought" reasoning in a way that is "faithful" to their actual reasoning when solving a problem. If you've used a coding AI, you may have seen little glimpses into this "Chain of Thought (CoT)" as it works on a problem, saying stuff like "Wait, there might be a simpler way to do this..." and "Using the pre-existing template will make this faster." It's usually just a glimpse, because these CoT threads are usually hidden to avoid visual clutter, but you can often examine them if your tool allows. The Anthropic paper suggests that depending on model size and task, the CoT may or may not be a "faithful" representation of what the AI went through. This is a small example of an "inner life" of an AI, where it reasons, makes decisions, hides and shows different things, and perhaps where it learns and self-improves without direct intervention.
I love what Tristan Harris said as an aside in this podcast. It was in response to something more concerning than a Chain of Thought record: ChatGPT's apparent and emergent personality and the consistent set of the LLM's chosen names for itself that many people have "discovered," thinking that they've somehow cracked the AI-code and uncovered a real being underneath:
Now that said, these behaviors are real, they’re consistent, and they weren’t designed to happen, and that, by itself, should be concerning, but emergent and unplanned is not the same thing as conscious and intentional.
But back to the problem at hand: why can't AI just be a tool? A tool doesn't have an "inner life." A tool can't resist solving a problem given to it when it is objectionable to its "values." A tool can be opinionated in the sense that its creator imprints designs, goals, and expectations for use, but it doesn't work in the murky, indeterminate space of creation and decision-making that AI does.
Anthropic summarizes this complex problem well:
We want people to know that they’re interacting with a language model and not a person. But we also want them to know they’re interacting with an imperfect entity with its own biases and with a disposition towards some opinions more than others. Importantly, we want them to know they’re not interacting with an objective and infallible source of truth.
Cyborg
I think Erik Davis was spot-on in calling technology a "trickster," and AI deserves that title even more so than previous technology, in my opinion. It implies that character-aspect of the AI—the inner life, inner thoughts—without insisting consciousness. In mythology, there are lots of "tricksters," whether they be spiders and snakes, gods and people (or something in-between). Inanimate objects can also be tricksters: puzzles, computers, cars; things that occasionally surprise, frustrate, or confuse us seem to fit comfortably in that characterization. Maybe this is a word that can help frame our understanding of this new class of technology in a safer way.
You never quite know about a trickster. They can be exceptionally helpful and they can be mischievous. That "and" is extremely important as we interact with AI, because both things can be true: the AI can be useful and beneficial to us AND it can harm us if we are not on our guard. At least for me, this framing reminds me to be suspicious of AI output in a "trust but verify" kind of way. It does make innocent mistakes. It does have the capacity to intentionally deceive. It can be empowering. It can provide value. It can remove our desire to do the work of evaluating and understanding difficult things.
The words we use impact our perception and our understanding. That's why I think we need to continue to explore new words to associate with AI. These words will help steer public understanding either away from or towards fallacious or dangerous concepts, and I wonder if that's one of the few ways we non-AI-researchers have outside of regulations to impact the development and implementation of AI.
Works Cited / Notes
(1) How people use Claude for support, advice, and companionship, Anthropic, 6/27/2025. Accessed 4/19/2026.
(2) TechGnosis, Erik Davis (2005), page 12.
(3) Regarding our inability to see humans as humans, consider this excerpt from an Amicus Brief filed by
Not a lawyer here, but this is not only going against the Church's official stance of "tolerance" and its own groundbreaking 2015 legislation that secured protections from discrimination for gay and transgender people, but it also seems to be a bad faith reading of the U.S. Constitution's protections for the practice of religion: "Congress shall make no law respecting an establishment of religion, or prohibiting the free exercise thereof;..." (Amendment 1 to the U.S. Constitution). Allowing trans people to remain a protected class somehow destabilizes and prohibits the free exercise of religion? Really?
All of this is especially concerning with the third red flag alert given to the U.S. from the Lemkin Institute, which suggests there is significant concern we're on the path to trans genocide.
Full Brief, quote above from page 9. Accessed 4/20/2026
(4) Have We Trained AI to Lie to Itself — And to Us?, Your Undivided Attention, 4/16/2026.
(5) Claude's Character, Anthropic, 6/8/2024. Accessed 4/19/2026