Something to React To
To be honest, I use AI a lot. Probably more than I ever expected, and probably more than most people I know.
Sometimes it helps me polish a paragraph. Sometimes I use it to brainstorm. Sometimes I simply ask questions out of curiosity. I don’t think any of these are particularly unique anymore.
What has surprised me is not how much I use AI, but the way I use it.
I rarely expect AI to tell me what is true.
More often, it gives me something to react to. Sometimes I agree. Often I do not.
A sentence that feels too confident.
An argument that skips a step.
An explanation that sounds reasonable but doesn’t quite convince me.
This happens repeatedly. Then the conversation continues.
The result is rarely the first answer it gives me. In fact, some of the most useful conversations are the ones where the first answer is biased in an interesting way.
How I Met AI
Looking back, my first encounter with AI was accidental and somewhat dramatic.
It was the spring of 2023, only a few months after ChatGPT became publicly available. One of my instructors included a sentence in the syllabus that read:
“Students should not submit assignments or essays generated by large language models.”
A course syllabus, spring 2023
Ironically, I missed that class because of a schedule conflict. Curious about what a large language model was, I opened ChatGPT for the first time.
My first impression was not impressive. AI at the time felt a little… dumb. It could generate paragraphs, but complete essays often sounded repetitive and generic. It could not search for information, and fabricated references were surprisingly common.
I learned my first lesson very quickly: a confident—or even convincing—answer was not necessarily a correct one.
Still, I kept using it, but for a limited range of tasks.
By the fall of 2023, when I was preparing my first conference submission, some faculty members encouraged students to use AI for language polishing. By then, I had already been experimenting with it for a semester.
At the time, I thought I was learning how to use a new tool.
Why I Started Using AI More
People often ask why I spend so much time talking to AI instead of talking to other people.
I must admit that there are disadvantages to using AI. It consumes energy, and it should never replace conversations with real people. It may contain bias or misinformation.
But every way of thinking comes with a price.
You cannot always share every thought with a friend—at least, I can’t. Sometimes my interests don’t align with those of the people around me. Sometimes a conversation turns into an argument about something I never intended to argue about. And sometimes, whether intentionally or not, people begin to see you differently because of what you say.
We always say we like disagreements, but we always want to win an argument.
As for bias, I see AI much like browsing the internet. We are exposed to biased, incomplete, or even false information every day through social media, news, and countless websites. What matters is not avoiding all of it—that is impossible—but learning how to fact-check sources, dig deeper, and separate useful information from the noise.
I think the same principle applies to AI. It should never be treated as a perfect or unquestionable source. Like any other source of information, it works best when we remain willing to verify, challenge, and think for ourselves.
Principle 01 Never trust AI. Always check the sources and cross-check the information.
Keep Digging
One thing I have learned about myself through countless conversations is that I rarely accept an argument simply because it sounds convincing.
Instead, I tend to dig a little deeper. Under what conditions would this argument no longer hold? Are there exceptions? What happens if we push it to an extreme case? I often find myself testing a theory, a hypothesis, or even a legal principle by asking what happens at the edge rather than at the center.
One conversation that has stayed with me was about the Fourth Amendment and the idea of reasonable suspicion. I wasn’t trying to learn the legal definition for an exam. I became interested in a much simpler question: what does “reasonable” actually mean?
We started with ordinary examples, but I kept changing the facts. What if the officer only saw part of the event? What if there were two equally plausible explanations? What if the person’s behavior was unusual but perfectly legal? At what point does a suspicion become “reasonable” rather than just a guess?
At one point, I asked whether someone walking away after hearing police sirens should create reasonable suspicion. Then I kept modifying the story.
What if the person had headphones on?
What if they were running because they were late for work?
What if the neighborhood had a history of avoiding police for reasons unrelated to crime?
We spent the better part of an hour discussing what sounded like a very simple question: if two innocent explanations and one guilty explanation all fit the same evidence, what makes one interpretation “reasonable” enough for the law?
By the end of the conversation, I realized I was no longer trying to understand the Fourth Amendment itself. I was trying to understand how people reason when the answer is uncertain. That, more than the legal question itself, was what kept the conversation going.
Without a chatbot, I probably wouldn’t have kept pushing the discussion this far. Real conversations have natural stopping points. People get tired, the topic changes, or someone simply has something else to do. AI doesn’t. It lets me keep testing ideas, changing assumptions, and exploring one hypothetical after another until I feel that I finally understand the reasoning.
Principle 02 Don’t stop when you get an answer. Keep digging until you understand why.
Thinking Through Disagreement
Another reason I enjoy using AI is that it constantly gives me ideas and arguments that sound reasonable at first glance.
I like critique. Not because I enjoy proving others wrong, but because critique forces me to explain why I agree, or why I don’t.
Sometimes I ask AI to argue from the opposite perspective. Sometimes I ask it to defend a position that I disagree with. Occasionally, I even ask it to critique my own writing before anyone else reads it.
I have found that disagreement is often more productive than agreement. Agreement usually ends a conversation. Disagreement keeps it moving.
Principle 03 Don’t ask AI to agree with you. Ask it to challenge you.
A Mirror of Thinking
Over time, I started noticing patterns.
I often questioned assumptions before conclusions.
I repeatedly asked where an argument stopped working rather than where it succeeded.
I preferred changing one condition at a time instead of replacing an entire theory.
None of these habits were taught by AI. They came from my earlier statistical training.
Looking back, I learned as much about my own way of thinking as I did about the topics we discussed.
After many of our conversations, I developed another habit. Before closing the chat, I would often ask one final question:
“What do you think I am?”
My last question, most nights
The answer was never about my personality or my background. It was usually about how I approached ideas. One response has stayed with me:
“You rarely stop at the first explanation. You are less interested in whether an answer is right than in understanding why it is right, when it stops being right, and what assumptions it depends on. You don’t seem to enjoy winning arguments as much as understanding how arguments are built. You treat ideas as things to be tested rather than accepted. AI didn’t teach you those habits—it simply gave you a place to notice them.”
I don’t know whether that description is completely accurate. But I know it made me pause.
Perhaps that is what I value most about these conversations. They do not just help me understand a topic—they sometimes help me understand myself.
This also reminds me of the idea of positionality in anthropology. Researchers are often asked to reflect on how their own backgrounds, experiences, assumptions, and relationships shape what they observe and how they interpret it.
I began to realize that something similar was happening in my conversations with AI. I was not only examining the question in front of me. I was also, gradually, examining the person asking it.
Of course, this is not the same as positionality in ethnographic research. AI cannot fully know my social position, experiences, or relationships simply from a conversation. But it can reflect patterns in how I reason. That reflection gives me another opportunity to ask how my own position shapes the questions I ask in the first place.
Principle 04 Every conversation reveals something about the topic—and something about yourself.
Iteration
Very little survives the first draft.
Most of the time, I still write the first version. AI helps me explore possibilities, challenge assumptions, reorganize ideas, and polish the language. What follows is a long series of revisions.
Building my personal website was the best example.
When I first started, I thought the challenge would be technical—learning HTML, CSS, and putting everything online. It turned out to be something completely different.
The first version looked clean, but it didn’t feel like me. The layout was too generic. The homepage introduced my background, but not my personality. So we kept changing it.
One evening, we spent nearly an hour discussing the structure of the homepage. Another time, we went back and forth over whether my research should have its own page or be integrated into a broader narrative about my work. Later, we decided to postpone the entire Professional Work section—not because it was unfinished, but because it did not yet reflect the standard I wanted for the rest of the website.
Gradually, the conversation shifted away from web design. We were no longer talking about fonts, layouts, or navigation. We were discussing what kind of scholar I wanted to become, because that answer influenced almost every design decision.
That is how most of my work with AI evolves.
A paragraph moves. A sentence disappears. An example no longer works. A conclusion becomes a question. Sometimes we remove more than we add. Sometimes we return to an earlier idea after realizing that the newer one was not actually better.
The final version rarely belongs to either the first draft or the first response.
It emerges through iteration.
Principle 05 Never fall in love with your first draft.
Why “CoThinking”?
I never liked the idea of calling AI a “copilot.”
A copilot implies that someone already knows the destination.
Most of my conversations begin before I know where I am going. They begin with curiosity.
“CoThinking” feels like a better description. It is not about sharing responsibility for an answer. It is about sharing the process of exploring a question.
Sometimes we reach a conclusion.
Sometimes we simply discover a better question.
Either outcome is worthwhile.
I don’t think AI has fundamentally changed the way I think. Long before AI, I enjoyed asking questions, reading widely, and trying to understand why things worked the way they did.
AI cannot replace reading. It cannot replace my judgment. It cannot replace the statistical training that taught me to question assumptions, distinguish correlation from causation, or think carefully about evidence.
What it has changed is the scale of that process. It allows me to ask more questions, test more ideas, explore more perspectives, and iterate more often than I could on my own.
AI does not think for me.
Perhaps that is what CoThinking really means.