What does it actually mean to think and reason?
I came across this definition of a ‘Reasoning AI model’ recently while reading the book “How to AI” by Christopher Mims.
According to the book, a reasoning model is a type of AI/large language model that talks to itself in order to “reason” through a problem.
While a technical definition of how an AI model ’thinks’ is more complicated, it still made me think - Do we also solve our problems by talking to ourselves? What’s the general process of reasoning for humans? Is it just talking to ourselves? Is it asking “Why?” to get to the root of a problem and then finding a solution?
The dictionary meaning of reasoning is: The act of thinking about something in a sensible way.
The meaning still has a problem that any other word in a dictionary usually has, at least for me. It doesn’t explain what’s ‘sensible’.
After digging a bit, here’s what I think makes thinking ‘sensible’. It’s a way of thinking that:
- is evidence-based
- uses our larger knowledge base as needed
- is open to correction
- factors in any uncertainty
and so on.
In a world where we have to act quickly and are heavily influenced by biases, it can feel intimidating to actually reason well. The act of reasoning, or perhaps thinking, can also be influenced by how we “feel” at that moment. Imagine thinking clearly when we are frustrated or extremely happy. Chances are we will arrive at poor decisions that we regret later.
Here are a few thoughts I gathered while trying to dig more into this:
- Adding a simple layer of “do I have enough evidence to support this reasoning” can save us a lot of time in decision making and improving quality of our decisions. For instance, if we don’t feel confident in taking up a new project, asking ourselves for evidence and probing why we can’t do it might give us an answer rather than getting stuck in the loop of whether we should do it or not.
- Bayesian updating: Simply put, it’s a way to improve our beliefs after we get new information. Perhaps we can say it’s an antidote to confirmation bias, which makes us believe what we already believe (in some cases despite getting new information). Reasoning something like “I used to believe I wouldn’t be able to take up this project, but now that I reviewed my past 5 projects and all of them were successful, it’s likely I can take up another one.”
- Respect what we don’t know as it’s difficult to fully know what we don’t know. But respecting that there’s uncertainty and embracing a chance for something to show up off-script helps us prepare, and perhaps makes a setback more manageable.
Always respect what we don’t know as it’s difficult to fully know what we don’t know.
While there are a lot more thinking tools and mental models to explore (and we probably should), this is a good time to pause and think about how we actually reason for complex decisions. Perhaps it’s also nice to watch the reasoning process of an AI model (we can just expand the thinking tasks/process in most of the AI models/tools today, as it happens) and see how we are different. Just a thought!
References
How to AI: Cut Through the Hype. Master the Basics. Transform Your Work. by Christopher Mims (Penguin Random House, 2026).
