How Does AI Work? An Introduction to Model Thinking

How Can I Become AI-Literate?

AI Literacy is two things: a) knowing how to use AI and b) knowing how AI works.

To achieve (a)—knowing how to use AI—there’s no substitute for hands-on experimentation. Start using AI tools, and explore them yourself. Because AI tools are evolving rapidly and new applications are constantly emerging, there aren’t yet “experts” on how to use every tool. It’s only through exploration and trial that we learn which tools are suited to which tasks.

But what about (b)—understanding how AI works? Where should you start?

The key is to think in terms of models. AI is based on models, but so is science. Good science relies on models to explain, predict, and test ideas about how the world works. This post is an introduction to “model thinking” and how it applies not only to AI but also to other areas of life, such as athletic training.

Why Do Most Exercise Regimes Fail?

In an earlier post, I shared a video by Olympic marathoner Frank Shorter, who won the 1972 Olympic marathon and is widely regarded as one of the best middle-distance runners of his time. I’m linking it here again because it’s well worth a watch—just five minutes, packed with insights. The video, Frank Shorter’s “Exercise Recommendations for Older (or Younger) Adults,” offers a few key principles that illustrate model thinking.

Shorter touches on several points, but I want to highlight two distinct models of athletic training. These models don’t just apply to physical training but extend to learning and skill development in general. I’ll save that connection for another post.

The Training Effect and Model Thinking

A central idea Shorter discusses is the “training effect,” which is the adaptation process our bodies undergo with sustained training. He notes that it typically takes about two months for the body to adapt to a new exercise regimen. Most people, however, give up before seeing significant results. What does this have to do with model thinking?

Most people’s mental model of exercise is linear. This is shown in the left figure. When we start exercising, we expect immediate, steady improvements in performance: the more effort we put in, the greater the boost in performance we expect. In this model, effort (e) is the slope of a straight line. However, this linear model of progress doesn’t match how our bodies actually respond to training.

In the right figure, we see a different model of athletic training and physical skill development. Each body has a stable equilibrium state—a point where it naturally settles. You can apply all sorts of efforts to shift this equilibrium, but immediate changes are unlikely. Shorter’s insight is that effective training requires sustained “stress” and “activity” over a period. Depending on factors like training intensity, your body type, and your starting state, this period could last around eight weeks, after which your body “jumps” to a new equilibrium—a measurable increase in performance.

Two Models of Effort and Performance

To summarize, I’ve presented two models of the relationship between time, effort, and performance:

1. Linear Model - Based on the assumption of steady improvement, where increased effort produces a proportional rise in performance.

2. Equilibrium Model - Based on states, where performance remains stable until enough consistent effort is applied to reach the next stable state.

Determining which model more accurately describes our reality is an empirical question. Through experiments, data collection, and revision of hypotheses, we test our assumptions. This process of using models to frame and test our understanding is essential to model thinking.

To become AI literate, embrace model thinking. The best way to understand AI is to view it as a set of models, just like in science or athletics. By thinking in terms of models, we can better grasp how AI functions and learn to apply it in various contexts.