A Working Picture of Intelligence

To train intelligence, you first need a picture of it—and not a fuzzy one. As long as "intelligence" stays a single vague quality, there is nothing concrete to aim training at. So it is worth starting somewhere more grounded: describing the recognizable ways intelligence actually shows up in people.

What intelligence looks like

Think of the things we point to when we call someone intelligent:

• Someone who can understand and dissect a complex argument.
• Someone who's good with words.
• Someone who keeps track of everything in a conversation.
• Someone who comes up with interesting, unique, and correct ideas.
• Someone who thinks fast.
• Someone who has good ideas come to them quickly—the solution surfacing mid-game, or mid-problem at work.
• Someone who takes in a whole situation at a glance.

None of these is "general intelligence" in the abstract. They are specific, recognizable, and—this is the point—each one comes apart cleanly into a small set of underlying cognitive processes.

The same parts, again and again

• Understanding and dissecting a complex argument draws on relational reasoning (tracking how the claims, words, and ideas relate), working memory (holding the pieces at once), and vocabulary—the last of which you accumulate over a lifetime.
• Being good with words draws on semantic access—how fast and how widely you retrieve words and meanings—together with the vocabulary you have to retrieve from, and relational reasoning to assemble it.
• Keeping track of a conversation is working memory, close to undiluted.
• Coming up with interesting, unique, and correct ideas draws on the breadth and reach of your semantic access (the novelty), the store of knowledge you are recombining, and relational reasoning to keep the result sound.
• Thinking fast is those same verbal processes—semantic access, relational reasoning—run at speed. Which names the parameter sitting underneath all of it: processing speed, the rate at which every other process runs.
• Having good ideas come to you quickly—the solution that surfaces mid-game—draws on the speed of that retrieval, on domain expertise (the patterns you have stored over time), on how quickly you take the situation in, and on a fast relational check that the idea fits.
• Taking in a whole situation at a glance is that last process on its own: perceptual processing speed—how much you register, and how fast—feeding everything downstream.

A small set of processes, in two kinds

Across every demonstration, the same short list keeps reappearing: relational reasoning, working memory, semantic access, and processing speed, with knowledge—vocabulary, expertise—woven through.

And they sort into two kinds.

Four are processes you can train directly:

• relational reasoning
• working memory
• semantic access
• processing speed—less a separate kind of intelligence than the rate underneath the others: when operations run faster, they finish before working memory fades, so more is held and more is computed at once.

The fifth is different. Knowledge—vocabulary, the expertise behind a fast solution—you do not train directly. It accrues as a return on the others: the better you reason, the more easily you learn, and the more you come to know over a lifetime. This is investment, and it is the only way crystallized knowledge enters this picture.

Why this matters for training

With any single demonstration, you can improve by sheer practice—argue enough and you will get better at arguing. But that practice tends to lift only the thing you practiced. What carries across all of them is the layer underneath: the shared processes every demonstration is built from. Strengthen those, and the gain shows up everywhere they are used.

Why those processes are shared in the first place—and why training them transfers while task-specific practice does not—is the subject of Process Overlap Theory and Does Cognitive Training Transfer?.