Brandon Marcus

There is an irresolvable tension between the practice of predicting human behavior and the belief in free will as part of our everyday life. A healthy degree of uncertainty about what is to come motivates us to want to do better, and it keeps possibilities open. The desire to leave no potential data point uncollected with the objective of mapping o... See more

And that’s what real taste is: a deep internal coherence. A way of filtering the world through intuition that’s been sharpened by attention.

The ability to choose what matters in a world drowning in what doesn’t.

I have elsewhere called these artificial intelligence systems “ looms ,” and I think they are best understood as external attention systems —just as writing can be called an external memory system. Recommendation engines, predictive feeds, and language models perform the various functions of attention: filtering, noticing, selecting what matters fr... See more

So: we have before us two distinct phenomena, both called “artificial intelligence,” both transformative, both rooted in attention, but profoundly different in their structure, effects, and affordances:

• The first consists of systems that externalize and perform the operations of attention—filtering information, reducing uncertainty, detecting patte

... See more

To be clear about the mechanism: our society selects for the affordances of a medium—speed, ease, efficiency—not for its effects . And it is the effects of literacy that hold its civilizational value. This is the critical point: those deep cognitive and ethical capacities are not being selected for. They are not easily monetized or optimized.

Productive friction has three main principles:

1. Immediate feedback: You understand when you've failed and can see why.


2. Cumulative learning: Each attempt builds your reference library.


3. Transferable principles: The specific teaches the general.

today’s machine intelligence structures the world to produce predictable data, attention, and behavior. Through continuous modeling and subtle feedback, human action is rendered legible and brought under algorithmic management. This marks a second enclosure — not of land, but of the cognitive commons itself.