Engagement vs Empowerment Algorithms

Pair of opposed design philosophies for software that competes for attention.

Engagement algorithms (today)

Optimized for time-on-app as the proxy metric.

Feed ranking maximizes scroll depth, return visits, dwell time
Notifications calibrated to maximize re-engagement frequency
Hooks (per Eyal-style design) chained to make leaving feel costly
Outcome variable: revenue per user, indirectly via ad impressions

The cost David named

"Free, cheap sources of pleasure — and the most destructive part is that not only is it freely available, but you don't have to exert any effort to get it. So if you can get a very high source of dopamine without doing anything, why would you ever exert more effort to chase something that is, in the short term, less dopamine?"

This is the structural problem: engagement algorithms shift everyone's dopamine-to-calorie ratio toward "low effort, high dopamine," which raises the marginal cost of any action.

David also flagged: "80, 90% of high-schoolers and children are addicted to social media, brainwashing their brains. It's jeopardizing their human brain system." Jacob: "And honestly, I think that's the single biggest problem."

Empowerment algorithms (the alternative)

Optimized for what you do after closing the app.

Filter content to remove "low vibration" stuff (per Jacob's local-LLM project)
Surface things that prompt action, not consumption
Auto-close tabs / sites flagged as unproductive
Outcome variable: cumulative user agency over time, however measured

Jacob's Empowerment Algorithm App is one implementation. Harrison's "pause" app is another (mentioned in passing).

The keystroke-shortcut interface

A small but interesting design detail Jacob describes:

"I just put a thing today where I would just go on with my regular day browsing the internet, and then just by keyboard shortcut, I can say, okay, this website is unproductive. And then it will categorize that website as unproductive and pull all of its metadata to train an underlying model on what unproductive looks like, what productive looks like."

Two-channel feedback: explicit tag (the shortcut) plus implicit features (page metadata, time-on-page, etc.). After enough labels, the model generalizes.

The "key" mechanic:

"For unproductive websites, it just instantly closes them. As soon as you open an unproductive website, it just closes the tab."

AI as part of the solution, not just the problem

Jacob's interesting twist:

"One of the reasons I'm really optimistic about AI is I think I've noticed it affects my social media behavior, because now I build stuff instead of scrolling as much, because the dopamine loops are so accessible with AI. And the friction level to build stuff is so decreased, like, I don't have to be ready to go super hardcore intellectual activity all the time. I can vibe-code when I'm fried and still be pretty good."

So: AI not just enabling better recommendation/engagement systems, but creating a competing dopamine source that happens to also be productive. Building stuff as the new scrolling.

Why this matters

Engagement vs. empowerment is the proxy battle for Vision for the World. If the dominant algorithms shape attention toward consumption, sparks of motivation never get to action. If they shape it toward agency, the same population becomes capable of collective intelligence work.

This is downstream of Wu Wei Disclosure and upstream of Level 1 (self) of the coherence ladder.

Wu Wei Disclosure — the underlying mechanic
Empowerment Algorithm App — the implementation
Sparks of Motivation — what better feeds let through
Harrison — built "pause," a sibling app