@HowToPrompt__ - Compression Progress Theory Explains

Open original

@HowToPrompt__ on X The post argues that beauty, humor, curiosity, boredom and musical hooks all come from one mechanism: the brain compressing incoming information and getting a dopamine reward when it finds a pattern. It credits a paper by AI researcher Jürgen Schmidhuber, written to give artificial agents intrinsic motivation, and frames the findings as unsettling for anyone who thinks human experience is beyond machines.

  • Predictable input feels boring because there is nothing left to compress, while pure randomness can't be compressed at all and gets tuned out.
  • The reward peaks in the middle: input that looks complex but suddenly resolves into a hidden pattern, which the post says is what we label curiosity, humor or beauty.
  • The underlying theory was written by an AI pioneer to build intrinsic motivation into machines, not to explain human psychology.

The core claim is that a single idea, the brain as a compression engine that rewards itself for finding patterns, explains beauty, humor, curiosity, boredom, and why babies explore and music has hooks. The mechanism is simple. Predictable input is already compressed, so there's nothing left to learn and you get bored. Pure noise can't be compressed at all, so you tune it out. The reward sits in between: something looks complicated, then suddenly a pattern clicks into place, and the brain releases dopamine. The author calls that jolt "compression progress" and treats it as what we experience as curiosity, humor, or beauty. A joke that seems senseless until the punchline makes everything fit is a clean example. So is a painting whose hidden symmetry pulls the chaos together. The article also makes a specific attribution claim, and it's the one most worth checking before you rely on it. It says the idea came from an AI researcher working on intrinsic motivation for artificial agents, not from a philosopher or neuroscientist trying to explain people. That framing is what gives the piece its "unsettling" tone: the same principle meant to make a machine curious may be what makes us curious. The post names no paper, author, or year, though, so treat the attribution as unverified until you find the source. The idea of compression progress as an intrinsic reward does have a real lineage in AI research on curiosity-driven learning, which makes the claim plausible, but plausible isn't the same as confirmed for this specific paper. The argument's weak point is the leap from "this principle explains a lot" to "our deepest experiences are just a mathematical reward." Explaining why a joke lands is not the same as showing the brain's reward circuitry runs on compression rate, and the post offers no experiments, effect sizes, or neural data. Plenty of the examples are also unfalsifiable as written: any pleasing thing can be redescribed as "a pattern you found," which makes the framework feel more explanatory than it may be. Worth saving if you want the idea and the original source to chase down. Skip the claim that machines now appreciate art on the strength of this post alone.

Ask about this article