[colinsane]

> If a map were to represent the territory with perfect fidelity, it would no longer be a reduction and thus would no longer be useful to us.

if a map perfectly represented the territory it would be very useful to everyone mentioned in this article. with a perfect representation of the territory, you can just simulate different strategies and deploy the best one. no need for risk management: your perfect map allows you to eliminate all risk.

a perfect map might not be possible if you’re an embedded actor, but that doesn’t mean one shouldn’t pursue the best possible map. the rest of the article is about recognizing flaws in your map. and guess what: when you identify a shortcoming in your map — which the author does and recommends others do — that’s identical in an information sense to just building a more detailed map.

> improbable and consequential events seem to happen far more often than they should based on naive statistics.

the author has quantified some thing (“consequential events”) and then stated that this thing occurs within some data set more frequently than would be consistent with that very dataset. i get what he’s trying to say, but when he phrases it this way it’s just a simple contradiction with an easy way out: build better maps.

so, yes: the map is not the territory. if you build a map without complete knowledge of the territory (which is the majority of maps), then it has unknowable error bars. but maps are unavoidable: you can either explicitly follow a map, or implicitly follow one. Warren Buffet uses a map when making sense of the world. is it good, or bad, that the map he follows is accessible to only a single mind and has not been digitized and shared more widely? the biggest case to be made for ditching digitized/formalized maps is because this allows you to retain more hidden information, which is the basis for gaining an edge in financial markets. but the author didn’t really argue the futility of maps based on embedded actors — it was mostly an argument that too many people are engaged in map-making without first understanding the boundaries of the territory. and that’s no argument that informal maps are intrinsically superior to explicit maps.

[pdonis]

> with a perfect representation of the territory, you can just simulate different strategies and deploy the best one. no need for risk management: your perfect map allows you to eliminate all risk

Such a perfect map is impossible because it would need to have infinite accuracy, and its consequences would not be computable.

> if you build a map without complete knowledge of the territory (which is the majority of maps)

Which is all maps. Complete knowledge of the territory is impossible.

[AA-BA-94-2A-56]

To agree with you and further what you've said...

Even if the map is infinitely perfect, your understanding of it is imperfect. Because your mental model of the map is the actual map you follow, it is the actual map that you follow. Because you are imperfect, even if your map is perfect, the map you follow is imperfect.

Therefore, the saying is always true.

[colinsane]

all maps of a physical territory, yes. that’s a consequence of ħ, at the very least. but once you’ve encoded this best of all possible maps, the uncertainties in the map are quantified (like probabilities). you no longer have “unknown unknowns” so you can just carry through all the PDFs and create a true risk assessment. and it’s these unknown unknowns which were, AFAICT, the heart of this article.

[pdonis]

> all maps of a physical territory, yes. that’s a consequence of ħ, at the very least.

Not really. If it's a consequence of anything, it's a consequence of relativity and the finite speed of light: you literally cannot know all of the information you would need to construct the "perfect map" you describe, because you only have access to information from a limited portion of the universe.

Quantum uncertainty just makes it worse, since even in the finite portion of the universe you have access to, you cannot know the exact state.

> once you’ve encoded this best of all possible maps

There is no such thing. Even leaving aside the relativity issue I raised above, non-commuting quantum observables are incompatible, so there is no single "best possible map" taking quantum uncertainty into account.

> you no longer have “unknown unknowns”

This is impossible. First, information outside your past light cone is always "unknown unknowns". Second, even if we limit attention to the data in our past light cone, someone can always measure a quantum observable that doesn't commute with the ones you have data on and invalidate your current model.

[astrange]

You can’t perfectly calculate the future of any situation ahead of time if you’re part of it, because it becomes chaotic.

This is part of the “computers are going to become AGI and enslave us” scam, they show people proofs AIXI is an optimal planner and don’t mention that it doesn’t include itself in the plans in order to make them computable.

If it was possible you could beat the stock market.

[tomcatfish]

You're aware that the AI folks aren't trying to build AIXI, right? I find it hard to believe you'd thrash a theoretical infinite-compute chess model as silly this hard; it's a useful model as a boundary case.

[thfuran]

>If it was possible you could beat the stock market

You say that like it's some natural law that that can't be done. But that's not the case.

[astrange]

There's some natural laws in the area. You can't win infinite money forever by one piece of good information, it's going to get priced in or the other traders are going to get wiped out.

Rational expectation/efficient market hypothesis works better than people think: https://someunpleasant.substack.com/p/economists-do-they-kno...