The article argues, but does not provide evidence. It specifically says the UN used surveys immediately after saying surveys don't work here. There's no validation that estimates from satellite imagery are better than the methods PNG used.

The fact the UN didn't adopt this report would certainly be an argument against it.

China is the best example, its estimated that their population is off by entire countries in some statisitics, either through disppeared girls, hidden covid deaths, local economic fraud. There is also no independently verifiable group in China and is actually explicitly banned to use non-government methods.

Why is the default assumption "just trust them bro, why would they lie!"?

That's not scientific. There's no verification or validation of data.

Your default assumption should be to question authority, especially if authority claims sole dominion over claims of fact, like "this is our population, because we say so."

They are humans with power, therefore they lie. If you don't have accountability feedback, you can never, ever check those lies, so you rely on proxies and legitimate models.

I highly recommend researching proxies you understand and can trust, and developing an understanding of the models that exist, and how to estimate confidence over a bounded range of values.

I don't think China has only 500 million people - that's a little silly. But I also don't think they have 1.4 billion, either, especially since one of their main justifications for that is "hey, we have this many phone accounts!" - their population control policies, their population decline, their cultural preference for male children and infant femicide, and so on don't jive with simple models of population growth based on human population growth constraints. If there's a deviation between properly error bounded models of populations over time in the hundreds of millions over the highest reasonably bounded value, something is suspicious.

You can take your reasonably bounded model and correlate with proxies - if the verifiable evidence supports the model over the claims, you can be more confident in the model than the claims.

Reliable proxies that can't be faked are difficult, and better models are going to be needed in the future as we get into AI slopageddon territory, where you can trivially fabricate entire identities and histories for billions of nonexistent people, even establishing social webs and histories for all of them, statistically indistinguishable from real people.

To perform a census, you need models constructed from verifiable data and first principles reasoning, with Bayesian certainty attached to each and every contributing factor, and then you need to set probabilistic bounds based on known levels of variability in things like population growth rates. Once you have an upper and lower bound, you can assign a certainty measure to the official claims - something like "this has a .01% chance of being true" - that's a good indication that reality diverges from those claims. It's not proof, it doesn't give you 100% certainty that some other number is precisely the case, but it's evidence.

The US government varies wildly in population counts, too, depending on which party is in power, which locales are being counted, the intent of the count, such as census, or estimation of population of illegal immigrants versus legal immigrants, etc. This is why census laws in the US forbid estimations or models or extrapolations; you need firsthand, auditable data collection, or fuckery occurs. The 2020 census was corrupted and then this was discovered by media and third party verification, for example. If you don't have a free press, things like that don't ever get revealed and confirmed, and authority is never held to account (in theory. In principle. In practice, power is rarely held to account anyway.)