Lesswrong disclaimer
This is a link post to a living document. What’s below may be an older version. Click link for latest version.
Also, I increasingly find it a waste of time to discuss such ideas on lesswrong, so don’t expect a very favourable response from me unless you quickly prove you’re acting in good faith (as opposed to self-interestedly protecting your EA/LW funded policymakers or whatever). I think we need a mass movement to protest against the US intelligence community which will accelerate towards ASI by default. Using persuasion alone won’t be enough. I want to empower the Big Mob as Vitalik puts it, and disempower companies and govts building ASI.
2026-03-01
Petapixel cameras won’t exist soon
Disclaimer
Quick Note
Contains politically sensitive info
Huge amount of background context on why I care about petapixel cameras
I have often speculated about the idea of a world with zero privacy for everyone.
Not one where only govts and companies can surveill individuals, but also where individuals can surveill govts and companies, and individuals can surveill each other. The platonic limit for this is a world where everyone is on public livestream 24x7x365
The most immediate motivation for me to study this is to help enforce an AI pause.
Currently govts can surveill semiconfuctor manufacturing facilities and GPU compute clusters.
However, individuals cannot directly surveil these facilities as easily. Individuals are currently at the mercy of whether the govt gives them access to this information or not. Both US and Chinese intelligence agencies are not under full democratic control.
Also, even govts may not be able to surveil these facilities in the limit if a) the existing stockpile of household GPUs is used to coordinate a distributed training run (using Tor and monero payments) b) continued algorithmic progress allows a small number of private individuals to use their existing stockpile of household GPUs and build ASI without building a large compute cluster
I also have other motivations for wanting to live in a world with zero privacy for everyone. Not discussed in this post. All those motivations are less urgent than the immediate motivation of trying to enforce an AI pause.
Govts already have a large variety of methods they can use to surveil even smaller or distributed compute clusters.
Spies
Stockpiles of zerodays that affect almost all computers on Earth
Agreements with Big Tech that give them access to lots of private data
Analysis of the entire public internet
Military satellites that fly at LEO and do facial recognition
Drone swarms (10,000 drones is enough to surveil a city and do facial recognition)
Gigapixel cameras flown by aircraft or stationary at ground level
A number of checkpoints and identity databases embedded into society, that make it hard for individuals to function in society. Example: license plate registration and highway checkpoints, CCTV surveillance of city centres, bank account registration, etc
Currently, individuals do not have access to many of these surveillance measures.
Individual whistleblowers can leak info to public only at actual risk to their life, such as Edward Snowden
Individuals do not have access to private Big Tech data
Individuals can increasingly crawl, backup and analyse the public internet, however this is still expensive and only a few groups in the world are doing today. It is easier if you narrow your attention to a small subset of the internet.
Govts have much larger zeroday stockpiles than most private hacker groups. Private hacker groups continue to exist but aren’t as well resourced, and finding zerodays is expensive in terms of developer time.
Commercial satellite footage is not allowed below 10 cm resolution that enables facial recognition. It is easy for a govt to shut down private satellites violating this law.
Many zones are declared no fly zones for drones. It is easy for a govt to shut down private drone swarm operators
Gigapixel cameras are being incresingly only worked on by military, and may not be as available to general public soon.
Identity databases are increasingly being sold on the dark web to individuals, and private investigators increasingly make use of them.
A brave one-off individual could operate some of the above tech, leak info to the public, and then flee the country or risk imprisonment. However this may or may not be a long-term sustainable way of ensuring individuals can do surveillance without govt involvement.
If the govt itself is the one trying to accelerate ASI, then many of the above measures fail. It may be better to live in a world where individuals can surveil ASI projects directly without having to involve the govt.
Petapixel cameras won’t exist soon
Gigapixel cameras allow you to do facial recognition (0.1 metre resolution) from a distance of 1-10 kilometres. You can fly them via aircraft/helicopter or place them at ground level.
For example, here is a 120 gigapixel photo of New York City
If govts don’t want individuals to have access to this tech, it is currently not that difficult for them to lock down the supply chain for this and prevent individuals from having access. (Not sure, this is my best guess, can read more.)
If an individual wanted to use such cameras anyway, in defiance of the govt, they would have to do from outside the borders of the country they want to surveil. You would then need to surveil at 0.1 metre resolution from a distance of 10,000 kilometres. This requires not gigapixels or terapixels, but petapixels.
Despite huge advances in sensor array tech driven by Moore’s law, I don’t think humanity is anywhere close to inventing petapixel cameras in the next 10 years. (Unless ofcourse ASI itself invents them, all bets are off the table then.)
You can’t expect anyone to “quickly prove they’re acting in good faith,” that’s a highly unreasonable bar.
Okay. I think what I want is feedback on tactics, not strategy. I don’t want to debate why AI pause political movement is required, yet again. I don’t want to debate why US intelligence community will accelerate development of ASI, yet again. I can debate details of how gigapixel cameras work.
And I don’t have time for actual “proof” yes, I’m just gonna make a guess. Which could involve me projecting stuff based on reference class of similar comments or similar people commenting.
I don’t think you understand what a “pixel” is.
If you want to see a 1 km x 1 km area at a resolution of 0.1m, then you will need 10 000 x 10 000 = 100 000 000 points on the image, AKA 100 megapixels. This is (ideally) independent of technology. You can walk around and take fifty 2 MP pictures and stitch them together, you can fly a drone a few hundred meters up and take a wide-angle shot, or you can fly a satellite overhead and take a picture from space. The distance doesn’t matter.
From that, it’s a simple extrapolation that it’ll (again, ideally) take 100 MP/km^2 * 778 km^2 = 77800 megapixels = 77.8 gigapixels to surveil the land area of New York City to that level of detail. Again, that could be from an array of low-resoultion cameras, a wide-angle camera nearby, or a telephoto camera far away.
(Also, a brief search suggests facial recognition needs about 3mm (0.003m) resolution to identify individuals, not 100 mm (0.1m))
A 1 petapixel camera could cover an area of 3162 km x 3162 km to that resolution, or roughly the entire United States in a single snapshot (ignoring practicalities like the curve of the earth, of course). It could also be used to count someone’s nosehairs if you set it up differently.
I understand all this. I am assuming a constant angular field of view, let’s just assume 120 degrees for now. I assuming a single photo from a single camera placed at a single location covering 10^15 pixels. I am not talking about multiple photos stitched together, or moving the camera around, and so on.
(Yes, the camera will necessarily have multiple sensor arrays and internally stitch the data together anyway)
And yes a petapixel camera with 120 degree (or some other large field of view) could cover the United States at 3 mm resolution.
I am not sure if we are actually disagreeing.
I am saying someone should be able to place a camera outside US borders and yet be able to do facial recognition of people inside from thousands of kilometres away.
I asked this question to Opus
It works out to 1 photon per year.
This math assumes a raw pixel with no optics, which is an absurd way to build a camera. With a 1m lens at 40km, you could get ~10⁵ photons per second (13 OOM better).
The problem here is the diffraction limit. At the 2,500 km ranges discussed, 3mm resolution requires a single aperture of ~500m or a constellation of ~7,500 JWST-scale telescopes tiling the coverage. Optical interferometry could theoretically reduce the count, but requires maintaining satellite relative positions to within a wavelength of light.
Thanks this comment is useful!
There’s no law of physics that prevents humanity from building either of these things. I’m just pessimistic about the engineering advancing to the point that we can build this in next 10 years. (without help of superhuman intelligence, that is)
Yeah, seems like ASI can be achieved well before the monitoring can be built.
Oh right, yeah, that makes sense.
The constant angular field of view is the disagreement. A camera in the mid-gigapixel to low-terapixel range could cover one city by using an appropriate lens at an arbitrary distance (including space).
Any sensor finer than that would either cover substantial amounts of “boring” area (e.g. nature preserves, agricultural areas), or increase the resolution beyond your target.
I am still not clear where we are disagreeing, sorry.
What do you think is the bottleneck to building a petapixel camera that lets you do facial recognition from outside national borders? I don’t think you can simply stitch a bunch of gigapixel cameras together and achieve this.
A camera that can do facial recognition from outside of national borders doesn’t need to be a petapixel one. A mid-gigapixel camera with good optics can cover an entire city at once (or at least it could if it wasn’t for all the buildings in the way).
The main barrier to petapixel cameras is that they don’t serve your goal of full public monitoring (regardless of whether it’s by the government or by everyone individually).
This is technically true. But yes, if you had the tech to build this it would also become trivial to built a petapixel camera too (for someone who can afford it). The hard part is doing 0.1 metre resolution from a 10,000 kilometre.
Thanks for this exchange btw, I guess in future I could be more precise.
Why?
Assume we had the tech to manufacture petapixel cameras, and individuals worldwide could purchase them (i.e. a govt couldn’t just lock down the supply chain). Why does this not eventually to a world with zero privacy for everyone?