Hacker News Reader

Comment by irrational | Hacker News Reader

> while others have criticized what they see as a flippant use of potentially harmful technology: showing how easy it is to access livestream footage and facial-recognition software. Many of these critics encouraged the artist never to make the A.I. he developed public.

Security through obscurity? If this artist can create this AI on his own, then many programmers can create it, even if the artist doesn’t make his code public. I feel like many non programmers don’t understand how easy some of this stuff is to replicate once they’ve seen some else build something.

View on HN

Something to remember is that "critics" generally just means "random people replying on twitter sampled by the article writing to create a narrative".

It's not even a good sampling of what "many non-programmers" think - especially since the single linked critic claims to be a Software Engineer.

Was there really the need for AI here? Geo locate the photo shoot, timestamp it and search for a camera that matches? And then it looks like some simple object classifier boxes from open cv to jazz up the images?

I think the bigger issue, one that is valuable is the discussion about privacy in public space. Or a better way to put it, you have no privacy at all in public spaces.

Hmm, yea privacy in public is an interesting thought. I don’t think its ridiculous to say you should have _some_ privacy, even in a public space. It feels like the expectation is that you can be physically seen, but beyond that I don’t see how someone loses all facets of privacy, unless you’re assuming a surveillance-state.

This adds less than you'd think since people are already tracked by their cellular location. Some off-shoot use cases for this are determining where someone travels and when along with who they interact with by looking at adjacent cellular signals.

You can quickly solve where they would be expected to go in their daily life, what route they'll take to get there, who they will interact with along the way, and when they will be doing it.

Combine this with their credit card purchase history and you can also work out what they're going to buy and how long it will take them to eat it.

I (along with many others who are privacy conscious or are just tech luddites) do not carry my cell phone all of the time.

Albeit noisier, utilizing facial recognition for the same purpose has been in use for a number of years now. They're both identification mechanisms but tracking IMSIs and Mac Addresses gives both better recall and precision than a network of cameras.

For the longest time in urban public spaces, there was a sort of privacy by means of the anonymity processed. You knew that you would never see any of the people around you. That's definitely completely gone now.

You do still legally have an expectation of privacy if you’re having a conversation and at least trying to ensure it’s somewhat confidential.

How do you geolocate a photo shoot and search for a camera that matches? I really don't know what you meant. Assuming you're not talking about phone metadata, of course.

It's phone geo data.

Not sure if Instagram makes the exif data available

No, exif data is wiped from photos by any of the large social media companies and public photo hosts and has been like that for years.

Edited to add comma.

I think you’re missing a critical comma?

”No, exif data…”

I think you'd be right!

Last I heard, Instagram strips EXIF data on upload.

Instagram strips it but people themselves add the location on their own, that you can maybe access through api or at least minimum by scraping.

Otherwise sites like this wouldn't exist: https://iknowwhereyourcatlives.com/

(taken from yesterday's posting of OPsec tools: https://news.ycombinator.com/item?id=32951406)

People geotag their own Instagram posts, you can just search Instagram for photos in Times Square and get loads.

If you only want to do this for a few photos, you just need to (eg) record the Times Square webcams for a couple weeks, then find a couple of recently tagged Times Square photos, make a good guess as to time-of-day and exact photo spot and scrub through your recorded footage from those times until you see someone in those clothes with that pose. It'll take a little while, but with this project you only need to do it a few times.

Keep in mind that the person doing this has all the context from the Instagram account's other photos and captions. Perhaps the same Instagram account made it really obvious what day this trip was happening, which would narrow it down enormously. We don't have that context, but the artist putting this project together has a free choice of which photos to investigate, and in Times Square or in front of Wrigley Field you can surely find some Instagram posts where it's obvious from the surrounding context what day and time it was taken, EXIF data aside.

And that's assuming the artist didn't just hire these people to show up at Times Square at a convenient time for him to press "record" on the webcam.

Yeah, there's no need for "AI" software on this project. From the images, it looks like a straight out-of-the-box use of YOLO v7 or something to make it seem more fancy, but this could be easily be done manually without any machine learning at all (or code for that matter).

> you have no privacy at all in public spaces.

Have we ever?

Nope. Only through obscurity - it used to be difficult/expensive to track somebody’s daily movements. Now it’s relatively trivial to do so.

This is where I have trouble. It obviously feels wrong to do at scale in an automated way but the only difference is that it got easier.

Yeah. I feel the same way. And I realize I facilitate it with my phone and other devices. But still, feels creepy at best and dangerous at worst.

To varying degrees, depending on your definition of privacy.

For example, it used to be possible to go to a public park with a < 5% chance of being in the background of some random photograph taken by a stranger (or 0% depending on how far we go back in history).

Now in many places you can barely walk 1 block without being in the background of someone's iPhone selfie or captured on someone's Ring camera.

You don't know when the image was taken, you need to go through the footage until you find the influencer.

May be this wouldn't be that hard to do manually, but making it with an AI shows that it can be don at scale; whether the author here did it or fake I don't know, but the message wouldn't be the same without the AI part.

Exactly. This is not technically remarkable given what's currently possible. I would actually be surprised is some variation of this isn't already used by law enforcement or security companies. This looks like something a college student might be able to make with something like OpenCV (scaling aside).

> If this artist can create this AI on his own, then many programmers can create it

Yes, the software can't be controlled, the knowledge and skill needed to make it is widespread. But what about the hardware? Production of suitable hardware is very centralized. As more people get burned and shaken by these new systems, I fear we may eventually reach a critical mass where most of the population is eager to restrict availability of new hardware to 'trusted' organizations; corporations and governments.

I don't think that hardware control is the solution here as there are no definitive bounds that can limit what is possible since the answer is often just "let it execute in more time", or "run it distributed over a network of other devices".

yeah - colleague here is outsourcing this kind of engineering for security Fortune 500 clients right now, and has been for 2+ years. There is big business going on right now building and shipping exactly this, as high end security camera installs, as far as I can tell. Its not new at all

> Many of these critics encouraged the artist never to make the A.I. he developed public.

We'll have to wait until OpenAI ethically releases an alternative to "geolocate criminals", as usual. :-)

this is just a time based version of photosynth

Reply on news.ycombinator.com