Amazon-powered AI cameras used to detect emotions of unwitting train passengers | Hacker News Reader

Amazon-powered AI cameras used to detect emotions of unwitting train passengers

wired.com

82 pointsamunozo47 comments

> some say the technology should be banned due to the difficulty of working out how someone may be feeling from audio or video

I see this red herring all the time in reporting. The problem isn’t that it’s unreliable. That will probably change in the future. The problem is, I don’t want to be spied on. Period. The fact that it’s inaccurate only adds insult to injury.

Ubiquitous surveillance in Xinjiang didn’t become more acceptable as their facial recognition tech got more accurate.

With falling costs for NPUs in SBCs and AI PCs, there will be ubiquitous human-activity-sensing cameras and WiFi radios in buildings and desks.

https://www.pcmag.com/news/how-qualcomms-always-on-camera-be...

> The “Always-On Camera” option that debuted with last year’s Snapdragon 8 Gen 1 is now the “Always-Sensing Camera.” This feature keeps a phone’s front camera on, in a low-power and low-resolution mode and available only to the chipset’s Qualcomm Sensing Hub secure enclave, to do basic checks of presence. For example, blanking the display if a second face appears in the background in a possible case of shoulder surfing.

https://www.pcgamer.com/intel-wifi-sensing-raptor-lake/

> 13th Gen Raptor Lake processors. The technology is called Wi-Fi Proximity Sensing (or just Wi-Fi Sensing), and since I'm over at Intel's lab in Haifa, Israel, I've been able to give it a go for myself. It really is as simple as it sounds. You just walk up to the PC and it will wake from sleep to your desktop or lock screen. Then, if you walk away, it'll go back to sleep again 30 seconds after you abandon it.

Upcoming activity detections include breathing rate, which can be a proxy for emotion. Who will own/secure the data?

For those who are uncertain, I think the term "trespasser" is a euphemism for suicides. At a terminus like Waterloo the trains are approaching pretty slowly to stop just short of the buffers, so much less risk.

It sometimes is, but it also includes cable thieves, people trying to collect a ball they've lost (as the article mentions), anti-social behavior from the kind of people who think it's fun to set a signalling cabinet on fire, and potential terrorists/saboteurs.

Do these stations not have platform screen doors? Those seem like the most obvious deterrent to suicides, as well as lost items and people generally getting onto the tracks.

You have different rolling stock departing from the same platform. Trains have doors at different locations. Some trains have them close to the center, some near the edges. There are also trains with slam doors - ones that open manually outside, not slide sideways automatically. Welcome to national rail.

> There are also trains with slam doors - ones that open manually outside, not slide sideways automatically.

Haven't seen one at Waterloo for decades. Could be wrong of course. I miss them.

Not on Waterloo, but Kings Cross had them, which is also mentioned in the article. Waterloo in turn had steam leasure trains departing.

No. Waterloo is a huge station used by a huge variety of trains with doors in different places, shifting to platform screen doors would be the work of decades if it's even possible at all.

Ironically mass surveillance is the only thing I might eventually kill myself over

All of these speculative use cases mentioned in the article and thread, while certainly possible using AI, were always possible before and sometimes done quite well too. So why weren't they used in excess before? A lack of a strong business proposition, its that simple. If you were able to get really clear, actionable information from these metrics, everyone would have been collecting them decades ago. Clearly, the business case is too insignificant to be worth while, but that's not going to stop companies from riding the AI wave and shilling junk as long as they can get in a room with a nontechnical corporate software buyer. There's always good money in selling the latest tool to hammer a nail like one always could have.

> were always possible before

It’s a question of viability not possibility here.

Dropping costs by 95+% dramatically changes things. So, you can “make a strong business proposition” for stuff that was never going to happen with the old method. Consider the economics around scam phone calls when long distance phone calls cost more than 2$/minute. They still happened but it was vastly more targeted and therefore orders of magnitude less common.

Now apply that same logic to facial recognition etc and suddenly new types of surveillance will become commonplace.

Costs were already low and machine learning is not cheaper than just doing something like a principle component analysis or LDA. States using 1990s hardware were already able to determine you didn't already have a drivers license in their database of millions of drivers.

Why is it even important to know the emotional state of passengers? Assuming it's even accurate.

Article covers that: upsells in retail and advertising

Sentiment analysis can be pretty valuable in marketing/propaganda.

That's only official part. Having cameras you can track a lot. 1. recognize people 2. track who is interacting with whom 3. track who personally pays attention to what ads on the screen 4. get some clues about health and wealth. personalized and average 5. if cameras come with microphone arrays, 'for gunshot detection', you can listen selectively. +recognition, remember? 6. Read smartphones' screens.

Now all this in hands of private company. Ready for sale.

This is really it. Private companies are paying lobbyists to fill lawmakers nonsense fake facts selling imaginary public safety value, while a huge amount of marketing value will be reaped by the private company in the hands of this public data.

Super duper use case: Figure out what makes travelers the angriest. Make improvements. Measure how efficient it was.

Less duper use case: Figure out the threshold of incompetence, delays, infrastructure management, overcrowding is likely to cause a riot and try to avoid riots.

> Less duper use case: Figure out the threshold of incompetence, delays, infrastructure management, overcrowding is likely to cause a riot and try to avoid riots.

This is what I was thinking. It's a product marketer's optimization dream: Find out the worst possible product you can make where customers will still (barely) buy it. If you can measure quantitatively how disgusted a customer is, and if you can quantify the level of disgust where he chooses not to buy, then you can find the exact cost floor for your product.

The incentive systems need to be aligned for us to start seeing these super duper use cases

Another thing I figure they might be trying is anticipate suicides, so they can stop them before they happen. But it's probably just to sell you a grimace shake or whatever.

It is not important, but “sentiment” was a pretty standard feature of early vision AI systems, so I guess that data was collected in the hope that it would perhaps be useful some time in the future. I used to sell customer journey tracking edge AI software, and though we never shipped sentiment in production it was a thing our customers would regularly ask for.

Such an ignorant waste of task payers money. Face expression is not emotion. Can we say the elementary school fact again: one's face expression is not their emotional state. People have memories that cause facial expressions, people play out scenarios in their imagination that cause facial expressions, and ordinary every day body pain and plain old fart suppression in public causes facial expressions. Are these everyday ordinary human behaviors going to require explaining to the authorities, or will people just start saying "no" to absolute nonsense? Seriously. This is "technology" absolute nonsense, absolute tax payer theft.

That's probably true for individuals, but what about for crowds?

If say most people in a crowd are angry some might have non-angry facial expressions because they are thinking about a non-angry memory or they are busy trying to suppress a fart but wouldn't there still likely still likely be a much higher proportion of angry facial expressions than in a crown with a normal number of angry people?

In the future you can't have any bad days or you will get pre-crimed.

"If you aren't a criminal you would be on Soma. The only reason people aren't on Soma is so that they can claim the AI was wrong and they were just having a bad day."

It's wild to get to live in a reality only dreamed about. On the plus side that train ride will be nicer when the people conducting 'fart suppression' get pulled before boarding, so I'd mark that one up as positive use case.

I've heard that Amazon/Google/Microsoft will let you opt out if you have _nomap tattooed on your face.

In futuristic scenarios people had weird facial cosmetics (think blacked out around eyes in Blade Runner) or implants (Cyberpunk 2077). Now we have know the use case for that.

People need to start makeup trends/temporary tattoos/jewelry that can confused AI and start making them socially acceptable.

https://web.archive.org/web/20240617202511/https://www.wired...

Link to avoid the hundreds of essential 'partners processing data' via cookies.

I wonder if these systems are in operation at Canary Wharf too, because you are not allowed to wear face masks there already.

BTW, while EU laws protect personal data. From being exported, for example. Raw video and images should be ok to export and do actual recognition somewhere else, where laws are more permissive. No border crossing of personal data.

Do they? https://signal.org/blog/pdfs/upload-moderation.pdf

As an European, I’ve lost faith on any EU law “protecting” people. The business of trading your data is just too good to ignore.

I wonder how much AI the train companies would have to install to work out that bad train service makes passengers unhappy.

Maybe A/B test having the trains run on time.

"Megacorp dupes government into buying its spy hardware to expand surveillance state because of bikes or something."

>It was established that, whilst analytics could not confidently detect a theft, but they could detect a person with a bike

Really groundbreaking stuff. I'm so happy public funds are being used to pump surveillance data to a mega-corporation so they can tell us who does or does not have a bike.

Ah yes, bike, no bike

It seems like there actually was some value to this part:

> allowed police to speed up investigations into bike thefts by being able to pinpoint bikes in the footage

Normal procedure as far as I know is you keep recordings for 48h or whatever, and if someone reports their bike as stolen a human reviews the footage to see if they can find anything. If you can use ML or something to tag the snippets in the last 24h that contain bikes being moved, that's more efficient use of taxpayer-funded police officers' time?

It's also something that somebody with little to no previous knowledge of machine learning could whip up with OpenCV in a weekend. Maybe they could hire an undergrad to do it as a student project instead of giving Amazon unfettered access to their surveillance data?

That's just not true. Maybe you could whip something up that could detect a bike under select circumstances, but a robust system that could actually be used as a tool is a much harder project.

The government doesn't need to be duped into expanding surveillance. They are desperate for more of it.

Right, the duping comes from getting them to install adtech instead of just regular old cameras.

> "use a range of advanced technologies across our stations to protect passengers"

Total surveillance. Of course, that's for your own good. Do they guarantee that after 'smile detection' image is deleted. Or they keep images 'only for testing'?

You seem not to love our Big Brother, do you?

I had even bigger "brother" for significant part of my life. That was more than enough.

Extremely expensive yet unreliable trains... Subsidized. Extremely high taxes.... And this is what is used for: Squeezing more from people and abusing their identities further while having very little transparency on the government dealings where it matters.

Bonus points for trying to "meddle" in Macedonia CIA style.

But hey, doing useful things like closing all those loopholes that allow elites to use tax havens while operating in London or punishing corruption is not really on the list.

Amazon? Oh hell no! No way. I'll email all my personal data to NSA and anyone else who's interested. But not to Amazon.

Amazon already has your data. The data that matters to them anyway.

Companies purchase bulk customer data. Your data and my data will be included in these purchases.

Anything else is correlative and immediate ensuring that your data and my data is current.

To pretend otherwise is naive or agenda serving.

> “We take the security of the rail network extremely seriously and use a range of advanced technologies across our stations to protect passengers, our colleagues, and the railway infrastructure from crime and other threats,”

But I think Amazon and co are the threat, and you're giving them exclusive access to collect huge amounts of data!

Reply on news.ycombinator.com