I do appreciate how explicit the software engineer that named that app was - this whole thing would probably have gone unnoticed with minor amounts of obfuscation.
I do appreciate how explicit the software engineer that named that app was - this whole thing would probably have gone unnoticed with minor amounts of obfuscation.
It's been a long time since I've ever approached a vending machine, but I'm curious as to why this is even necessary. The only (non-nefarious) reason I can even think of is power saving reasons,...
Adaria Vending Services told MathNEWS that "what’s most important to understand is that the machines do not take or store any photos or images, and an individual person cannot be identified using the technology in the machines. The technology acts as a motion sensor that detects faces, so the machine knows when to activate the purchasing interface—never taking or storing images of customers."
It's been a long time since I've ever approached a vending machine, but I'm curious as to why this is even necessary. The only (non-nefarious) reason I can even think of is power saving reasons, but I find it difficult to believe continuously running a camera and consuming CPU cycles to detect faces consumes less power than alternatives. (I also find it difficult to believe they even care about the power usage, but that's neither here nor there.) Especially when some of those machines have flashing lights or are constantly showing things on their screen when not in use. I mean, even just a couple of motion sensors working in tandem to work as a proximity sensor would likely be cheaper; I have that same tech on my garbage can, and it keeps it powered for a year on a couple of AA batteries.
I've poked my head into vending a little bit in the past, so I took a look and found a PDF for a smart vending machine from one of the vendors listed in the article, Invenda. I don't get the...
I don't get the impression it's power saving. The PDF says one of the reports available is "demographics," which reminds me of machines I saw for Japan that could detect the gender of the customer and highlight specific products. The PDF goes on to specifically call out some specific metrics: "foot traffic" and "gender/age/etc"
They also mention AI, because of course they do.
Do away with the limitations of traditional machines. Increase sales thanks to a multi-product basket and numerous upselling features: combo deals, promotions, animations*, enhanced imagery, AI product recommendations, gamification, cross-selling and sampling.
The machine could be using AI to do image processing and categorization of customers, then just storing the anonymized data and sending that to the cloud for reporting.
To me, at least, actively reinforcing the current state of things is a fairly large step past just acknowledging it. I think explicitly describing it as racist or sexist depends pretty heavily on...
To me, at least, actively reinforcing the current state of things is a fairly large step past just acknowledging it. I think explicitly describing it as racist or sexist depends pretty heavily on one’s own interpretation of those terms, but I’m pretty comfortable in saying it’s harmful.
Advertising is agnostic to whether it harms or helps the target group, just so long as it makes them spend. If reinforcing biases and stereotypes, even at the expense of the target group, is effective then that’s what will be done - and it’s a short step from there to a good amount of evidence that the stereotyping and pigeon-holing of groups is detrimental to some or all of the people in those groups.
[Edit] Who’s to blame is a very interesting question, though… It’s rare for an individual ad campaign or technique to shift societal thinking on its own, and even in the vanishingly small number of cases where they have we’re then looking to second- and third-order effects to measure the harm. Then we have to ask who and how many of the many interconnected individuals who contributed can be considered “responsible”, if any or all even can? And then we have to ask if they could reasonably have foreseen the negative outcomes, if they acted in honest good faith, if…, if…, if…
Having the ability to (try to) identify a customer based on their appearance then change the sales behavior according to that categorization goes beyond mere targeted advertising and is, by...
If their programming reflects actual market trends, is it really racist or sexist? It might miss the mark every once in a while, but presumably less so than traditional untargeted advertising, otherwise they wouldn't spend the money to develop the tech and buy the demo info.
Having the ability to (try to) identify a customer based on their appearance then change the sales behavior according to that categorization goes beyond mere targeted advertising and is, by definition, racist or sexist.
Imagine if the prices offered were higher when a female-presenting person stood in front of the machine. Or higher because the system detected traditionally Asian facial features. Or it simply claimed to be out of lower-price options to steer certain types of people toward higher price options.
The quote from the vendor literature was "combo deals, promotions, ... and sampling", so those behaviors are certainly within the machines capability. The actual behaviors are likely to be more subtle, but differences of degree, not kind.
Doing something that obviously discriminatory seems short sighted on the vendors part, but here in the US, we have seen people lean in to catering to white supremacist groups to establish a power base, so catering to them to increase sales doesn't seem outside the Overton window any more.
Obviously I'm describing behaviors far beyond what the original story described, so we are into "slippery slope" territory. And what I've described is the most extreme option. It's likely the actual discriminatory behaviors would be less obvious and harder to detect.
The reason to ban this kind of thing at a policy level, even if the machines behave neutrally for now, is that once the capability is there, the profit motive makes actually doing these things only a matter of time.
Aside from the potentially discriminatory aspects of demographic targeting, this sounds like design for dark patterns: I don't want f*cking junk food slot machines providing operant conditioning...
Aside from the potentially discriminatory aspects of demographic targeting, this sounds like design for dark patterns:
combo deals, promotions, animations*, enhanced imagery, AI product recommendations, gamification, cross-selling and sampling.
I don't want f*cking junk food slot machines providing operant conditioning via intermittent reward, with personalized targeting. In fact, it would delight me if there was a preemptive legislative ban on putting this technology anywhere under-18s might have access.
I'm a little hesitant to completely deprive adults of particular choices. I'm not philosophically libertarian, I recognize that humans are animals with cognitive vulnerabilities that can be...
I'm a little hesitant to completely deprive adults of particular choices. I'm not philosophically libertarian, I recognize that humans are animals with cognitive vulnerabilities that can be hacked. I'd rather legislatively foreclose opportunities to exploit those vulnerabilities profitably (mandatory warnings, public advertising limits, sin taxes, limited licenses, age restrictions, harm minimization, etc.) than outright ban anything except non-consensual activity.
I know it sounds like bureaucratic hell, but that's what a maze of guardrails looks like, rather than straightforward prison bars cutting off access.
I agree. "Gamification" in particular is a pet peeve of mine, as I think it's often a messy system with no benefit that's tacked on after the fact for marketing purposes. It's funny you mention...
A lot of machines use motion detection to turn on their displays and lights. I also believe the person quoted in the article is probably correct in that it doesn't store actual facial pictures. It...
A lot of machines use motion detection to turn on their displays and lights. I also believe the person quoted in the article is probably correct in that it doesn't store actual facial pictures. It probably makes an estimate on sex and age and adds that to the record of the sale. I can see why marketers would want that, but it feels pretty gross to not disclose.
I'm reminded of Japan's approach to nuclear deterrent. They have no nuclear weapons but take the needless expense to run their nuclear energy plants with enhanced uranium. The deterrent is, if...
I'm reminded of Japan's approach to nuclear deterrent. They have no nuclear weapons but take the needless expense to run their nuclear energy plants with enhanced uranium. The deterrent is, if they ever felt the need to develop nukes, it would only take a few months because the infrastructure is already there.
Compare this to companies, who could make a killing selling face data if they were allowed (it's legally grey right now I believe?), installing facial recognition into everything for seemingly innocuous reasons. The infrastructure is already there, so when the legal green light comes through...
Most nuclear power plants operate on some level of enriched uranium. Generally between 3 and 5%. You might be thinking about their stockpile of separated plutonium, which (pre-Fukushima) was to be...
Most nuclear power plants operate on some level of enriched uranium. Generally between 3 and 5%. You might be thinking about their stockpile of separated plutonium, which (pre-Fukushima) was to be used for breeder reactors. That could potentially be used in a weapon, although it isn’t optimal, especially for a first design. There’s a reason most countries with nukes run/ran dedicated material production reactors, the isotopics are better.
I do appreciate how explicit the software engineer that named that app was - this whole thing would probably have gone unnoticed with minor amounts of obfuscation.
So true but "Invenda.Vending.TotallyInnocuousApp.exe" was already taken.
It's been a long time since I've ever approached a vending machine, but I'm curious as to why this is even necessary. The only (non-nefarious) reason I can even think of is power saving reasons, but I find it difficult to believe continuously running a camera and consuming CPU cycles to detect faces consumes less power than alternatives. (I also find it difficult to believe they even care about the power usage, but that's neither here nor there.) Especially when some of those machines have flashing lights or are constantly showing things on their screen when not in use. I mean, even just a couple of motion sensors working in tandem to work as a proximity sensor would likely be cheaper; I have that same tech on my garbage can, and it keeps it powered for a year on a couple of AA batteries.
I've poked my head into vending a little bit in the past, so I took a look and found a PDF for a smart vending machine from one of the vendors listed in the article, Invenda.
I don't get the impression it's power saving. The PDF says one of the reports available is "demographics," which reminds me of machines I saw for Japan that could detect the gender of the customer and highlight specific products. The PDF goes on to specifically call out some specific metrics: "foot traffic" and "gender/age/etc"
They also mention AI, because of course they do.
The machine could be using AI to do image processing and categorization of customers, then just storing the anonymized data and sending that to the cloud for reporting.
You only say this because you clearly have a Cheddar & Sour Cream Ruffles face.
Huh. So now vending machines can be racist and sexist.
To me, at least, actively reinforcing the current state of things is a fairly large step past just acknowledging it. I think explicitly describing it as racist or sexist depends pretty heavily on one’s own interpretation of those terms, but I’m pretty comfortable in saying it’s harmful.
Advertising is agnostic to whether it harms or helps the target group, just so long as it makes them spend. If reinforcing biases and stereotypes, even at the expense of the target group, is effective then that’s what will be done - and it’s a short step from there to a good amount of evidence that the stereotyping and pigeon-holing of groups is detrimental to some or all of the people in those groups.
[Edit] Who’s to blame is a very interesting question, though… It’s rare for an individual ad campaign or technique to shift societal thinking on its own, and even in the vanishingly small number of cases where they have we’re then looking to second- and third-order effects to measure the harm. Then we have to ask who and how many of the many interconnected individuals who contributed can be considered “responsible”, if any or all even can? And then we have to ask if they could reasonably have foreseen the negative outcomes, if they acted in honest good faith, if…, if…, if…
Having the ability to (try to) identify a customer based on their appearance then change the sales behavior according to that categorization goes beyond mere targeted advertising and is, by definition, racist or sexist.
Imagine if the prices offered were higher when a female-presenting person stood in front of the machine. Or higher because the system detected traditionally Asian facial features. Or it simply claimed to be out of lower-price options to steer certain types of people toward higher price options.
The quote from the vendor literature was "combo deals, promotions, ... and sampling", so those behaviors are certainly within the machines capability. The actual behaviors are likely to be more subtle, but differences of degree, not kind.
Doing something that obviously discriminatory seems short sighted on the vendors part, but here in the US, we have seen people lean in to catering to white supremacist groups to establish a power base, so catering to them to increase sales doesn't seem outside the Overton window any more.
Obviously I'm describing behaviors far beyond what the original story described, so we are into "slippery slope" territory. And what I've described is the most extreme option. It's likely the actual discriminatory behaviors would be less obvious and harder to detect.
The reason to ban this kind of thing at a policy level, even if the machines behave neutrally for now, is that once the capability is there, the profit motive makes actually doing these things only a matter of time.
Aside from the potentially discriminatory aspects of demographic targeting, this sounds like design for dark patterns:
I don't want f*cking junk food slot machines providing operant conditioning via intermittent reward, with personalized targeting. In fact, it would delight me if there was a preemptive legislative ban on putting this technology anywhere under-18s might have access.
I'm a little hesitant to completely deprive adults of particular choices. I'm not philosophically libertarian, I recognize that humans are animals with cognitive vulnerabilities that can be hacked. I'd rather legislatively foreclose opportunities to exploit those vulnerabilities profitably (mandatory warnings, public advertising limits, sin taxes, limited licenses, age restrictions, harm minimization, etc.) than outright ban anything except non-consensual activity.
I know it sounds like bureaucratic hell, but that's what a maze of guardrails looks like, rather than straightforward prison bars cutting off access.
I agree. "Gamification" in particular is a pet peeve of mine, as I think it's often a messy system with no benefit that's tacked on after the fact for marketing purposes. It's funny you mention slot machines though, as some Japanese vending machines also already randomly dispense an extra drink or gift after vending your purchase.
You could just detect motion, it would likely work just as well..
A lot of machines use motion detection to turn on their displays and lights. I also believe the person quoted in the article is probably correct in that it doesn't store actual facial pictures. It probably makes an estimate on sex and age and adds that to the record of the sale. I can see why marketers would want that, but it feels pretty gross to not disclose.
I'm reminded of Japan's approach to nuclear deterrent. They have no nuclear weapons but take the needless expense to run their nuclear energy plants with enhanced uranium. The deterrent is, if they ever felt the need to develop nukes, it would only take a few months because the infrastructure is already there.
Compare this to companies, who could make a killing selling face data if they were allowed (it's legally grey right now I believe?), installing facial recognition into everything for seemingly innocuous reasons. The infrastructure is already there, so when the legal green light comes through...
Most nuclear power plants operate on some level of enriched uranium. Generally between 3 and 5%. You might be thinking about their stockpile of separated plutonium, which (pre-Fukushima) was to be used for breeder reactors. That could potentially be used in a weapon, although it isn’t optimal, especially for a first design. There’s a reason most countries with nukes run/ran dedicated material production reactors, the isotopics are better.