AI Shopping Startup's AI Was Actually Just Low Cost Workers in the Philippines (theinformation.com) 43
Some startups are bold and original. And some, like Nate, had more modest goals: automatically filling out shoppers' contact and payment information on retailers' websites. In exchange for sparing them a minute or two of data entry on their phones, Nate charged shoppers $1 per transaction. But it struggled to turn even that vision into reality. The Information: While the company said it was using artificial intelligence to populate customer information during the checkout process, it had actually hired workers in the Philippines to manually enter the data on retailers' sites for a significant portion of the transactions Nate facilitated in 2021, according to two people with direct knowledge of the company's practices. That meant customers' orders were sometimes placed hours after they clicked the buy button through the Nate app. Nate didn't disclose its decidedly low-tech methods to at least some of the investors from whom the startup tried to raise money, according to a person with direct knowledge of fundraising discussions.
Fake it til you make it (Score:4, Insightful)
Mechanical turk is probably cheaper than the upfront spend on AI training
Re: (Score:3, Insightful)
If not that then it certainly is a time-honoured silly valley ploy.
The more mindboggling thing is to presume people wouldn't mind paying a dollar extra just so they don't have to fill in some stupid forms in their webbrowser. This probably says something about the circles the founders of the company and thus floaters of the idea run with.
Re:Fake it til you make it (Score:4, Interesting)
The problem with this business plan is twofold:
1)I'll pay to avoid paperwork, but not. this little paperwork. You need to be a bit more complex than this to make it worthwhile.
2)This is a built in feature in most browsers.
I'd pay an upfront fee one time for a better version of the browser feature. I won't pay a per use fee on it, that's insane.
Re: (Score:2)
You forget point 3)
Potential for your data to be leaked out by those working low end data entry.
Re: Fake it til you make it (Score:2)
Re:Fake it til you make it (Score:5, Interesting)
Mechanical turk is probably cheaper than the upfront spend on AI training
Mechanical Turk annotating isn't cheaper than AI training--- it is an essential part of the upfront spend on of AI training.
In order to know how good you AI is, you need annotated examples. In order to train your AI, you need annotated examples. The obvious way to get these annotated examples is to start a service backed by Mechanical Turk: you get annotated examples, the customer gets the service they desire and you get income. What's not to like?
Re:Fake it til you make it (Score:4, Informative)
This is like the ACTUAL Mechanical Turk [wikipedia.org].
It was allegedly a chess playing automaton in the late 18th to early 19th century. What it actually was, was a guy inside a wooden statue playing for the "automaton". Same thing here. Supposed to be AI, but in reality just a human pretending to be AI.
Re: Fake it til you make it (Score:2)
DALLE2 and GPT3 are actually just some third world children really fast/good at their job. (j/k)
I knew it: So-called 'AI' is FAKE (Score:4, Funny)
Re: (Score:3)
You would suck at driving too if you can't see a damn thing and get knocked out regularly by the horrible feet smell!
Re: (Score:2)
Just made me think of a Rick and Morty episode. Summer uses one of Ricks inventions to increase her breast size but accidentally makes her self 100 feet tall.
On the side of the device is a phone number and Beth calls it. The 3 "support" guys tell her to open a side panel on the device and then they promptly escape out, leaving Beth standing there clueless.
Uncanny Valley (Score:2)
Wonder what tipped off people interacting with it? Beyond that, would people believe a voice-based AI that speaks perfect English or would they be more likely to believe a foreign accent?
Isn't this a horrific privacy violation (Score:5, Insightful)
If a shopper entered enough info to complete a transaction and then some low paid worker in the Philippines had to manually enter it - doesn't that mean all those workers had access to payment information too? Couldn't any of those workers have used the info they were entering to place an order anywhere else?
Re:Isn't this a horrific privacy violation (Score:5, Interesting)
doesn't that mean all those workers had access to payment information too?
Welcome to reality. Do you know how the company setup that payment terminal? Do you know the store clerk ran up the charges correctly? What happens with that credit card information you typed into that website? We have to have some form of trust with the data both personal and financial we hand over, and I guarantee your that little "I agree to the terms and conditions" checkbox at the bottom almost certainly said that their workers and subcontractors have access to your personal information.
This is one of the reasons I like the Dutch iDEAL system so much. Your payment details are never given to the merchant. Paying via iDEAL redirects you to the bank website who then does a challenge-response authentication with you (usually using the bank's own app, or the bank's 2FA token), and the merchant only gets told if payment was successful or not.
Re: (Score:1)
Do you know how the company setup that payment terminal?
If they use Apple Pay I do, that's why I use it where possible.
Do you know the store clerk ran up the charges correctly?
There again, this is why I save credit receipts and check charges online against the receipt - and also just scan it over at time of purchase. Even if they aren't out to scam you sometimes mistakes are made so it's always good to look.
This is one of the reasons I like the Dutch iDEAL system so much.
Similar to why I like Apple Pay.
Re: (Score:2)
If they use Apple Pay I do, that's why I use it where possible.
Nope. You're just trusting a different third party in your case. You feel warm and fuzzy inside because we're not running a story on Apple pay outsourcing the handling of your data to the Philippians.
There again, this is why I save credit receipts and check charges online against the receipt
After the fact is not a mechanism for prevention, it's a mechanism for damage reduction / mitigation. If we run a system where we need to do this kind of check afterwards then the system has failed to meet *your* required level of trust.
Similar to why I like Apple Pay.
Apple pay is a useless 3rd party who doesn't need to be part of a transacti
Re: (Score:2)
Would be a criminal act in Europe. You are no allowed to send or process privacy relevant data in ways and places you did not get explicite informed consent for.
Re: (Score:1)
Startups move fast and break rules.
Re: (Score:2)
They do. Last year I had to use a sentence like "if not fixed, this finding may result in personal criminal liability for the decision makers" in an audit-report for a start-up in the insurance sector. That problem got fixed really fast....
Re: (Score:1)
Hard to sue somebody in Timbuktu from Europe. Europe may require a local office with a signer, but they can usually skip town faster than authorities can hunt them down if things go sour.
Re: (Score:2)
"Vanishing" is not that easy. Typically you lose most of your money when you do something like that. Also, any kind of international travel becomes a risk. Not a good option in any way. In the case at hand, no chance at all, unless they wanted to go into "sans papier" status.
Re: (Score:1)
> Typically you lose most of your money when you do something like that.
Only a couple of week's worth if you regularly xfer your paycheck out.
> any kind of international travel becomes a risk.
If HQ pays you handsomely during your tenure, many would be okay with that risk.
Re: Isn't this a horrific privacy violation (Score:2)
Cool (Score:1)
I personally don't see the problem, apart from the social problems common to gig workers (probably don't get fixed salary or health insurance). I'd probably be happy to pay 1 â to have someone in the Philippines make tedious paperwork for me. Rich Filipino people have maids. I'd feel like I my butler Alfred to bring the fine whiskey for me. If I was not that lazy I'd open a remote Filipino maid business immediately.
Re:Cool (Score:5, Insightful)
The problem isn't doing it this way, its lying to your investors and customers about how you're doing it.
To investors- you're probably breaking federal law by falsely claiming to have tech you don't.
To customers- you're giving my personal info to gig workers in the Phillipines with no data protection and little legal protection, without me knowing.
If they're honest about this, let everyone choose their own risk tolerance and I'm ok with it. Lying about it is the problem.
Re: (Score:2)
without me knowing.
But you were told this would happen. You DID read the ToS right? Honestly if we actually read the ToS two things would happen:
a) we'd be so alarmed we'd not use most services.
b) we'd be so tired from the long read at the end that we forgot why we wanted to use the service in the first place.
Re: (Score:2)
Burying things in a ToS is not acceptable. Things like this need to be stated out front in clear english, not legalese meant to maximize their flexibility later on. My lawyer has to sit with those things for a significant amount of time to fully understand them, the average person has no chance.
Reminds me the Russian robot (Score:2)
So it was AAI (Score:2)
Artificial Artificial Intelligence.
Does that work like a double negative equating to a positive? If so, it was just "Intelligence".
Welcome to Movie Phone! (Score:2)
"Why don't you just tell me the name of the movie you selected..."
Re: (Score:2)
Great reference, but at this point most people will need a link...
https://youtu.be/gSQ6q_rGpI8 [youtu.be]
Sounds like Algo (Score:2)
Reminds me of Algo AI.
Supposed "smart assistant" but it was just a group of developers running SQL queries overnight via batch jobs to get "answers".
Still wondering if anyone found out it's a ruse yet?
https://en.wikipedia.org/wiki/Mechanical_Turk (Score:2)
Will have to be renamed.
Re: (Score:1)
If it is impossible, fake it! And still get rich! (Score:2)
Well, that is a "maybe" about the "get rich" part. But I am not surprised at all, that "AI" still fails at supposedly easy tasks like filling in addresses. It nicely shows how dumb these systems actually are.
I also have now seen a few sample "dialogs" of the Google AI that is supposedly going to become sentient eventually (grande, baseless claim by the project lead, IMO a direct lie, same of which we now have seen so many by sleazy AI researchers). This machine is utterly dumb and understands absolutely not
Re: (Score:2)
> It nicely shows how dumb these systems actually are.
AI's can be smart. What this shows, is how dumb webpage designers can be, as a collective.
Re: (Score:2)
AI's can be smart.
Citation needed. Because there are zero instances of that happening that I am aware off. And I have been following the field for 35 years now.
Wha? (Score:2)
I mean the summary has to be missing some key detail, yes? If not, I am really, REALLY in the wrong business. I should be out there bilking venture cap firms.
Natural Intelligence Processing (Score:2)