Pizza Hut's AI system caused 'cascading' problems and $100M in damages, franchisee alleges in new suithttps://www.businessinsider.com/pizza-hut-ai-system-dragontail-lawsuit-franchisee-2026-5

[haliphax@hachyderm.io]

@rysiek If you read further...

> The complaint says DoorDash drivers began waiting to batch multiple orders together after gaining virtual visibility into kitchen systems, allowing them to see when pizzas would come out of the oven.

> Instead of immediately leaving with a completed order, the suit claims drivers waited "up to fifteen (15) minutes" for additional deliveries, increasing the time between when a pizza is removed from the oven rack and when it leaves the building to be delivered. That delay slowed deliveries, disappointed customers, and caused a sharp drop in sales, the suit says.

[wizbean@tech.lgbt]

@rysiek instead of baking the pizzas and just hoping and praying that an independent contractor (door-dasher) sees the order in an app and chooses to take it, they should consider hiring delivery people directly and pay them a good wage and pension

would literally solve all their problems

[mighty_orbot@retro.pizza]

@rysiek Not exactly right: “Instead of immediately leaving with a completed order, the suit claims drivers waited ‘up to fifteen (15) minutes’ for additional deliveries, increasing the time between when a pizza is removed from the oven rack and when it leaves the building to be delivered. The lawsuit also alleges Dashers could see tip amounts and whether orders were cash payments, making some drivers less likely to accept certain deliveries.”

The AI software did exactly what it was supposed to do. The delivery drivers were never meant to have that much access.

[cmthiede@social.vivaldi.net]

@trebach @grob @rysiek and before that it was a place to take the family to sit down and share a pizza together, pick out some music on the jukebox and play video games while waiting for the pizza to arrive as the parents got to have a 1:1 conversation with the kids entertained. But technology took care of all that!

[trebach@functional.cafe]

@cmthiede @grob @rysiek Provided you had enough quarters. I liked the lunch buffet they had

[henrik@eliitin-some.fi]

@jollysea @rysiek

And the root cause for the dashers to optimize is a crappy earning levels to begin with? (just a guess)

[nini@oldbytes.space]

@rysiek Did the AI somehow break through a "guardrail" and give drivers way too much info? Can't see how that'd go wrong, not like anyone's ever played an AI against itself.

[cmthiede@social.vivaldi.net]

@trebach @grob @rysiek yeah, it was a decent place when it still had a soul.

[grob@mstdn.social]

@trebach @rysiek no lies detected

[jackryder@infosec.exchange]

@rysiek

[tymwol@hachyderm.io]

@rysiek unpaywalled link https://archive.is/6NvBm

[gimulnautti@mastodon.green]

@rysiek Put a robot to do a human’s job..

[chemicaleyeguy@mstdn.science]

@rysiek #AI is #clankers all the way down, and that’s why #wankers love it *so much!*

Even *more* than hot pizza , apparently!

#Resist #AIslop.

[androcat@toot.cat]

@rysiek I have the same issue with deductive logic applied to the real world.

'Everybody is just "doing it wrong", it's really a good tool, otherwise.'

Somehow the total preponderance of "human error" in that domain is not a fundamental fault in the tool, can't be, it's infallible, you know!!

[szescstopni@circumstances.run]

@rysiek CC @davidgerard

[androcat@toot.cat]

@cnr

There are mentions of it several times, though.

The failure here is a particular one, not the usual hallucinations - The company wanted to make something, anything at all, with "AI".
So they replaced a functioning system that had purpose-guided information flows with "something, anything with 'AI'".
And because it was "with 'AI'" no thought was given to the purposes involved, old knowledge was disregarded, and I am willing to bet, SMEs were told to "work with it, not against it" or "fail fast so we can improve it", without actually checking if this "something, anything" was a good idea.

@jollysea @rysiek

[pi@vnil.de]

@rysiek@mstdn.social > The complaint says DoorDash drivers began waiting to batch multiple orders together after gaining virtual visibility into kitchen systems, allowing them to see when pizzas would come out of the oven. Instead of immediately leaving with a completed order, the suit claims drivers waited "up to fifteen (15) minutes" for additional deliveries[...]

The lawsuit also alleges Dashers could see tip amounts and whether orders were cash payments, making some drivers less likely to accept certain deliveries.

this one really is "just" a labor problem (food delivery services underpay while doing over-optimized pay schemes to the drivers) / suboptimal product decision (oversharing someone else's data) that's not unique to AI though, unless we assume that some AI did the entirety of PM's job and got rubberstamped. One can share the kitchen data in plain text with drivers and get the same outcome.

[nemobis@mamot.fr]

@rysiek «The complaint says DoorDash drivers began waiting to batch multiple orders together after gaining virtual visibility into kitchen systems»

This sounds like an intentional plan to extract money from restaurants and pump it into the gatekeepers' bank accounts (a joint DoorDash/Pizza Hut goal I presume).

The main contribution of "AI" is presumably the hope that it provides plausible deniability for said high street robbery.

[milagemayvary@mstdn.social]

@tymwol @rysiek

When possible, use archive.org

archive.today also owns the urls:
archive.fo
archive.is
archive.li
archive.md
archive.ph
archive.vn

Wikipedia blacklists Archive.today, starts removing 695,000 archive links

If DDoSing a blog wasn't bad enough, archive site also tampered with web snapshots.

Ars Technica (arstechnica.com)

Archive.today has begun rolling out Google's reCaptcha as well, which is concerning when it can be altered.

I do not have evidence, but I can see how this could possibly be a malware vector in the future.

[jeantranscene@mastodon.social]

@androcat @rysiek Here you have a tool that's non deterministic in nature, trained on human behavior and which impact have not been fully tested. What were we expecting exactly? But yes, it's absolutely the tool's fault.