McDonald’s is removing artificial intelligence (AI) powered ordering technology from its drive-through restaurants in the US, after customers shared its comical mishaps online.
A trial of the system, which was developed by IBM and uses voice recognition software to process orders, was announced in 2019.
It has not proved entirely reliable, however, resulting in viral videos of bizarre misinterpreted orders ranging from bacon-topped ice cream to hundreds of dollars’ worth of chicken nuggets.
McDonald’s told franchisees it would remove the tech from the more than 100 restaurants it has been testing it in by the end of July, as first reported by trade publication Restaurant Business.
This one has a clear economic issue as well: AI synthesized speech burns a shitton of money but somehow replaces minimum wage work. No idea how they imagined it would work. And tokens only getting more expensive since they’re currently heavily subsidized
What annoys me is that automating order taking doesn’t require a full LLM. They could easily program a voice to text system with a library of food items, basic voice commands, and have the outputs display on the screen so that the user can verify and correct the order verbally.
Like smart tech doesn’t need to know anything beyond it’s job. But they’re trying to normalize these massively superflous systems because the economics of big corporations has completely decoupled from reality. && Fascism.
I recall pretty reliable hotlines with just that, and when their schematic script was failing you, they refered you to a human operator. But it doesn’t have an AI buzz around it, and on paper it isn’t as simple and universal as a chatbot that can verbally answer any question by spending tokens on 3rd party’s servers. I can see why this idea sells and why technical, ethical, economical and usability issues get overlooked by linkedin managers, but it doesn’t make it less frustrating.
It’s basically the hologram from i,Robot. https://youtu.be/ZKxr0wyIic4
The main thing is that a human with subject matter knowledge designed the phonetic library. AI sloppers are leaving that for the LLM to hallucinate.
Naw, audio doesn’t take much, definitely not as much as a full time employee. The text models probably don’t need to be very large for this, the whole thing can probably run on a modern phone.
With all the optimization that has been done, everything can be done pretty efficiently except for video and good image gen.
The main issue with all this is probably audio quality (wind + car running + garbage microphones)
Inference is actually relatively cheap. A company the size of McDonald’s could also negotiate a deal.
Tokens are expensive for you but for the labs tokens are cheap, training is expensive.