An interesting article in The Atlantic about the slow adoption of automation in the fast food industry that covers (briefly) many of the points that we had in our discussion about Optimus.
The TL;DR - there’s not a lot of roboticization in fast food. Robots can’t do a whole lot of fast food jobs because it’s hard for robots to manipulate food, and hard to teach them to do it because the internet’s worth of data that power the training for large language models (like ChatGPT) aren’t available. But also because it’s hard to fit a robot that can only do one or two things into a fast food restaurant.
Looking at Musk’s dreams for Optimus, that’s a mixed bag. Optimus solves the size problem - a humanoid robot will fit much more easily into a place designed for humans. But there’s a ton of work that needs to be done on the manipulation problem - tactile sensing for robots is still in its infancy, and lack sufficient algorithms to predict how objects will react and deform when manipulated.
As someone who worked short order cooking I can tell you that most human activities require very high levels of very rapid highly varied tactile sensing leading immediately to precise but always slightly varied actions.
E.G. You are cooking 7 orders of eggs in three different styles with different levels of doneness along with attached bacon sausage patties for three different tables. In the bits realm this is as easy as it gets, but in reality it is MESSY. Same is true of leading a s/w team with insanely opposed temperaments working on deadline to satisfy internally contradictory idiot specs in a changing management environment.
AI dealing with unreal social media burps and mumbles makes (bad) sense, but for the real world? I am still very much at “show me”.
I’m reminded of Tysons comments on robots butchering chickens. People are developing them but so far humans do it better. Humans get more of the more valuable cuts increasing profits.
Maybe one day but not ready for prime time.