4 ms·

The discussion about AI-driven workforce reductions misses the point entirely: on a macro level the real problem isn’t job losses — it’s the persistent stagnati

The discussion about AI-driven workforce reductions misses the point entirely: on a macro level the real problem isn’t job losses — it’s the persistent stagnation in productivity growth, particularly in developed economies. Historically, automation has driven significant increases in productivity, which in turn raised living standards and created new industries. However, recent decades have failed to deliver those gains. Productivity growth in the U.S., for instance, has been below 1% for most of the last 15 years, despite massive advances in technology. Given that AI should be increasing productivity, I tend to think that the actual utility of most of our advances in AI is massively out out of proportion with the reality.

foxyv 2y ago
I remember back in 2012 I was working for UPS as a package sorter. All along the aisle were leftover machines covered in dust and grease. They were all leftover from a time when some bright spark tried to automate package sorting. The idea seems pretty easy at first. Read the label, send it to the correct area. But if you work at it for a few hours you realize how much skill it takes to do properly. Labels are obstructed, damaged, crossed out, and modified. I've seen boxes with 5-6 different labels all in various states and defaced in various ways. Then comes the part where you need to get the packages from the input belt onto the correct outputs. Every package was a different shape, size, and condition. A lot had broken open. Some were nothing but masking tape in the shape of a mushy bowling ball. We would then often have to buffer the packages while one belt went down. I would sometimes spend hours thinking about what it would take to automate my $15/hour job and just kept coming up with ridiculously complicated robotic projects with AI that is only possible now with extremely expensive computation. Even then it would need a ton of human supervision.

foxyv 2y ago

I remember back in 2012 I was working for UPS as a package sorter. All along the aisle were leftover machines covered in dust and grease. They were all leftover from a time when some bright spark tried to automate package sorting. The idea seems pretty easy at first. Read the label, send it to the correct area. But if you work at it for a few hours you realize how much skill it takes to do properly. Labels are obstructed, damaged, crossed out, and modified. I've seen boxes with 5-6 different labels all in various states and defaced in various ways. Then comes the part where you need to get the packages from the input belt onto the correct outputs. Every package was a different shape, size, and condition. A lot had broken open. Some were nothing but masking tape in the shape of a mushy bowling ball. We would then often have to buffer the packages while one belt went down. I would sometimes spend hours thinking about what it would take to automate my $15/hour job and just kept coming up with ridiculously complicated robotic projects with AI that is only possible now with extremely expensive computation. Even then it would need a ton of human supervision.