It was steam-engine pioneer James Watt coined and quantified “horsepower” as a sales gimmick to sell steam engines. The point was that steam engines could replace your power-horses, not multiply...
"We have, however, had trouble ensuring that all of those “something else” pay decently and come bundled with status and security."
An equally serious problem is deciding who "we" is. (The people in Cupertino? Americans in general? the people of the world?) For Kodak, it was the citizens of Rochester. But Apple sells to the world. A job with Foxconn in China making iPhones might come with decent pay, status, and security, but with wages and working conditions no American or European would tolerate. So Apple can be a good global employer, even while being a bad local employer.
Consider that the proper analogy may be musicians rather than software engineers. About 90% of the money now earned in the business is earned by one family: Jay Z and Beyoncé.
What is the AI equivalent of a horsepower? The measures I usually see are either watts or tokens, but those are inputs not outputs. For software, it's obviously not lines of code. A programmer who can solve a problem with fewer lines of code is generally recognized as a better programmer than one who requires more. There's a whole technology for reducing the number of lines of code that is so intensive that programmers are warned against applying it prematurely.
How useful is AI technology? Self driving cars and trucks are currently deployed and the technology is supposedly ripe for wider adoption. AI as it is currently defined had nothing to do with it. Waymo's operating software is not built using LLM technology but rather a technological stack built to order and including a broad range of technologies like machine vision, text recognition, spatial navigation and linear prediction that were once considered AI but have not been so considered for decades. How many watts or tokens would one have to burn to have an LLM coding tool produce code to drive a car like a Waymo taxi?
How many jobs is AI actually going to kill? Writing jobs of all kinds, for example, have been vanishing for decades. It had nothing to do with AI. Ditto for bank tellers. Ditto for bellmen. Ditto for auto mechanics. Ditto for illustrators. Ditto for railroad workers. Ditto for typists. We are told that AI will eliminate programming jobs, but it will be hard to tell in the short run. There was almost panicky hiring during COVID, but now jobs are vanishing as that panic has passed. We're also seeing a lot of programming jobs being eliminated by the failure of VR or various video game sequels to catch on with consumers.
One thing that can be predicted is that without political change, any gains in GDP per capita will accrue to those who do not have to sell their labor on the market as has been the case for the last four decades. Per capita GDP may have doubled, but the share of it that an hour of labor has bought has halved. Again, this had nothing to do with AI.
You identify two separate issues: "share of the gains", and "bundled with status and security." The latter is the most important part. As Tocqueville observed long ago, Americans will tolerate economic inequality if leavened by social equality, at least among People who Count. (Jim Crow was more tolerant of economic than social equality: a point that Booker T. Washington perceptively seized on.)
Reducing economic inequality is conceptually easy: the old left-neoliberal notion of tax-and-redistribute. Social equality--a very heavy lift that will, at least, entail a lot more democracy, if not lampposts.
Much as I have enjoyed some of Petzold's writing, I am unclear on how it bears on your essay. But I am too lazy to go through the motions of quizzing an AI on the matter.
"We have, however, had trouble ensuring that all of those “something else” pay decently and come bundled with status and security."
An equally serious problem is deciding who "we" is. (The people in Cupertino? Americans in general? the people of the world?) For Kodak, it was the citizens of Rochester. But Apple sells to the world. A job with Foxconn in China making iPhones might come with decent pay, status, and security, but with wages and working conditions no American or European would tolerate. So Apple can be a good global employer, even while being a bad local employer.
Consider that the proper analogy may be musicians rather than software engineers. About 90% of the money now earned in the business is earned by one family: Jay Z and Beyoncé.
Touché...
What is the AI equivalent of a horsepower? The measures I usually see are either watts or tokens, but those are inputs not outputs. For software, it's obviously not lines of code. A programmer who can solve a problem with fewer lines of code is generally recognized as a better programmer than one who requires more. There's a whole technology for reducing the number of lines of code that is so intensive that programmers are warned against applying it prematurely.
How useful is AI technology? Self driving cars and trucks are currently deployed and the technology is supposedly ripe for wider adoption. AI as it is currently defined had nothing to do with it. Waymo's operating software is not built using LLM technology but rather a technological stack built to order and including a broad range of technologies like machine vision, text recognition, spatial navigation and linear prediction that were once considered AI but have not been so considered for decades. How many watts or tokens would one have to burn to have an LLM coding tool produce code to drive a car like a Waymo taxi?
How many jobs is AI actually going to kill? Writing jobs of all kinds, for example, have been vanishing for decades. It had nothing to do with AI. Ditto for bank tellers. Ditto for bellmen. Ditto for auto mechanics. Ditto for illustrators. Ditto for railroad workers. Ditto for typists. We are told that AI will eliminate programming jobs, but it will be hard to tell in the short run. There was almost panicky hiring during COVID, but now jobs are vanishing as that panic has passed. We're also seeing a lot of programming jobs being eliminated by the failure of VR or various video game sequels to catch on with consumers.
One thing that can be predicted is that without political change, any gains in GDP per capita will accrue to those who do not have to sell their labor on the market as has been the case for the last four decades. Per capita GDP may have doubled, but the share of it that an hour of labor has bought has halved. Again, this had nothing to do with AI.
these are all the good questions. I do not know anyone I am confident has good answers...
You identify two separate issues: "share of the gains", and "bundled with status and security." The latter is the most important part. As Tocqueville observed long ago, Americans will tolerate economic inequality if leavened by social equality, at least among People who Count. (Jim Crow was more tolerant of economic than social equality: a point that Booker T. Washington perceptively seized on.)
Reducing economic inequality is conceptually easy: the old left-neoliberal notion of tax-and-redistribute. Social equality--a very heavy lift that will, at least, entail a lot more democracy, if not lampposts.
"How did it do?"
Much as I have enjoyed some of Petzold's writing, I am unclear on how it bears on your essay. But I am too lazy to go through the motions of quizzing an AI on the matter.
I didn’t understand the part about not-raw-food. Do you mean food preparation?