Firstly, you wrote quite a lovely article! I enjoyed the deep dive into A.I. vs. Automation, so I have to thank you for linking it under my note yesterday.
Secondly, I was just writing an article on a closely related subject under A.I., and happened to come to the same opinion as yours. I would like to note, it does appear that a lot of the concern and fear mongering towards A.I. stems from the fact that a lot of the fear mongerers/exaggerated reports cast fear on the potential for AGI. While, I don't believe that it will follow the same principles of economics as automation and other innovations have, it is comforting to know there is a decent chance it will.
I would like to know, what is your thoughts on the general landscape of reporting of A.I.? I've found it to be either overly positive without criticism, to overtly negative and down right terrifying.
Thank you for the kind words! Regarding the coverage of AI, I agree with you that it feels like the coverage follows these extremes. It would be beneficial if the authors of any pieces stated a few of their assumptions at the onset of their articles/opinion piece. It would be much more interesting and easier to discuss their working model to determine whether their thoughts are internally consistent, even before they extrapolate them to the new AI technology.
Great article. I tend to agree with your predictions.
There was a good recent article in the Journal of Economics Teaching on how to teach the perspective that AI will substitute for human jobs in a principles class. I'll be having my students look at a couple different models and studies and argue which they think is more likely to happen.
Quite a nifty idea in the paper in creating a joint supply curve. I've been thinking about how supply/demand curves (which ultimately are reflections of an underlying model) sometimes may seem 'incorrect' (the supply/demand curves are not wrong themselves, but rather we're making an incorrect assumption in the underlying model). This simple idea that labor supply should instead be a joint supply of a task that can be done by labor or a 'robot' gives a different perspective on what matters in the model (not just total labor, but type of labor).
Yeah. I've discussed with some of the econ department whether new tech should be thought of as capital in a production function, or as a trading partner in a comparative advantage model. Or something else. I was interested to notice in this paper that he mentioned the robots' comparative advantage several times. I don't know the Acemoglu + Restreppo paper well enough to say whether their task-based model dovetails with comparative advantage. But it does emphasize the importance in creating new types of work if other types of work are substituted for. Which is an intuition that has been around for a while.
second of all, what I was taught in uni, is that neural networks (a language model is a neural network as far as I know) are really good at interpolating the data, but not great at extrapolating. So basically what chatgpt does, is figuring out whats the most likely response that someone would get on the internet to a question that they asked, and responding with that. This works well for things that have already been done, but when you have to do something that is geniuinely new, you will not get much help from it. Of course, 90% of work for developers are tasks that have been done a milion times over, but at the same time its not what takes the most of the time (the everpresent 80/20 rule). So I think some jobs will become obsolete / demand will decrease a lot - for example writing copy, or generating otherwise not that creative content (just ask the chat gpt to write the description of a company or product), while others (devs, doctors or similar) will be greatly enhanced by it.
Have you (the Author, not Autor :D) tried using chat gpt to help you write the article? Could be interesting to see how it deals with more precise economical terminology and logic connections.
Personally, I'm also leaning that at least just neural networks will not go far beyond what you say. As many so far point out, it's great at sounding like it knows things but so far rarely captures nuances (and especially thinking outside the box)
I haven't yet used it for my writing work here. I have tried using the Bing search engine but wasn't too great, although prompting is a skill now. I did test the Bing in summarizing the articles I've written to see how it does and also if what I wrote makes sense. That is a great feature.
There is also an interesting economics approach - one economist is basically using ChatGPT bots as agents in economic models to see what they would do. A way to possibly test behavior - https://twitter.com/johnjhorton/status/1606364947335741453
Some interesting results here, even including understanding political predispositions.
Firstly, you wrote quite a lovely article! I enjoyed the deep dive into A.I. vs. Automation, so I have to thank you for linking it under my note yesterday.
Secondly, I was just writing an article on a closely related subject under A.I., and happened to come to the same opinion as yours. I would like to note, it does appear that a lot of the concern and fear mongering towards A.I. stems from the fact that a lot of the fear mongerers/exaggerated reports cast fear on the potential for AGI. While, I don't believe that it will follow the same principles of economics as automation and other innovations have, it is comforting to know there is a decent chance it will.
I would like to know, what is your thoughts on the general landscape of reporting of A.I.? I've found it to be either overly positive without criticism, to overtly negative and down right terrifying.
Thank you for the kind words! Regarding the coverage of AI, I agree with you that it feels like the coverage follows these extremes. It would be beneficial if the authors of any pieces stated a few of their assumptions at the onset of their articles/opinion piece. It would be much more interesting and easier to discuss their working model to determine whether their thoughts are internally consistent, even before they extrapolate them to the new AI technology.
Great article. I tend to agree with your predictions.
There was a good recent article in the Journal of Economics Teaching on how to teach the perspective that AI will substitute for human jobs in a principles class. I'll be having my students look at a couple different models and studies and argue which they think is more likely to happen.
https://journalofeconomicsteaching.org/a-note-on-teaching-automation-in-undergraduate-courses-casey/
Thank you for the comment and the link!
Quite a nifty idea in the paper in creating a joint supply curve. I've been thinking about how supply/demand curves (which ultimately are reflections of an underlying model) sometimes may seem 'incorrect' (the supply/demand curves are not wrong themselves, but rather we're making an incorrect assumption in the underlying model). This simple idea that labor supply should instead be a joint supply of a task that can be done by labor or a 'robot' gives a different perspective on what matters in the model (not just total labor, but type of labor).
Yeah. I've discussed with some of the econ department whether new tech should be thought of as capital in a production function, or as a trading partner in a comparative advantage model. Or something else. I was interested to notice in this paper that he mentioned the robots' comparative advantage several times. I don't know the Acemoglu + Restreppo paper well enough to say whether their task-based model dovetails with comparative advantage. But it does emphasize the importance in creating new types of work if other types of work are substituted for. Which is an intuition that has been around for a while.
first of all: https://i.redd.it/hnwjtobij0381.jpg
second of all, what I was taught in uni, is that neural networks (a language model is a neural network as far as I know) are really good at interpolating the data, but not great at extrapolating. So basically what chatgpt does, is figuring out whats the most likely response that someone would get on the internet to a question that they asked, and responding with that. This works well for things that have already been done, but when you have to do something that is geniuinely new, you will not get much help from it. Of course, 90% of work for developers are tasks that have been done a milion times over, but at the same time its not what takes the most of the time (the everpresent 80/20 rule). So I think some jobs will become obsolete / demand will decrease a lot - for example writing copy, or generating otherwise not that creative content (just ask the chat gpt to write the description of a company or product), while others (devs, doctors or similar) will be greatly enhanced by it.
Have you (the Author, not Autor :D) tried using chat gpt to help you write the article? Could be interesting to see how it deals with more precise economical terminology and logic connections.
Personally, I'm also leaning that at least just neural networks will not go far beyond what you say. As many so far point out, it's great at sounding like it knows things but so far rarely captures nuances (and especially thinking outside the box)
I haven't yet used it for my writing work here. I have tried using the Bing search engine but wasn't too great, although prompting is a skill now. I did test the Bing in summarizing the articles I've written to see how it does and also if what I wrote makes sense. That is a great feature.
There is also an interesting economics approach - one economist is basically using ChatGPT bots as agents in economic models to see what they would do. A way to possibly test behavior - https://twitter.com/johnjhorton/status/1606364947335741453
Some interesting results here, even including understanding political predispositions.