The astonishingly sudden advance that GPT-4 represented, is being forgotten, a moment when everyone was floored, astonished, and amazed. So, the idea that AI is actually not AI and, is silly.
Second, the limiting factor in all these scenarios of change is how quickly can society assimilate these technologies. How quickly can they be translated into products that can be deployed. How quickly can employers start to use these new technologies.
I think we already know that AI actually IS different. I don't mean that anyone knows what the longer-term outcomes are going to be, but rather that we already have very good evidence that we have a uniquely fast feedback loop on our hands.
It's not the first technological feedback loop. You could argue that the entire industrial revolution was such a loop. Very roughly, harnessing fossil fuels, steam, and machines made industrial operations more effective, which in turn made us better at fossil fuel extraction and machine production. However, that feedback loop was slow because it involved lots of laborious and long lead time work like digging mines and building factories. The current AI loop involves writing code, coming up with new algorithms, and training better AIs, and it has already made some of those processes on the order of 10% more efficient. We're also investing trillions in compute, model training, and related improvements, rapidly growing the amount of technology and capital we can apply to AI.
Based on AI-driven productivity and investment, it seems pretty clear to me that we are already in the fastest technology adoption and deployment cycle in history. In contrast, VisiCalc adoption didn't directly improve VisiCalc. Spreadsheets have only improved incrementally over the last 20-30 years. Would we expect the same for the next 20 years of AI?
In terms of outcomes, I agree that the result isn't going to be some binary scenario like no work for anyone. Outcomes by type of work will also depend highly on how elastic demand for that type of work is. For instance, let's say software development becomes 10X more efficient but we have 20X untapped demand for more code if it can be produced faster and at a lower cost. We could actually see software employment increase by 2X. But that will not be the case for most fields. If copywriting becomes 10X more efficient, it's likely that much of that will just result in cost savings. The same way the word processor made the typing pool obsolete, may forms of cognitive work will become obsolete or a small part of a bigger job.
From our perspective, the disruptions of the industrial revolution seem relatively benign, but at the time they were big upheavals. I think we also need to keep an open mind that what's coming down the road with AI, in terms of changing the future of work, will be significantly more dramatic.
The astonishingly sudden advance that GPT-4 represented, is being forgotten, a moment when everyone was floored, astonished, and amazed. So, the idea that AI is actually not AI and, is silly.
Second, the limiting factor in all these scenarios of change is how quickly can society assimilate these technologies. How quickly can they be translated into products that can be deployed. How quickly can employers start to use these new technologies.
I think we already know that AI actually IS different. I don't mean that anyone knows what the longer-term outcomes are going to be, but rather that we already have very good evidence that we have a uniquely fast feedback loop on our hands.
It's not the first technological feedback loop. You could argue that the entire industrial revolution was such a loop. Very roughly, harnessing fossil fuels, steam, and machines made industrial operations more effective, which in turn made us better at fossil fuel extraction and machine production. However, that feedback loop was slow because it involved lots of laborious and long lead time work like digging mines and building factories. The current AI loop involves writing code, coming up with new algorithms, and training better AIs, and it has already made some of those processes on the order of 10% more efficient. We're also investing trillions in compute, model training, and related improvements, rapidly growing the amount of technology and capital we can apply to AI.
Based on AI-driven productivity and investment, it seems pretty clear to me that we are already in the fastest technology adoption and deployment cycle in history. In contrast, VisiCalc adoption didn't directly improve VisiCalc. Spreadsheets have only improved incrementally over the last 20-30 years. Would we expect the same for the next 20 years of AI?
In terms of outcomes, I agree that the result isn't going to be some binary scenario like no work for anyone. Outcomes by type of work will also depend highly on how elastic demand for that type of work is. For instance, let's say software development becomes 10X more efficient but we have 20X untapped demand for more code if it can be produced faster and at a lower cost. We could actually see software employment increase by 2X. But that will not be the case for most fields. If copywriting becomes 10X more efficient, it's likely that much of that will just result in cost savings. The same way the word processor made the typing pool obsolete, may forms of cognitive work will become obsolete or a small part of a bigger job.
From our perspective, the disruptions of the industrial revolution seem relatively benign, but at the time they were big upheavals. I think we also need to keep an open mind that what's coming down the road with AI, in terms of changing the future of work, will be significantly more dramatic.