Are AI productivity gains misjudged by focusing on ease of application?
Core argument: Goldman's AI report scales prospective productivity gains to the ease of applying generative AI rather than to the availability of low-hanging economic fruit.
We caution that Goldman’s approach is fairly traditional, in that it treats the prospective productivity gains across various disciplines as a function of the ease of applying these generative AI technologies, as opposed to the availability of low-hanging economic fruit. We think the gains from making it easier to produce presentations and documents and emails are much less than touted. On other hand, we stand by our argument that the gains of generative AI in other occupations are large and larger than Goldman anticipates in this report. Again, and more specifically, we think that the gains from producing software are very large, given that, as we’ve argued, we have underproduced for decades, and software sits in the sweet spot with respect to its combination of a clean grammar and logical structure, and given that it has been too costly to produce for most of its existence, hence why it’s been so slow to “eat the world,” as has sometimes been said is happening.





