Recent searches
Search options
"Economic impacts of AI-augmented R&D"
This @tamaybes@twitter.com @nicholaskemery@twitter.com @ProfNeilT@twitter.com paper is a tour de force. They argue that deep learning makes R&D more capital-intensive and productive, which in turn makes the economy more productive.
https://arxiv.org/abs/2212.08198
They use a model of technological progress in computer vision to estimate the capital intensity of deep learning. A highlight is that they estimate the human capital in a paper with a deep learning model - this is a great example of impactful, deep-learning augmented R&D. Meta!
Their results suggest that if, thanks to deep learning, other domains of R&D became as capital intensive as computer vision, this could ~ double productivity growth.
They conclude with a nice discussion of the generalisability of findings to other domains of R&D / domains with low-data regimes, and even whether DL-enabled knowledge could generate the same spillovers as other types of R&D. More research needed!
PS. I am also interested in whether DL-derived predictive knowledge is a complement or substitute for explanatory knowledge in the production function, what mixes of human capital (e.g. technical / domain knowledge) complement AI capital, how DL in changing...
... the composition of capital in different domains of R&D (e.g. structural biology), and on how the adoption of AI is changing the structure of knowledge production in science e.g. does it make it more or less centralised and what does this mean for productivity...
PPS. Here is @tamaybes@twitter.com's thread.
https://twitter.com/tamaybes/status/1609659231682334720