• Humans have achieved greater living standards than other animals due to two great meta-innovations: history and science.
• History is about recording knowledge in language, while science is about discovering generally applicable principles about how the world works.
• Science is often done in a lab, but can also be done by observing nature. Mathematics is a powerful tool for expressing laws of the universe.
• Despite the success of science, some complex phenomena have so far defied the approach of discovering si1mple, generalizable laws, leading to the idea that some domains of human knowledge may never be described by such principles.
• Leo Breiman’s essay “Statistical Modeling: The Two Cultures” demonstrated that algorithmic models (early machine learning techniques) were yielding better predictions than data models, even though the former were far less easy to interpret.
• Alon Halevy, Peter Norvig, and Fernando Pereira argued that in the cases of natural language processing and machine translation, applying large amounts of data was effective even in the absence of simple generalizable laws.
• AI may always be powerful yet ineffable, performing frequent wonders, but prone to failure at fundamentally unpredictable times.
• Natural experiments are a different tool than science and history, as they allow us to verify causal links.
• Khachiyan et al. used deep neural nets to look at daytime satellite imagery, in order to predict future economic growth at the hyper-local level, with astonishing accuracy.
• AI may revolutionize fields of endeavor where traditional science has run into diminishing returns, leading to a leap in human power and flourishing.
Published December 31, 2022. Visit Noahpinion to read Noah Smith’s original post.
