Our future of abundant intelligence

In 2019 Rich Sutton, a Canadian computer scientist, wrote an essay called "The Bitter Lesson". His observation is that in the prior 70 years of AI research, the field's painstaking focus on crafting approaches based on expert human knowledge has been consistently beaten by general methods, like search and learning, simply by throwing increasingly large amounts of computation at the problem.

One example of the bitter lesson is the victory of "statistical" machine learning approaches to language translation over rule-based, grammatical approaches. The lesson is bitter because so many researchers invested their careers into a spectrum of human-knowledge-based approaches. They did this only to find those methods outperformed by simpler general techniques, rendering their hard-won progress obsolete as sufficient compute arrived post about 2015.

Rich Sutton's observation was a comment on research efforts to create AI systems. But a similar Bitter Lesson is evident in the AI application layer, too.