Everyone can extrapolate. One of the most irritating tendencies of public intellectuals is the assumption that only they understand the word exponential, and then insist on asserting that every trend they can lay their eyes on must be an exponential trend (or if it's clearly not, then it will be soon).
Progress comes in fits and spurts. Sometimes there's fast progress, and then the field matures and it slows down. It was ever thus. Measured in tech demos, AI progress has been impressive. Measured in social impact it has way underperformed, with the applications until November of last year being mostly optimizations to existing products that you wouldn't even notice unless paying close attention. That's what 10+ years of billion-dollar investments into neural nets got us: better Gmail autocomplete and alt tags on facebook images.
Now we have a new toy to play with at last, and AI finally feels like it's delivering on the hype. But if we extrapolate from past AI experience it's going to mostly be a long series of cool tech demos that yields some optimizations to existing workflows and otherwise doesn't change much. Let's hope not!
Progress comes in fits and spurts. Sometimes there's fast progress, and then the field matures and it slows down. It was ever thus. Measured in tech demos, AI progress has been impressive. Measured in social impact it has way underperformed, with the applications until November of last year being mostly optimizations to existing products that you wouldn't even notice unless paying close attention. That's what 10+ years of billion-dollar investments into neural nets got us: better Gmail autocomplete and alt tags on facebook images.
Now we have a new toy to play with at last, and AI finally feels like it's delivering on the hype. But if we extrapolate from past AI experience it's going to mostly be a long series of cool tech demos that yields some optimizations to existing workflows and otherwise doesn't change much. Let's hope not!