7 ms·

The science knowledge base grows monotonically but not necessarily at a steady rate. Actually far from it. There was a time in the 19th century when people tho

The science knowledge base grows monotonically but not necessarily at a steady rate. Actually far from it. There was a time in the 19th century when people thought physics discoveries were complete. A few decades later all hell broke lose. The problem we have right now is that we are stymied on the complexity front (materials, chemistry, biology, medicine etc) Our mental / mathematical models that we use to understand the world and even enable us to ask original scientific questions are losing their edge when things get complex. These tools are 19th century stuff (including all the AI junk) and we have milked them for all they are worth I would hypothesise that a new period of discovery will correlate with some breakthrough in the mathematics of complexity. We had a few false dawns but not the real thing

feoren 4y ago
> These tools are 19th century stuff (including all the AI junk) I'm not really sure what you're looking for here. I mean, you could claim that modern cryptography is based on ideas from the ancient Greeks (factoring numbers), and is therefore old and busted. Is every idea just derivative unless it uses a completely new kind of math? Software developers (maybe not you, but a large percentage of the users of this site) tend to automatically think older ideas are worse. But that's not true in math. Mathematical ideas don't expire. The idea of the number 7 is aeons old; the number 0 over a thousand years; the complex number i 450 years. But the numbers 7 and 0 and i are still extremely useful, in infinite combinations. If you think we have milked those numbers for all they're worth, you have an extremely limited imagination. Sure, most of the "AI junk" is a combination of old ideas like linear algebra and gradient descent, but used at a scale orders of magnitude larger than their creators ever imagined. (And quantity has a quality all its own.) It's that combination of ideas that is so powerful. Besides, backpropagation is a pretty damned novel idea. If you're looking for the one truly new idea from AI, it's backpropagation. You are vastly underestimating the volume of the idea space available to explore simply by combining the ideas and materials we already have. Crossover is a better source of variety than mutation is. You may be right that truly new concepts are the "raw materials" of invention, but we never run out of the raw ideas we have, and they prove to be useful in their combination up to the limits of our imaginations. It's actually surprising how likely you are to come up with a reasonably good idea by simply considering a few randomly-selected ideas together. > we are stymied on the complexity front (materials, chemistry, biology, medicine etc) Again you seem like your standard is: if it's not an entirely new thing where nothing like it has ever existed before, then it's boring!. If you think the volume of the idea-space from combining the materials, chemistry, biology, and medicine we already have/understand is X, it's almost certainly at least 2^X, or even X! (factorial). Besides, we are coming up with brand new materials all the time! Materials science is one of the most exciting frontiers of human technology.
- college_physics 4y ago
  Dont get me wrong, there is plenty of scope for interesting / useful applications of software and pattern matching at scale but those are more a form of "tech assisted social evolution" rather than the steady march of fundamental inventions and discoveries across the entire spectrum of knowledge that characterised the last few centuries. Old math can still have an impact at the margin (eg speeding up protein folding calculations etc) but its not the cognitive leap that would allow us to steer discoveries in more effective manner. Materials, medicine etc all very exciting and very active but also hampered by their intrinsic complexity.

feoren 4y ago

> These tools are 19th century stuff (including all the AI junk) I'm not really sure what you're looking for here. I mean, you could claim that modern cryptography is based on ideas from the ancient Greeks (factoring numbers), and is therefore old and busted. Is every idea just derivative unless it uses a completely new kind of math? Software developers (maybe not you, but a large percentage of the users of this site) tend to automatically think older ideas are worse. But that's not true in math. Mathematical ideas don't expire. The idea of the number 7 is aeons old; the number 0 over a thousand years; the complex number i 450 years. But the numbers 7 and 0 and i are still extremely useful, in infinite combinations. If you think we have milked those numbers for all they're worth, you have an extremely limited imagination. Sure, most of the "AI junk" is a combination of old ideas like linear algebra and gradient descent, but used at a scale orders of magnitude larger than their creators ever imagined. (And quantity has a quality all its own.) It's that combination of ideas that is so powerful. Besides, backpropagation is a pretty damned novel idea. If you're looking for the one truly new idea from AI, it's backpropagation. You are vastly underestimating the volume of the idea space available to explore simply by combining the ideas and materials we already have. Crossover is a better source of variety than mutation is. You may be right that truly new concepts are the "raw materials" of invention, but we never run out of the raw ideas we have, and they prove to be useful in their combination up to the limits of our imaginations. It's actually surprising how likely you are to come up with a reasonably good idea by simply considering a few randomly-selected ideas together. > we are stymied on the complexity front (materials, chemistry, biology, medicine etc) Again you seem like your standard is: if it's not an entirely new thing where nothing like it has ever existed before, then it's boring!. If you think the volume of the idea-space from combining the materials, chemistry, biology, and medicine we already have/understand is X, it's almost certainly at least 2^X, or even X! (factorial). Besides, we are coming up with brand new materials all the time! Materials science is one of the most exciting frontiers of human technology.

college_physics 4y ago
Dont get me wrong, there is plenty of scope for interesting / useful applications of software and pattern matching at scale but those are more a form of "tech assisted social evolution" rather than the steady march of fundamental inventions and discoveries across the entire spectrum of knowledge that characterised the last few centuries. Old math can still have an impact at the margin (eg speeding up protein folding calculations etc) but its not the cognitive leap that would allow us to steer discoveries in more effective manner. Materials, medicine etc all very exciting and very active but also hampered by their intrinsic complexity.