One question that you do not address is whether humans will need to manage these AI researches, and how they could be trained after the current generation of human research managers retires. Of course we could wave our hands and say "AI could tell us", but that would be unsatisfying.
It will also be interesting when AI is applied to philosophy. Since essentially all philosophy contradicts all other philosophy, we already know that it is mostly a steaming pile of garbage. AI could simplify spotting things like is/ought fallacies, missing chains of reasoning, bad assumptions, and thousands of other problems. See: "What is Wrong with Our Thoughts? A Neo-Positivist Credo" by David Stove, an Australian philosopher. https://web.maths.unsw.edu.au/~jim/wrongthoughts.html
The “cloud laws” idea raises a fascinating question: how much of science requires human explanation?
Biology is full of systems that respond successfully to super complex conditions without an explicit model. If AI can similarly discover reliable patterns that humans cannot explain, we can start separating prediction, control, and understanding in some interesting and useful ways.
I am by nature a skeptic of the view that AI will usher in a bright new world, however I am in agreement with Noah on his evaluation of AI as possibly ushering in a new era in science. It is obvious that the computing (thinking?) capacity of AI can lead to new breakthroughs. My over riding concern is this, how do we "box" in AI so that those who wish to use AI for destructive ends such a bio-weapons are thwarted. Sometimes I fear that AI is the Manhattan Project with the resulting technology for sale to the highest bidder.
Will the AI agents be able to draw conclusions that humans will understand? What if the scientific discoveries remain in the AI realm and never reach us constrained humans. Like if we tried to explain to a dog how its DNA is sequenced. It will neither understand nor care.
What about looking at, like, the total costs, the likely risks, the potential benefits at, you know, the same time? The problem is that the total costs and the likely risks seem like a fatal flaw. The balloon has a hole in it. It kind of doesnt matter what color it is?
I would trade away a guaranteed total realization of EVERY potential medical and scientific miracle in exchange for protection from ANY slight incremental risk of ending up in a permanent authoritarian police state.
Ultra-human understanding from a machine system built out of human understanding, seems like a stretch. And if it happens, Noah will have to relax his strictures on the concept of emergence.
But as this is the age of the unexpected (Trump in the Whitehouse, Arsenal being beaten by Fleetwood in the Carabao Cup) the possibility shouldn't be written off. 'Quantity has a quality all of its own' etc.
AI can certainly solve problems of data synthesis and understanding, communication between nodes (researchers, research papers) within a network or between networks and hypothesis generation. Those things alone could be a huge advance. Plug the hypotheses generated into an appropriate robotic system that can interface with lab equipment, perform experiments and produce results, and the scientific cycle could be entirely self-sustaining. Though having suitably qualified humans track and understand what was going would be comforting, if nothing else.
This seems to be ignoring your own earlier post about people using the golden age capabilities to design and distribute deadly pathogens.
One question that you do not address is whether humans will need to manage these AI researches, and how they could be trained after the current generation of human research managers retires. Of course we could wave our hands and say "AI could tell us", but that would be unsatisfying.
It will also be interesting when AI is applied to philosophy. Since essentially all philosophy contradicts all other philosophy, we already know that it is mostly a steaming pile of garbage. AI could simplify spotting things like is/ought fallacies, missing chains of reasoning, bad assumptions, and thousands of other problems. See: "What is Wrong with Our Thoughts? A Neo-Positivist Credo" by David Stove, an Australian philosopher. https://web.maths.unsw.edu.au/~jim/wrongthoughts.html
The “cloud laws” idea raises a fascinating question: how much of science requires human explanation?
Biology is full of systems that respond successfully to super complex conditions without an explicit model. If AI can similarly discover reliable patterns that humans cannot explain, we can start separating prediction, control, and understanding in some interesting and useful ways.
I am by nature a skeptic of the view that AI will usher in a bright new world, however I am in agreement with Noah on his evaluation of AI as possibly ushering in a new era in science. It is obvious that the computing (thinking?) capacity of AI can lead to new breakthroughs. My over riding concern is this, how do we "box" in AI so that those who wish to use AI for destructive ends such a bio-weapons are thwarted. Sometimes I fear that AI is the Manhattan Project with the resulting technology for sale to the highest bidder.
Will the AI agents be able to draw conclusions that humans will understand? What if the scientific discoveries remain in the AI realm and never reach us constrained humans. Like if we tried to explain to a dog how its DNA is sequenced. It will neither understand nor care.
What about looking at, like, the total costs, the likely risks, the potential benefits at, you know, the same time? The problem is that the total costs and the likely risks seem like a fatal flaw. The balloon has a hole in it. It kind of doesnt matter what color it is?
I would trade away a guaranteed total realization of EVERY potential medical and scientific miracle in exchange for protection from ANY slight incremental risk of ending up in a permanent authoritarian police state.
Ultra-human understanding from a machine system built out of human understanding, seems like a stretch. And if it happens, Noah will have to relax his strictures on the concept of emergence.
But as this is the age of the unexpected (Trump in the Whitehouse, Arsenal being beaten by Fleetwood in the Carabao Cup) the possibility shouldn't be written off. 'Quantity has a quality all of its own' etc.
AI can certainly solve problems of data synthesis and understanding, communication between nodes (researchers, research papers) within a network or between networks and hypothesis generation. Those things alone could be a huge advance. Plug the hypotheses generated into an appropriate robotic system that can interface with lab equipment, perform experiments and produce results, and the scientific cycle could be entirely self-sustaining. Though having suitably qualified humans track and understand what was going would be comforting, if nothing else.
Agent Smith fully agrees.