The goal is not to produce a single hardcoded circuit which builds houses, but to produce modular circuits for all of the neural processes which humans do, but which use much less resources (plus are more suitable to analysis). Since human neural tissue underwent stochastic evolution, exact equivalents are not expected for all neural tissue components, but the goal is produce numerous formulas (which allow conversion to hardcoded circuits) which do what the individual parts of the human neural tissue can do, such as one formula which reduces groups of pixels into primitive shapes, one which reduces primitive shapes into more abstract concepts (such as inanimate object versus animal), plus for tasks (modular circuits for tasks such as to compute how to move the end of the arm to a position, plus another circuit to compute positions to move to assemble some product).
https://github.com/SwuduSusuwu/SusuPosts/blob/preview/posts/... shows pseudocode for how to use low-cost interferometers to produce accurate pointclouds (which outdoor autonomous tools can use to move around obstacles). Source code is all functional (not blackbox artificial intelligence, but more close to calculus, suitable for analysis at technical schools).
Oops, the URL was supposed to go to https://github.com/SwuduSusuwu/SusuLib/discussions/52#discus... (which discusses how to produce formulas suitable for hardcoded circuits with 100% reproducible results which scholars/professors can do analysis of to prove that such autonomous tools will perform at worst 99% as good as humans (for most conditions, will improve).
Preview (of future additions to https://github.com/SwuduSusuwu/SusuLib/blob/preview/posts/Ar...):
The goal is not to produce a single hardcoded circuit which builds houses, but to produce modular circuits for all of the neural processes which humans do, but which use much less resources (plus are more suitable to analysis). Since human neural tissue underwent stochastic evolution, exact equivalents are not expected for all neural tissue components, but the goal is produce numerous formulas (which allow conversion to hardcoded circuits) which do what the individual parts of the human neural tissue can do, such as one formula which reduces groups of pixels into primitive shapes, one which reduces primitive shapes into more abstract concepts (such as inanimate object versus animal), plus for tasks (modular circuits for tasks such as to compute how to move the end of the arm to a position, plus another circuit to compute positions to move to assemble some product).
https://github.com/SwuduSusuwu/SusuPosts/blob/preview/posts/... shows pseudocode for how to use low-cost interferometers to produce accurate pointclouds (which outdoor autonomous tools can use to move around obstacles). Source code is all functional (not blackbox artificial intelligence, but more close to calculus, suitable for analysis at technical schools).
Oops, the URL was supposed to go to https://github.com/SwuduSusuwu/SusuLib/discussions/52#discus... (which discusses how to produce formulas suitable for hardcoded circuits with 100% reproducible results which scholars/professors can do analysis of to prove that such autonomous tools will perform at worst 99% as good as humans (for most conditions, will improve).