7 ms·

The physics of information processing superobjects: Jupiter brains (1999) [pdf]

78 points by inlineint 10y ago

gwern 10y ago
see also http://aleph.se/andart2/megascale/just-how-efficient-can-a-jupiter-brain-be/ http://aleph.se/andart2/megascale/just-how-efficient-can-a-j... and https://www.gwern.net/docs/1999-bradbury-matrioshkabrains.pdf https://www.gwern.net/docs/1999-bradbury-matrioshkabrains.pd... http://aleph.se/andart2/megascale/energy-requirements-of-the-singularity/ http://aleph.se/andart2/megascale/energy-requirements-of-the...
- nabla9 10y ago
  Jupiter brains must be hollow spheres. Reason for this: Volume grows cubically, surrounding area has just quadratic growth. Fusion power generation in the core of the Sun is just 270 watts/m^3 similar to human metabolism, but but because the area grows much slower, getting rid of the heat makes the sun very hot. --- What is the computational complexity for Jupiter brain thinking process? Few options: 1. Tightly connected (like a neural network). O(n^2.373) ... O(n^3) Best modeled as distributed dense matrix multiplication. I think these algorithms are processing time limited, so the communication latency is not a problem. Node to node communication trough hollow sphere helps to remove physical communication bottlenecks. 2. MapReduce like. O(n*log n). 3. Chaotic scale‐free networks. O(n) I think.
  - gwern 10y ago
    > Jupiter brains must be hollow spheres. Reason for this: Volume grows cubically, surrounding area has just quadratic growth. See the first link. As well, Robin Hanson has some discussions of scaling heat dissipation with optimal pipe technology & branching on his OB blog and in the book form of _Age of Em_.
    - nabla9 10y ago
      Any technical solution like convection pipes just rearranges the area needed for heat transfer around. If you have have effective heat transfer you can have more compact computing core. For example, massive radiator disk with edge towards the sun. I think the choice depends on whether brains are latency limited, local computation limited or communication volume limited. I speculate that they are local computation limited in the largest scale.
  - idlewords 10y ago
    Power per cubic centimeter in the sun's core is even lower than that, similar to reptile metabolism. Think of the sun as a giant glowing space lizard.
    - arethuza 10y ago
      The one I remembered was that it's about the same energy generating density as a compost heap: https://en.wikipedia.org/wiki/Solar_core#Energy_conversion https://en.wikipedia.org/wiki/Solar_core#Energy_conversion
    - M_Grey 10y ago
      ...And that lizard is only as warm as it is because of quantum effects (tunneling).
      - zeroer 10y ago
        Could you elaborate?
        Baeocystin 10y ago
        He's referring to overcoming the Coulomb Barrier in stars that allows fusion to happen at temperatures/pressures that are otherwise too low to fuse. https://en.wikipedia.org/wiki/Quantum_tunnelling#Nuclear_fusion_in_stars https://en.wikipedia.org/wiki/Quantum_tunnelling#Nuclear_fus...
        M_Grey 10y ago
        Thanks, that's exactly what I was referring to, I should have been more clear.
  - schiffern 10y ago
    > Jupiter brains must be hollow spheres. So... it's the same shape as the biosphere. This seems to support my current working hypothesis: "sufficiently advanced civilizations are indistinguishable from nature." I just came across some surprising electron micrograph videos. They show "clumps" of hundreds of cell nuclei streaming through fungal hyphae networks one cell thick, and even some "swimming" against the current.[1] The nucleus isn't just a dumb hard drive holding DNA, but does information processing, sensing, and response as well. There's about 8 miles of fungal hyphae in a cubic inch of forest soil. So overall this represents an astounding potential for computing and sensing/effecting. I wonder if any bioinformaticists have estimated the theoretical power of such a computing architecture? [1] https://www.youtube.com/watch?v=za-vIdak3bI&t=25m15s https://www.youtube.com/watch?v=za-vIdak3bI&t=25m15s
    - hwillis 10y ago
      >This seems to support my current working hypothesis: "sufficiently advanced civilizations are indistinguishable from nature." How about "large scale optimization processes over long periods of time tend to converge onto functional solutions that meet basic dimensionality requirements."
      - M_Grey 10y ago
        I read these comments, and I just think how much good it would do someone like Dan Brown to spend an hour with people like you. sigh More broadly your collective point applies to larger megastructures like Matrioshka Brains (concentric Dyson Spheres). At some point maybe the convergence starts to look like an equation of state for some form of degenerate matter, arranged in very very thin concentric spheres.
khana 10y ago
Wow. Its all happening NOW.

78 points by inlineint 10y ago

gwern 10y ago

see also http://aleph.se/andart2/megascale/just-how-efficient-can-a-jupiter-brain-be/ http://aleph.se/andart2/megascale/just-how-efficient-can-a-j... and https://www.gwern.net/docs/1999-bradbury-matrioshkabrains.pdf https://www.gwern.net/docs/1999-bradbury-matrioshkabrains.pd... http://aleph.se/andart2/megascale/energy-requirements-of-the-singularity/ http://aleph.se/andart2/megascale/energy-requirements-of-the...

nabla9 10y ago
Jupiter brains must be hollow spheres. Reason for this: Volume grows cubically, surrounding area has just quadratic growth. Fusion power generation in the core of the Sun is just 270 watts/m^3 similar to human metabolism, but but because the area grows much slower, getting rid of the heat makes the sun very hot. --- What is the computational complexity for Jupiter brain thinking process? Few options: 1. Tightly connected (like a neural network). O(n^2.373) ... O(n^3) Best modeled as distributed dense matrix multiplication. I think these algorithms are processing time limited, so the communication latency is not a problem. Node to node communication trough hollow sphere helps to remove physical communication bottlenecks. 2. MapReduce like. O(n*log n). 3. Chaotic scale‐free networks. O(n) I think.
- gwern 10y ago
  > Jupiter brains must be hollow spheres. Reason for this: Volume grows cubically, surrounding area has just quadratic growth. See the first link. As well, Robin Hanson has some discussions of scaling heat dissipation with optimal pipe technology & branching on his OB blog and in the book form of _Age of Em_.
  - nabla9 10y ago
    Any technical solution like convection pipes just rearranges the area needed for heat transfer around. If you have have effective heat transfer you can have more compact computing core. For example, massive radiator disk with edge towards the sun. I think the choice depends on whether brains are latency limited, local computation limited or communication volume limited. I speculate that they are local computation limited in the largest scale.
- idlewords 10y ago
  Power per cubic centimeter in the sun's core is even lower than that, similar to reptile metabolism. Think of the sun as a giant glowing space lizard.
  - arethuza 10y ago
    The one I remembered was that it's about the same energy generating density as a compost heap: https://en.wikipedia.org/wiki/Solar_core#Energy_conversion https://en.wikipedia.org/wiki/Solar_core#Energy_conversion
  - M_Grey 10y ago
    ...And that lizard is only as warm as it is because of quantum effects (tunneling).
    - zeroer 10y ago
      Could you elaborate?
      - Baeocystin 10y ago
        He's referring to overcoming the Coulomb Barrier in stars that allows fusion to happen at temperatures/pressures that are otherwise too low to fuse. https://en.wikipedia.org/wiki/Quantum_tunnelling#Nuclear_fusion_in_stars https://en.wikipedia.org/wiki/Quantum_tunnelling#Nuclear_fus...
        M_Grey 10y ago
        Thanks, that's exactly what I was referring to, I should have been more clear.
- schiffern 10y ago
  > Jupiter brains must be hollow spheres. So... it's the same shape as the biosphere. This seems to support my current working hypothesis: "sufficiently advanced civilizations are indistinguishable from nature." I just came across some surprising electron micrograph videos. They show "clumps" of hundreds of cell nuclei streaming through fungal hyphae networks one cell thick, and even some "swimming" against the current.[1] The nucleus isn't just a dumb hard drive holding DNA, but does information processing, sensing, and response as well. There's about 8 miles of fungal hyphae in a cubic inch of forest soil. So overall this represents an astounding potential for computing and sensing/effecting. I wonder if any bioinformaticists have estimated the theoretical power of such a computing architecture? [1] https://www.youtube.com/watch?v=za-vIdak3bI&t=25m15s https://www.youtube.com/watch?v=za-vIdak3bI&t=25m15s
  - hwillis 10y ago
    >This seems to support my current working hypothesis: "sufficiently advanced civilizations are indistinguishable from nature." How about "large scale optimization processes over long periods of time tend to converge onto functional solutions that meet basic dimensionality requirements."
    - M_Grey 10y ago
      I read these comments, and I just think how much good it would do someone like Dan Brown to spend an hour with people like you. sigh More broadly your collective point applies to larger megastructures like Matrioshka Brains (concentric Dyson Spheres). At some point maybe the convergence starts to look like an equation of state for some form of degenerate matter, arranged in very very thin concentric spheres.

khana 10y ago

Wow. Its all happening NOW.