[HN Gopher] New work extends the thermodynamic theory of computa...
       ___________________________________________________________________
        
       New work extends the thermodynamic theory of computation
        
       Author : dlojudice
       Score  : 90 points
       Date   : 2024-05-14 13:24 UTC (9 hours ago)
        
 (HTM) web link (www.santafe.edu)
 (TXT) w3m dump (www.santafe.edu)
        
       | dlojudice wrote:
       | Every computing system, biological or synthetic, from cells to
       | brains to laptops, has a cost. This isn't the price, which is
       | easy to discern, but an energy cost connected to the work
       | required to run a program and the heat dissipated in the process.
       | 
       | Researchers at SFI and elsewhere have spent decades developing a
       | thermodynamic theory of computation, but previous work on the
       | energy cost has focused on basic symbolic computations -- like
       | the erasure of a single bit -- that aren't readily transferable
       | to less predictable, real-world computing scenarios.
        
         | yodon wrote:
         | If you're going to directly quote the first two paragraphs of
         | the article, it's generally considered good form to indicate
         | you're quoting from the article (for example by starting your
         | text with ">")
        
       | gryn wrote:
       | if you find this interesting you'll also find the publications of
       | J.P Cruthfield interesting, he seem to have worked in this same
       | institute up to 2004 before moving to UC Davis. he has 20+ years
       | worth of papers on the topic, I keep procrastinating reading them
       | esp the ones about what he calls epsilon machines.
       | 
       | I don't really understand this topic, but find the premise
       | interesting enough.
       | 
       | here's a paper about
       | 
       | the intrinsic cost of modularity
       | https://journals.aps.org/prx/abstract/10.1103/PhysRevX.8.031...
       | 
       | Anatomy of a Bit: https://arxiv.org/abs/1105.2988
       | 
       | Modes of Information Flow https://arxiv.org/abs/1808.06723
       | 
       | stuff related to Landauer's bound
       | https://arxiv.org/abs/1812.11241 https://arxiv.org/abs/1909.06650
       | https://www.researchgate.net/publication/350794561_Refining_...
        
       | idontknowtech wrote:
       | From the headline, I thought this was yet another attempt by some
       | silicon valley bro to handwave furiously about everything being
       | thermodynamics. Thankfully, this was not the case.
       | 
       | I'm kinda surprised nobody's done this before, given how
       | important estimating wastage is.
        
         | gs17 wrote:
         | The thermodynamic "limit" is pretty small compared to any other
         | waste in a system (from Wikipedia):
         | 
         | > At room temperature, the Landauer limit represents an energy
         | of approximately 0.018 eV (2.9x10^-21 J). Modern computers use
         | about a billion times as much energy per operation.
        
       | colmmacc wrote:
       | One of Feynman's lesser known works is his lectures on
       | computation.
       | 
       | https://www.amazon.com/Feynman-Lectures-Computation-Frontier...
       | 
       | Feynman is not as expert on the topic as he is on his core
       | research, but his gift for great explanation carries over and
       | makes the material more accessible. The lectures are ground in
       | thermodynamics and the related information theories, and there's
       | a very accessible lecture in there too about Maxwell's Daemon.
       | Most of the material is very foundational and still correct, so
       | it's a good read for anyone who is interested in the area. I'm
       | glad I read it before I had to deal with more complicated and
       | statistical approaches to computation and entropy.
        
         | gnatman wrote:
         | Fenyman worked with Thinking Machines on their first
         | supercomputer, The Connection Machine in the 80s. Here's some
         | great background and anecdotes from the founder, Danny Hillis:
         | 
         | https://longnow.org/essays/richard-feynman-connection-machin...
        
         | winwang wrote:
         | He also proposed using quantum computers to solve "hard"
         | physical systems.
         | 
         | https://arxiv.org/pdf/2106.10522
         | 
         | Though, I'm unsure if he were the first to do so in a semi-
         | public manner.
        
       | anonzzzies wrote:
       | This is important for our resources and connected to the (at
       | least for me) very interesting field of reversible computing.
        
       | rulalala wrote:
       | Is SFI alive after all his founders are long gone? I think so,
       | but really it is increasingly difficult to discern its current
       | unique contribution to the global scientific landscape.
        
       | pmayrgundter wrote:
       | Adjacent work from Fields & Levin 2021 on the thermodynamics of
       | cellular process, arguing that due to the information processing
       | demands and classical thermo minimal power requirements per
       | operation for loose estimates of protein control, aggregate
       | cellular metabolism power requirements are off by 10-15 orders of
       | magnitude of available. They then conclude that cellular
       | processes are using quantum coherent processes for major work
       | internally and externally (p17).
       | 
       | https://chrisfieldsresearch.com/quantum-cells-pre.pdf
        
         | delichon wrote:
         | So perhaps like consciousness.
         | 
         | https://www.youtube.com/watch?v=R6G1D2UQ3gg
        
           | darby_eight wrote:
           | I smell Penrose bullshit in the air
           | 
           | Of course the brain uses quantum mechanics, it exists in the
           | real world. What has yet to be demonstrated is any link to
           | consciousness.
        
             | delichon wrote:
             | The link addresses your point.
        
               | darby_eight wrote:
               | I am not watching a video
        
         | spacetimeuser5 wrote:
         | If Levin wouldn't produce so much fluff on pseudoscientific
         | concepts like "selves" and "computational psychiatry", he would
         | have had the mice regenerate better and already proceed to
         | humans, like Sinclair.
         | 
         | How would quantum theory explain the thermodynamics of
         | increased mental clarity reported by humans on 3-5 day of
         | fasting induced ketosis? Or multiple biocomputational benefits
         | of mTOR inhibition (by e.g. rapamycin for longevity purposes),
         | reduced rate of metabolism?
         | 
         | Extrapolating to modern mechanotech, this would be similar to
         | an airplane flying faster and better upon regular reduction of
         | fuel to it. Imagine a computer running more loads and faster
         | upon regular reduction of AC power to it. Thus linear
         | application of thermodynamics seems rather ridiculous and
         | quantum theory may not save, because it disintegrates beyond
         | the Plank's scale.
        
       | slwvx wrote:
       | The underlying paper:
       | https://journals.aps.org/prx/pdf/10.1103/PhysRevX.14.021026
       | 
       | I skimmed it; it's not super accessible.
        
       | nico wrote:
       | What are irreversible problems? Any good/fun and maybe practical,
       | examples?
        
         | sponaugle wrote:
         | From a boolean logic point of view - AND is irreversible, since
         | (1 .AND. 0) is 0, as is (0 .AND. 0.) Going from the result (0),
         | you can't reverse and get the starting inputs.
         | 
         | NOT however is reversible, since 1 .NOT. -> 0, and you can get
         | back to 1 with another .NOT.
        
           | nico wrote:
           | Thank you for the examples
           | 
           | It seems like AND is similar to calculating a binary
           | derivative (https://news.ycombinator.com/item?id=40328821)
           | 
           | In a certain way you could say you can integrate AND
           | 
           | For example 1 AND 0 = 0
           | 
           | Now if I only have 0 as a result, it means one of the
           | arguments was 0, and the other is either 0 or 1
           | 
           | So the "integral" of AND(0,x) = 0, is (0, (either 0 or 1)) or
           | just x = (either 0 or 1)
           | 
           | [0 or 1 meaning one means either. It doesn't mean 0 OR 1,
           | which would be 1]
           | 
           | This is analog to the result of an integral being expressed
           | as a function + C (constant). The C part is the uncertainty
           | of the result, just like having (0 or 1) for AND
           | 
           | The (0 or 1) part can also be expressed as a probability
           | distribution, for example p(0)=0.5,p(1)=0.5 which then you
           | can use to "sample" the integral of AND by randomly taking a
           | (0 or 1) and using it as argument to AND(0, x)
        
       | spacetimeuser5 wrote:
       | >>Every computing system, biological or synthetic, from cells to
       | brains to laptops, has a cost.
       | 
       | Obviously so, fetching tiny potatoes from point A to point B by
       | monkeys has a cost.
        
       ___________________________________________________________________
       (page generated 2024-05-14 23:01 UTC)