4 ms·

IMHO this is a special case of something much more profound. I studied biologically inspired and evolutionary computing very intensely in college, and I reached

by api 10y ago

IMHO this is a special case of something much more profound. I studied biologically inspired and evolutionary computing very intensely in college, and I reached the conclusion fairly early that the way we do compute is very unlike the natural world. Our computers are designed to compute finite values using instruction sets that are explicitly designed to be precise, which has the side effect of these instruction sets and encodings being highly brittle and un-evolvable. This is what makes things like neural nets and genetic programming hard-- making these systems work involves creating virtual anti-computers within our anti-biological existing ones that are fuzzy, mutable, evolvable, etc. I speculate that this is because we've built computers explicitly to do things we are not good at: crunching massive amounts of data very precisely. We are pretty good at more "organic" forms of cognition so there was little evolutionary "market need" for us to augment ourselves in that way.

j1vms 10y ago
Though for this specific discussion (which illustrates the uncountably infinite nature of the set of real numbers), it is useful to note that the class of realizable computers includes both analog and digital variants, though usually we are more familiar with the digital in the form of our desktop/laptop computers or servers. Analog computers can be indeed be used for these types of problems, and their usage in the field predates their digital siblings. Sampling of the result of an analog calculation is of course limited by analog noise (e.g. thermal noise), which we must contend with in this universe. On the other hand, processing in the analog domain gets us further than quantization noise and heat dissipation (due to clock signal & switching) would in the digital domain. Edit: If you are interested further, here is a discussion of the application of analog computing in solving differential equations: http://chalkdustmagazine.com/features/analogue-computing-fun-differential-equations/ http://chalkdustmagazine.com/features/analogue-computing-fun... Also, someone on here once posted a link to a recent dissertation comparing the efficiency of solution finding for a problem tackled analog vs. digital, wherein given the problem's nature, the analog computer outperformed its digital counterpart both in terms of performance and energy consumption. Edit2: Here is the dissertation http://www.cisl.columbia.edu/grads/gcowan/vlsianalog.pdf http://www.cisl.columbia.edu/grads/gcowan/vlsianalog.pdf And the comment it was from: https://news.ycombinator.com/item?id=11728186 https://news.ycombinator.com/item?id=11728186
wyager 10y ago
Being a messy computer doesn't make one better at messy computations. The reason humans are decent at certain tasks is usually because we have a lot of (low-quality) hardware directed at the task and we've had many years to evolve a workable hack-job. There is a common pattern in AI where a problem that humans do easily seems intractable, right up until we find a cheap trick that gets the job done and seems to have similar quirks to the biological equivalent. If we're lucky, we find an even better way than what evolution has stumbled upon. For example, computers are way better than humans at landmark recognition (using SIFT-inspired algorithms), but it took us a while to find the tricks to do that efficiently.

by api 10y ago

j1vms 10y ago

Though for this specific discussion (which illustrates the uncountably infinite nature of the set of real numbers), it is useful to note that the class of realizable computers includes both analog and digital variants, though usually we are more familiar with the digital in the form of our desktop/laptop computers or servers. Analog computers can be indeed be used for these types of problems, and their usage in the field predates their digital siblings. Sampling of the result of an analog calculation is of course limited by analog noise (e.g. thermal noise), which we must contend with in this universe. On the other hand, processing in the analog domain gets us further than quantization noise and heat dissipation (due to clock signal & switching) would in the digital domain. Edit: If you are interested further, here is a discussion of the application of analog computing in solving differential equations: http://chalkdustmagazine.com/features/analogue-computing-fun-differential-equations/ http://chalkdustmagazine.com/features/analogue-computing-fun... Also, someone on here once posted a link to a recent dissertation comparing the efficiency of solution finding for a problem tackled analog vs. digital, wherein given the problem's nature, the analog computer outperformed its digital counterpart both in terms of performance and energy consumption. Edit2: Here is the dissertation http://www.cisl.columbia.edu/grads/gcowan/vlsianalog.pdf http://www.cisl.columbia.edu/grads/gcowan/vlsianalog.pdf And the comment it was from: https://news.ycombinator.com/item?id=11728186 https://news.ycombinator.com/item?id=11728186

wyager 10y ago

Being a messy computer doesn't make one better at messy computations. The reason humans are decent at certain tasks is usually because we have a lot of (low-quality) hardware directed at the task and we've had many years to evolve a workable hack-job. There is a common pattern in AI where a problem that humans do easily seems intractable, right up until we find a cheap trick that gets the job done and seems to have similar quirks to the biological equivalent. If we're lucky, we find an even better way than what evolution has stumbled upon. For example, computers are way better than humans at landmark recognition (using SIFT-inspired algorithms), but it took us a while to find the tricks to do that efficiently.