I was almost out the door to pet shops and animal shelters so I could build a 16 cat-core setup over Thanksgiving. Then I figured I should at least read the full article before going through the trouble. It's a good thing too. Now I know that I should probably go for 64 cat-cores and treat it more like a quantum computer with the potential for rapid decoherence.
Like Searle's Chinese Room, but with a cat in it, where observing it or asking it to do anything at all necessarily yields nothing, because it's a cat. Searledingers Catbox?
But more generally I'd ask the question of, if you can train a conscious animal to execute 8 brainfuck instructions (+, -, <, > , etc..), could they also develop a writing system and express higher order languge concepts? Asking why animals don't write assumes they can't, as maybe we just need some kind of intermediate symbolic phoneme concept, as though - like BF for computation, there are some basic unit instuctions that you can combine and parse into anything, and if a border collie can distinguish words for hundreds of different things, I'd wonder if what we're missing is a set of metaphors that represent this underlying core instruction set.
Things like musical notation and knitting stitch patterns are also thought to be Turing complete, so it implies the representation doesn't matter for computation, and computation is sufficient for representing most things, so the idea of finding some underlying instruction set for producing languages across species seems both fantastic yet loosely plausible.
There are several Youtube channels where people gradually train cats and dogs to use boards full of buttons (20+ buttons in some cases) to communicate their wants. It definitely seems to work in general but doesn't make the animals any less themselves... one of the funnier examples being a video of a cat repeatedly pressing the "outside" button, staring at a yard full of snow, and then stalking back and pressing the "mad" button several times. Simple compound phrases (for example, "play friend" or "love grandma") are fairly common, but any kind of actual syntax is basically nonexistent.
I'm sure you could train dogs, monkeys, or parrots to reasonably reliably perform logical operations. From that you could build a simple processor, but the error rate might be too high for it to scale much. It's not so much getting them to do the tasks, it's getting them all to reliably do them at the same time in synchronisation.
Of course as with the guy in the Chinese Room that doesn't speak Chinese, none of the individual animals would have any idea what the overall system was doing.
There's a classic Charles Stross short story where, to spoil the story, some old men apparently "randomly" playing a board game to pass the time, are performing combinatorial magick to fight a demonically possessed ... well don't want to spoil the entire story. The background of the story is magick is real and based on math and computer logical operations and such. cstross used to post here occasionally and worked as a sysadmin in the olden days.
Anyway, aside from fantasy/sci fi stories, the interesting analogy with the story is just because I'm not smart enough to debug the system given no blueprints or explanation, does not mean an anthill is not running some novel and possibly interesting computational problem. Or bees nest or giant fungal mycelia in the dirt or herds of lobsters.
There are randomness detection and evaluation algos (distantly related to compression algos) and now that we have big data I suspect the next decades will have those algos rubbed up against big data and we'll discover interesting things about hive insects and what the hive is thinking as opposed to what individual insects are thinking. Or run it on dolphins or ancient human dwelling architecture.
And it was an "IBM 1602" although pretty obviously based very closely on the actual 1620.
The real world 1620 was a 50s era RTL scientific computer (IBM used to separate math processors from business processors, until the famous "360" that did everything pretty well). This used the SMS style hardware cards that everything IBM made used until the 360-era SLT cards. Each SMS card holds enough circuits to be roughly one 7400 series TTL chip.
He also has another series which he just finished that is very good as well about a family with the ability to step between diffrent timelines that's an interesting deconstruction of the trope.
While it starts as starts as a fantasy series, it eventually comes out their magic abilities are remnants of advanced nanotech from an interdimensional empire. Apparently a publisher he didn't like working with had the rights to his next Scifi series, so he had to start it as fantasy to get around the contract.
but the error rate might be too high for it to scale much
That just means you need a redundant array of parrots operating in parallel, with a regulator to detect different answers to the same problem and retry.
I'm glad you figured that 64 cat-cores is basically pure chaos. Well, for you, for the cores it's just their way of life: do random stuff for no apparent reason. Which is like the exact opposite of a computer.
Yes, if I can't get the system up & running for computational tasks, I figure I can always pivot the business model and sell its output as a world-class entropy source. My only problem is the lack of a moat around the model since anyone with a steady supply of cats, automated laser pointers, and a few hundred $/month in kitty kibble can reproduce my setup and make the whole thing a race to the bottom on price. I think at that point I could turn my accumulated knowledge into an excellent cat-entropy system servicing business, fine-tuning and optimizing other people's setups, advising on breed mixtures, laser speed, etc.
The business seems fraught with hairy (furry?) problems.
The laser system has promise, but you need a source of random data to drive the movement of the laser pointer (maybe strap them to other cats?). Otherwise, an attacker who can predict the movement of your pointers may be able to predict the output of your entropy stream after observing the behavior of your cat colony (certain cats may prefer to hunt at different times of the day, or prefer to lay in wait and pounce vs giving chase, etc).
Alternatively, you could observe a field (bed) of cats and record their chirality, assigning 0 to sinistral cats and 1 to dextral cats. The science[1] seems to indicate no preference for one or the other per cat, so your readings should be random. However, you may only observe one or two bit-flips per hour and during transitions it may take some time for the system to settle back down to a steady-state.
Good point. Would strapping a laser pointer to the back of one of the cats work to seed the system? After that I could divert some output into g-code movement instructions to the automated lasers.
One thing I'm unsure of is room size. I figure I could map 1/0 to each tile in a grid based on the presence of a cat, but that means with too large or small of a room I'll have varying degrees of bit density. Stakes are pretty high too, since I'm pretty sure that putting cats in the mix might allow me to exceed the theoretical maximum of 1 bit of entropy per bit.
I’ve heard that in order for this to work it is important to ensure that one of the cats is the Model M version and the other must be a Model F version.
Equally important, you must respond “No” when asked whether you want the “Neutering” feature enabled. This applies to both cats.
They're trying to use the unique nature of cats to exceed the theoretical maximum of 1:1 bits of entropy. This would create an entropic contagion event and that spread a heat death out into the universe in all directions.
Comments
I was almost out the door to pet shops and animal shelters so I could build a 16 cat-core setup over Thanksgiving. Then I figured I should at least read the full article before going through the trouble. It's a good thing too. Now I know that I should probably go for 64 cat-cores and treat it more like a quantum computer with the potential for rapid decoherence.
Like Searle's Chinese Room, but with a cat in it, where observing it or asking it to do anything at all necessarily yields nothing, because it's a cat. Searledingers Catbox?
But more generally I'd ask the question of, if you can train a conscious animal to execute 8 brainfuck instructions (+, -, <, > , etc..), could they also develop a writing system and express higher order languge concepts? Asking why animals don't write assumes they can't, as maybe we just need some kind of intermediate symbolic phoneme concept, as though - like BF for computation, there are some basic unit instuctions that you can combine and parse into anything, and if a border collie can distinguish words for hundreds of different things, I'd wonder if what we're missing is a set of metaphors that represent this underlying core instruction set.
Things like musical notation and knitting stitch patterns are also thought to be Turing complete, so it implies the representation doesn't matter for computation, and computation is sufficient for representing most things, so the idea of finding some underlying instruction set for producing languages across species seems both fantastic yet loosely plausible.
There are several Youtube channels where people gradually train cats and dogs to use boards full of buttons (20+ buttons in some cases) to communicate their wants. It definitely seems to work in general but doesn't make the animals any less themselves... one of the funnier examples being a video of a cat repeatedly pressing the "outside" button, staring at a yard full of snow, and then stalking back and pressing the "mad" button several times. Simple compound phrases (for example, "play friend" or "love grandma") are fairly common, but any kind of actual syntax is basically nonexistent.
All of the videos of these kinds I've seen were clearly fake.
Here's one example: https://www.youtube.com/channel/UCGMTesZlKa0Lokb7ZNqOJXQ
And another: https://www.youtube.com/channel/UCEa46rlHqEP6ClWitFd2QOQ
I'm sure you could train dogs, monkeys, or parrots to reasonably reliably perform logical operations. From that you could build a simple processor, but the error rate might be too high for it to scale much. It's not so much getting them to do the tasks, it's getting them all to reliably do them at the same time in synchronisation.
Of course as with the guy in the Chinese Room that doesn't speak Chinese, none of the individual animals would have any idea what the overall system was doing.
There's a classic Charles Stross short story where, to spoil the story, some old men apparently "randomly" playing a board game to pass the time, are performing combinatorial magick to fight a demonically possessed ... well don't want to spoil the entire story. The background of the story is magick is real and based on math and computer logical operations and such. cstross used to post here occasionally and worked as a sysadmin in the olden days.
Anyway, aside from fantasy/sci fi stories, the interesting analogy with the story is just because I'm not smart enough to debug the system given no blueprints or explanation, does not mean an anthill is not running some novel and possibly interesting computational problem. Or bees nest or giant fungal mycelia in the dirt or herds of lobsters.
There are randomness detection and evaluation algos (distantly related to compression algos) and now that we have big data I suspect the next decades will have those algos rubbed up against big data and we'll discover interesting things about hive insects and what the hive is thinking as opposed to what individual insects are thinking. Or run it on dolphins or ancient human dwelling architecture.
Can you share the name of that short story?
Found it: "Down on the farm"
https://www.tor.com/2008/07/20/down-on-the-farm/
And it was an "IBM 1602" although pretty obviously based very closely on the actual 1620.
The real world 1620 was a 50s era RTL scientific computer (IBM used to separate math processors from business processors, until the famous "360" that did everything pretty well). This used the SMS style hardware cards that everything IBM made used until the 360-era SLT cards. Each SMS card holds enough circuits to be roughly one 7400 series TTL chip.
Not sure of the exact short story he's referring to, but it must take place in his Laundry Files series
https://en.m.wikipedia.org/wiki/The_Laundry_Files
He also has another series which he just finished that is very good as well about a family with the ability to step between diffrent timelines that's an interesting deconstruction of the trope.
https://en.m.wikipedia.org/wiki/The_Merchant_Princes
********Spoilers*******
.
.
.
While it starts as starts as a fantasy series, it eventually comes out their magic abilities are remnants of advanced nanotech from an interdimensional empire. Apparently a publisher he didn't like working with had the rights to his next Scifi series, so he had to start it as fantasy to get around the contract.
Strike "animals" out, and it sounds like any worker in a big enough org. (Maybe I should stop reading Kafka)
That just means you need a redundant array of parrots operating in parallel, with a regulator to detect different answers to the same problem and retry.
64 cat-cores ... rapid decoherence
I'm glad you figured that 64 cat-cores is basically pure chaos. Well, for you, for the cores it's just their way of life: do random stuff for no apparent reason. Which is like the exact opposite of a computer.
Yes, if I can't get the system up & running for computational tasks, I figure I can always pivot the business model and sell its output as a world-class entropy source. My only problem is the lack of a moat around the model since anyone with a steady supply of cats, automated laser pointers, and a few hundred $/month in kitty kibble can reproduce my setup and make the whole thing a race to the bottom on price. I think at that point I could turn my accumulated knowledge into an excellent cat-entropy system servicing business, fine-tuning and optimizing other people's setups, advising on breed mixtures, laser speed, etc.
The business seems fraught with hairy (furry?) problems.
The laser system has promise, but you need a source of random data to drive the movement of the laser pointer (maybe strap them to other cats?). Otherwise, an attacker who can predict the movement of your pointers may be able to predict the output of your entropy stream after observing the behavior of your cat colony (certain cats may prefer to hunt at different times of the day, or prefer to lay in wait and pounce vs giving chase, etc).
Alternatively, you could observe a field (bed) of cats and record their chirality, assigning 0 to sinistral cats and 1 to dextral cats. The science[1] seems to indicate no preference for one or the other per cat, so your readings should be random. However, you may only observe one or two bit-flips per hour and during transitions it may take some time for the system to settle back down to a steady-state.
[1] http://blog.dimview.org/math/2017/07/28/cat-chirality.html
Good point. Would strapping a laser pointer to the back of one of the cats work to seed the system? After that I could divert some output into g-code movement instructions to the automated lasers.
One thing I'm unsure of is room size. I figure I could map 1/0 to each tile in a grid based on the presence of a cat, but that means with too large or small of a room I'll have varying degrees of bit density. Stakes are pretty high too, since I'm pretty sure that putting cats in the mix might allow me to exceed the theoretical maximum of 1 bit of entropy per bit.
You could also buy just two then wait a while and through some inexplicable magic, you'll gave 64 cores of cat.
I’ve heard that in order for this to work it is important to ensure that one of the cats is the Model M version and the other must be a Model F version.
Equally important, you must respond “No” when asked whether you want the “Neutering” feature enabled. This applies to both cats.
But be aware the resulting catsplosion will grind the simulation to a halt at some point
Are these the big/little cores I've been hearing so much about?
Ah yes, the famous "Meow's Law" of kitty density doubling.
So, the 'cat-bit' is worth 1 if the cat is alive and 0 if the cat is dead? Now I understand why quantum computing requires so much funding...
Eh, cats are cheap. Most of the money is spent on drugs and liquor. Quantum computing is well known for its party atmosphere.
Going for the 5dm or 3dm node size?
So the crazy cat ladies are really hardcore compute nerds that are just letting their cores rest?
They're trying to use the unique nature of cats to exceed the theoretical maximum of 1:1 bits of entropy. This would create an entropic contagion event and that spread a heat death out into the universe in all directions.
Cat ladies are evil nihilists.
Apple makes their own cats now days.
There's the makings of a science fiction novel in there, somewhere.
Do these cat-cores need cryogenic cooling?