so, how would one go about understanding how a computer works. well, unfortunately, the answer to this question largely depends on the motivation for asking it. for instance, if one knows how a PC works, but not a mac, then the answer would look very different then the answer to the same question from somebody who knows nothing about computers at all. thus, to appropriately answer the question: "how should one go about trying to understand how the brain works?" one should probably first specify a set of motivating factors (and maybe also desirata). so, some things we'd like include:
1) memory augmentation: our memories are limited in ways that are often detrimental to quality of life. how often is one arguing about whether or not one spouted a particular utterance? also, if i could remember everything i learned, i wouldn't need to look things up again, or even keep books after reading them.
2) optimal learning: we spend a large fraction of our lives learning knowledge that we then apply to our daily lives, whether it is in the form of wisdom or semantic information or something else, it often takes many years for us to achieve satisfactory mastering of a discipline. if we could learn more faster, technology and development could advance faster as well.
3) creative juices: while "creativity" is not particularly well understood feature of the human experience (at least i haven't found a satisfactory account of it), i still feel comfortable suggesting that if we were more creative, we could more quickly find better solutions to current problems.
4) "objective" perspectives: our lack of predictive power often results in our experiences being severely biased by previous experience. if we could somehow "objectify" our perceptions, we may be better at predicting the responses of others, and therefore be able to have more fruitful relationships.
5) better language: surely, languages are a very cool invention (or discovery, depending on perspective). and yet, they are not quite as expressive as they could be. some ideas are very difficult to express using language (for instance, abstract concepts such as the wave-particle duality). it would be great if somehow we could make languages have more expressive power.
6) love: something we all (or nearly all) of us want more of, both on the receiving and giving sides. if everybody loved one another, i imagine that many of todays problems would cease. clearly, this is a very hippie-dippie idea, and maybe seems somewhat in juxtaposition with a set of desirata for contemporary neuroscience. but, who better to figure out such a thing?
7) introspective accuracy: a common finding in psychological studies is that introspection is simply not that accurate. it'd be great if when we reflected on why we responded in the way we did, we could be more accurate.
although i have just enumerated 7 desirata, only the first two are "real" neuroscientific questions, in the sense that only two can even be expressed as questions using a neural vocabulary. the relationship between neural hardware and creativity, perspective, language, love, and introspection is so tenuous at this point, that it is probably not even worth considering until further notice. furthermore, the first two: learning and memory, are conserved evolutionarily, so we can start out by studying significantly simpler systems.
as a side note, people may argue with my desirata. ok, i'm not for everyone, and neither are my ideas. i'm happy to hear about alternative desirata...
Saturday, April 5, 2008
a good analogy for trying to figure out how the brain works
imagine finding a TV screen on the ground, and not knowing what it did or how. how could one figure out how it works. one strategy would be to look at each pixel in isolation, and see how it responded to various inputs, and then try to determine the "pixel-code", mapping the inputs to statistical regularities of the outputs. this might work. alternately, one could start with the simplest TV one could find that shares the same essential properties, and begin exploring the circuit underlying the behavior (ie, the mechanisms). upon developing the underlying operational principles, ie, the functions that resistors, capacitors, etc. perform, one could then scale up to increasingly complex systems. eventually, one could build up to something like a jumbotron, but it would be ill-advised to study a jumbotron, without first understanding a 4" black and white TV.
i think a similar argument applies to neuroscience. we could just start sticking electrodes in the brains of primates and humans, and hope that we can figure things out. or we could start with a much simpler system, and try to unravel the basic governing principles at work. this analogy breaks down, however, in a number of places.
first, i think the things that are especially cool about brains, are things that humans definitely do, and other animals do to a lesser degree. as the brain becomes less complex, the megacool properties become less pronounced. for instance, here is something especially cool that we do. you tell me the meaning of a word, and i then understand it, possibly forever. it is not clear what a homolog of that is in the animal kingdom. fortunately, other supercool attributes of human cognition do seem to have homologs. for instance, our ability to recognize objects. this is a megahard problem computationally. somehow, however, pretty much all animals have figured out how to do it. it is a necessary condition for behavior, at least at a very coarse level (ie, determining whether objects are predators or prey). so, this fear may be mitigated by studying properties that are conserved evolutionarily.
second, analog circuit elements are relatively simple as compared with neurons. this fear assumes that the fundamental (ie, "atomic") unit of neural computation is a neuron. so, one way to mitigate this fear is to postulate that the fundamental unit is something much simpler, ie, a synapse. while synapses are still much more complicated than analog circuit elements (eg, modeling a synapse "accurately" probably requires several states or dimensions, whereas analog circuit elements only require one), they are certainly closer, and it probably doesn't make much sense to postulate anything more atomic than a synapses. on this perspective, neurons become somewhat like integrated circuits, and then the brain becomes the whole circuit board.
third, one could argue that brains are much more general devices than TV's. but maybe that is not true. input to brains come in several possible forms: visual, auditory, etc. similarly, input to TV's come in several possible forms: tuners, cable, DVD's, etc. the output of brains also only have a few possibilities: speech, body language, movements, etc. similarly, TV's output only audio and visual signals. but brains seem to have something that TV's don't: internal states. ok, technically, TV's have 2 internal states: on and off. even if one postulates that different channels correspond to different internal states, the number of possible internal states for a TV pales in comparison to those of a brain, which are innumerable. so, let's switch the analogy from a TV to a computer. a computer (with all the appropriate dressings like an OS, programs, etc.) has may possible internal states. one may think of an internal state as follows: for a particular input, the output is different for a different internal state. so, for a computer, when running one program, a particular key stroke may lead to saving a document, whereas in another program, that same exact keystroke will lead to sending the document. in that sense, each program may be considered to correspond with a different internal state. and yet, computers are still insufficient, as the number of internal states for a computer is discrete and finite (eg, about 1 per program). however, in brains, the number of internal states may not be finite, and is certainly not discrete. for instance, i could be in a relatively good mood, in which case if i get hit by a car, it is not quite that bothersome; whereas if i were in a bad mood, it might be infuriating.
thus it seems as if even analogizing with the most sophisticated devices that humans understand (ie, computers) is insufficient, as the complexity of the human brain - at the level of internal states - is incomparably more complex. nonetheless, this seems as if its the best analogy that we can come up with, so we must work from there.
i think a similar argument applies to neuroscience. we could just start sticking electrodes in the brains of primates and humans, and hope that we can figure things out. or we could start with a much simpler system, and try to unravel the basic governing principles at work. this analogy breaks down, however, in a number of places.
first, i think the things that are especially cool about brains, are things that humans definitely do, and other animals do to a lesser degree. as the brain becomes less complex, the megacool properties become less pronounced. for instance, here is something especially cool that we do. you tell me the meaning of a word, and i then understand it, possibly forever. it is not clear what a homolog of that is in the animal kingdom. fortunately, other supercool attributes of human cognition do seem to have homologs. for instance, our ability to recognize objects. this is a megahard problem computationally. somehow, however, pretty much all animals have figured out how to do it. it is a necessary condition for behavior, at least at a very coarse level (ie, determining whether objects are predators or prey). so, this fear may be mitigated by studying properties that are conserved evolutionarily.
second, analog circuit elements are relatively simple as compared with neurons. this fear assumes that the fundamental (ie, "atomic") unit of neural computation is a neuron. so, one way to mitigate this fear is to postulate that the fundamental unit is something much simpler, ie, a synapse. while synapses are still much more complicated than analog circuit elements (eg, modeling a synapse "accurately" probably requires several states or dimensions, whereas analog circuit elements only require one), they are certainly closer, and it probably doesn't make much sense to postulate anything more atomic than a synapses. on this perspective, neurons become somewhat like integrated circuits, and then the brain becomes the whole circuit board.
third, one could argue that brains are much more general devices than TV's. but maybe that is not true. input to brains come in several possible forms: visual, auditory, etc. similarly, input to TV's come in several possible forms: tuners, cable, DVD's, etc. the output of brains also only have a few possibilities: speech, body language, movements, etc. similarly, TV's output only audio and visual signals. but brains seem to have something that TV's don't: internal states. ok, technically, TV's have 2 internal states: on and off. even if one postulates that different channels correspond to different internal states, the number of possible internal states for a TV pales in comparison to those of a brain, which are innumerable. so, let's switch the analogy from a TV to a computer. a computer (with all the appropriate dressings like an OS, programs, etc.) has may possible internal states. one may think of an internal state as follows: for a particular input, the output is different for a different internal state. so, for a computer, when running one program, a particular key stroke may lead to saving a document, whereas in another program, that same exact keystroke will lead to sending the document. in that sense, each program may be considered to correspond with a different internal state. and yet, computers are still insufficient, as the number of internal states for a computer is discrete and finite (eg, about 1 per program). however, in brains, the number of internal states may not be finite, and is certainly not discrete. for instance, i could be in a relatively good mood, in which case if i get hit by a car, it is not quite that bothersome; whereas if i were in a bad mood, it might be infuriating.
thus it seems as if even analogizing with the most sophisticated devices that humans understand (ie, computers) is insufficient, as the complexity of the human brain - at the level of internal states - is incomparably more complex. nonetheless, this seems as if its the best analogy that we can come up with, so we must work from there.
apocalypse now
quote from about 1000 years ago: "if you don't start accepting my g-d, the world as we know it will end."
quote from about now: "if you don't start recycling or fighting terrorism, the world as we know it will end."
arguments seems about the same to me. new set of beliefs, same old scare tactics. how about we focus our energies on solving real world current problems, famine, disease, poverty, crime, psychiatric illness, etc., instead of spending so much time, energy, and resources trying to prevent problems that may or may not happen?
quote from about now: "if you don't start recycling or fighting terrorism, the world as we know it will end."
arguments seems about the same to me. new set of beliefs, same old scare tactics. how about we focus our energies on solving real world current problems, famine, disease, poverty, crime, psychiatric illness, etc., instead of spending so much time, energy, and resources trying to prevent problems that may or may not happen?
Friday, April 4, 2008
putative theory of optimal interactions
when interacting with another human being, i think a useful guiding principle is: "do that which maximizes the expected connection between oneself and the other." i do not mean here immediate connection, but rather long term. if one analogizes with dynamics or kinetics equations, we are interested in maximizing the expected value of the steady-state of the system. in game-theoretic terms, this may be akin to the pareto optimal solution (but i don't understand that stuff well enough just yet). in buddhist thought, this may be related to the idea of searching to unify oneself with the other, to the point that one does not even consider oneself different from another (again, i'm not sure i get this stuff yet). bob marley may have referred to this idea as "one love. one heart. let's get together and feel alright," again referring to breaking down the barriers between us and just coming together. so, clearly, this is not a novel or unique idea. nonetheless, it does not seem to be central to many contemporary moral or ethical systems (at least western ones). i came to this thought because i used to act according to a very similar principle: "do that which maximizes the immediate connection between oneself and the other." sadly, this often leads to suffering, as immediate gains often come at the cost of long term suffering.
i find that i am able to act according to this new principle in a way that is not at all paternalistic. this is somewhat in contrast to: "do unto others that which you would like done unto you." i think the appropriate way to understand this thought lies in considering *psychological impact*, not actions. if i like eating cake, but you don't, then i shouldn't get you cake. if i only consider actions, then i would want cake, so i should get cake for others. instead, the important point is i would want things that *i* like, so when getting presents for others i should get them things that they like. however, acting according to this principle, even upon the interpretation that i prefer (a psychological impact emphasis rather then the "act" itself emphasis), leads me to paternalistic behavior or controversy. for instance, a woman wants to sleep with me. she says she'll like it and feels good about the decision. assuming i want to as well, i could go with it, and hope everything would be cool. or i could not go with it, the rationale being that i don't believe it is what she really wants deep down, or i think it would be better for her not to in the long run. either one of those rationales is somewhat dissatisfactory. however, if i consider our long term connection, i would concur only if i believed that it was beneficial for our relationship. this thought process somehow side steps the issue of considering her, or myself, of being paternalistic or selfish. the point is our interaction (which is really the thing at stake).
so, i'm pretty happy with this approach so far. if anybody has some qualms about it, i'd love to hear it. of course, it is not a complete theory of ethical behavior (eg, how might i consider interactions with people i've never met, such as sudanese refugees), nor does it solve all interpersonal relationship problems, but i think its a pretty good place to start.
i find that i am able to act according to this new principle in a way that is not at all paternalistic. this is somewhat in contrast to: "do unto others that which you would like done unto you." i think the appropriate way to understand this thought lies in considering *psychological impact*, not actions. if i like eating cake, but you don't, then i shouldn't get you cake. if i only consider actions, then i would want cake, so i should get cake for others. instead, the important point is i would want things that *i* like, so when getting presents for others i should get them things that they like. however, acting according to this principle, even upon the interpretation that i prefer (a psychological impact emphasis rather then the "act" itself emphasis), leads me to paternalistic behavior or controversy. for instance, a woman wants to sleep with me. she says she'll like it and feels good about the decision. assuming i want to as well, i could go with it, and hope everything would be cool. or i could not go with it, the rationale being that i don't believe it is what she really wants deep down, or i think it would be better for her not to in the long run. either one of those rationales is somewhat dissatisfactory. however, if i consider our long term connection, i would concur only if i believed that it was beneficial for our relationship. this thought process somehow side steps the issue of considering her, or myself, of being paternalistic or selfish. the point is our interaction (which is really the thing at stake).
so, i'm pretty happy with this approach so far. if anybody has some qualms about it, i'd love to hear it. of course, it is not a complete theory of ethical behavior (eg, how might i consider interactions with people i've never met, such as sudanese refugees), nor does it solve all interpersonal relationship problems, but i think its a pretty good place to start.
MIND08
so, while in nyc, i decided to take advantage of the scene, and attend a day long conference called "mind08: the design and elastic mind symposium." here are the highlights, and my reflections:
chuck hoberman: this dude invented the ball that can expand and contract. he is an architect using that idea to build dynamic spaces. turns out, dynamic spaces mostly means motorized blinds with weird shapes. they have a certain aesthetic appeal, but it is not clear whether there is any other practical advantage.
paul steinhardt: this dude has a theory about the universe. essentially, our universe lies on a membrane (or brane for short). apparently, some of the dilemmas posed by string theory can be solved by postulating the existence of another brane, orthogonal to ours. according to his theory, our brane and the other brane lie out the boundary of a dimension, and the branes cyclically collide and then spread apart again, repeating infinitely many times. thus, no single big bang, but rather many big collisions. seemed like a smart and nice dude, and he wrote some contemporary popular science books, the cool thing is that this theory actually makes predictions that should be testable within a few years (we gotta wait for technological developments).
janna levin: she's totally cool. she tackles the question of whether the universe is infinite in size, and if not, does it have boundaries. consider earth: it is both finite and boundaryless. potentially, the universe could be like that too. on earth, if one looks straight ahead, and light bends around the earth, then one would see one's own back side. similarly, if one looked backward, one would see oneself looking backward. if the universe had the same property of finiteness yet boundarylessness, then when we looked at the stars, we might be looking at the same stars at different points in time. it would be difficult to determine whether the different things we saw were actually the same thing in different times or not. that's cool.
kevin slavin: this dude makes games. turns out, games historically have taken place in "somewhere else." typically, some fantastical place that lacks any ties to our material world. he made some games that do not ascribe to that convention. in particular, shark-runners is a game about chasing sharks. you pretend to be on a boat trying to intersect shark paths. you are playing against sharks - real sharks. they implanted sharks with a little GPS, so they are actually playing against you in their real space. Crossroads: a 2 player game, played on one's cell phone. the goal is to run to as many intersections in lower manhattan as possible, in a 30 minute period. this requires actually passing through the intersections. a bad guy may be chasing you as well, so people apparently run away from a virtual villain, but they are running in real space. Plundr is a pirate game, where you live in a virutal world, and can steal or sell or trade goods. however, you can only do things in the place that you currently occupy in real space. if you go to a place on earth that nobody has played from yet, you can create your own island there, make taxes, etc.
henry markram: he is doing something called the "blue brain project" in collaboration with IBM. they want to simulate a human brain, a feat they claim to be able to accomplish in 10 years. besides the fact that it takes 200 PB to even store all the info one would need to simulate, it is not clear to me that the parameters are identifiable, given current experimental constraints. as such, they are limited to finding an equivalence class of parameters for the small stimulus space they are able to explore, and therefore probably lack the ability to simulate anything novel to make useful predictions. time will tell.
overall, it seems as if some people were literally trying to transform the way we understand reality, and others were simply making things that look cool. in general, while i was impressed with many of the designers' grasp of science, it did not seem as if they were able to make a useful contribution to science or understanding, but rather just something pretty that abstractly connected to a cool new scientific idea. also, calling this conference "mind08" seems like a misnomer, as almost nothing was actually about the mind. maybe something like "real design" would have been more appropriate. nonetheless, i'm glad i went.
chuck hoberman: this dude invented the ball that can expand and contract. he is an architect using that idea to build dynamic spaces. turns out, dynamic spaces mostly means motorized blinds with weird shapes. they have a certain aesthetic appeal, but it is not clear whether there is any other practical advantage.
paul steinhardt: this dude has a theory about the universe. essentially, our universe lies on a membrane (or brane for short). apparently, some of the dilemmas posed by string theory can be solved by postulating the existence of another brane, orthogonal to ours. according to his theory, our brane and the other brane lie out the boundary of a dimension, and the branes cyclically collide and then spread apart again, repeating infinitely many times. thus, no single big bang, but rather many big collisions. seemed like a smart and nice dude, and he wrote some contemporary popular science books, the cool thing is that this theory actually makes predictions that should be testable within a few years (we gotta wait for technological developments).
janna levin: she's totally cool. she tackles the question of whether the universe is infinite in size, and if not, does it have boundaries. consider earth: it is both finite and boundaryless. potentially, the universe could be like that too. on earth, if one looks straight ahead, and light bends around the earth, then one would see one's own back side. similarly, if one looked backward, one would see oneself looking backward. if the universe had the same property of finiteness yet boundarylessness, then when we looked at the stars, we might be looking at the same stars at different points in time. it would be difficult to determine whether the different things we saw were actually the same thing in different times or not. that's cool.
kevin slavin: this dude makes games. turns out, games historically have taken place in "somewhere else." typically, some fantastical place that lacks any ties to our material world. he made some games that do not ascribe to that convention. in particular, shark-runners is a game about chasing sharks. you pretend to be on a boat trying to intersect shark paths. you are playing against sharks - real sharks. they implanted sharks with a little GPS, so they are actually playing against you in their real space. Crossroads: a 2 player game, played on one's cell phone. the goal is to run to as many intersections in lower manhattan as possible, in a 30 minute period. this requires actually passing through the intersections. a bad guy may be chasing you as well, so people apparently run away from a virtual villain, but they are running in real space. Plundr is a pirate game, where you live in a virutal world, and can steal or sell or trade goods. however, you can only do things in the place that you currently occupy in real space. if you go to a place on earth that nobody has played from yet, you can create your own island there, make taxes, etc.
henry markram: he is doing something called the "blue brain project" in collaboration with IBM. they want to simulate a human brain, a feat they claim to be able to accomplish in 10 years. besides the fact that it takes 200 PB to even store all the info one would need to simulate, it is not clear to me that the parameters are identifiable, given current experimental constraints. as such, they are limited to finding an equivalence class of parameters for the small stimulus space they are able to explore, and therefore probably lack the ability to simulate anything novel to make useful predictions. time will tell.
overall, it seems as if some people were literally trying to transform the way we understand reality, and others were simply making things that look cool. in general, while i was impressed with many of the designers' grasp of science, it did not seem as if they were able to make a useful contribution to science or understanding, but rather just something pretty that abstractly connected to a cool new scientific idea. also, calling this conference "mind08" seems like a misnomer, as almost nothing was actually about the mind. maybe something like "real design" would have been more appropriate. nonetheless, i'm glad i went.
Subscribe to:
Posts (Atom)