Science & Technology Q&A for Kids (and others) [Part 62]

hi everyone welcome to another episode of science and technology q a for kids and others i'm back in my usual environment here um and looks like we have some questions saved up from a previous time so let me try and look at a few of these all right here's a slightly different one from a toure what do i think of psychological personality tests and what type am i so interesting you know people are different they react to the world differently how we come to be different is an interesting question that is probably a mixture of nature sort of how we are constructed and our genetics and nurture how we end up growing up and um one of the things that people have been doing for about a hundred years now is having these kind of tests where you say answer these questions and then characterize sort of what personality type are you based on these questions and so a really common one is the myers-briggs tests which i think are based on things from psychologist called carl jung from from sometime in the in the early part of the 1900s um but in any case the um uh usually people sort of characterize with these four letters like i'm typically you know if i do these things i'll typically be an entj i'm not sure i can remember what all the letters mean but the e is extrovert so each each one of these attributes there's one uh there's two possible values you can either be an extrovert or an introvert and i'm afraid i can't necessarily remember what the other ones are um but uh what does that mean well you know there's some people like kind of me who tend to in a situation i will express myself i'll kind of be out there you know talking to people if i'm uh with people i don't know i'll typically be hello let me introduce myself type thing that's kind of the typical extrovert story the introvert story tends to be more let me keep to myself if you know if there's a situation where it's like i could go and just say hello i'm pleased to meet you type thing versus i could just stay away and wait for somebody else to come and say hello to me that tends to be the introvert type thing and so it's kind of a question of what are you really measuring what's really going on in these situations well it's pretty clear that different people have different personality types can we really characterize the personality types that people have in terms of let's say you know four letters each with two possible values so that's some uh you know 16 possible overall personality type so to speak does that really work are there really 16 kinds of people or is that just a terribly coarse approximation and perhaps even a misleading one i don't think anybody knows i think that it's a little bit like trying to characterize things like emotions you know happy versus sad this versus that people have tried to make some characterization of the kind of space of possible emotions i was going to say human emotions but i don't need to say human emotions because we know it's a famous work of charles darwin's actually that uh dogs let's say uh express how seem to have emotional responses not completely different from humans and they even you know they have they they they do things equivalent to smiles and growls and things with that with their mouths that aren't that different from what we humans do and so the question then is is there a sort of space of possible emotions uh how how detailed is it and so on with emotions we have some vague idea that perhaps the levels of different neurotransmitters in our brains are related to different emotional states and while in our brains there might be you know a hundred billion neurons that are all doing different things it could be that some characteristics of emotions are more a feature of these kind of uh sort of these these chemicals that sort of wash through our brains to have an overall effect rather than the detailed characteristics of of individual neurons and so on and it's conceivable that the different neurotransmitters of which they're a comparatively small number might have something to do with these different dimensions of sort of emotional response it's not clear if that's correct when it comes to personalities i think it's it's you know one might wonder is it the case that sort of personalities are associated with some perhaps physical aspect of the brain it used to be the case a hundred years ago or more a little bit more than 100 years ago there was this strange subject of phrenology and the idea of phrenology was based on the shape of somebody's head you could tell aspects of that person's personality now it's it's an interesting thing because they're these kind of ideas in science which at the time everybody's like oh yeah of course this is right and subsequently people are like that was just a stupid idea how could anybody have thought that that was right and and you would have these things there would be these little models where people would you know a model of sort of a head where you'd mark off the different regions that corresponding to this or that thing and in the end it was a quite discredited uh kind of uh concept it's i'm not completely sure what sort of scientific work was done to fully discredit it but it sounds a bit goofy at least in any case the question would be if you have different personality types um that you could measure like extrovert versus introvert or something which seems to be a somewhat real thing because one knows just from from sort of anecdotal experience that there really are different kinds of people with respect to that i'm not sure that it's you're either an extrovert or you're an introvert it's more that's kind of a scale and it might depend on which day of the week it is which one of those you were any particular person is more into but one can sort of imagine that that's a thing now the question is if you've got a brain you say is this an extrovert or an introvert how would you tell nobody knows right now uh there's there's the sort of the so far as i know there's no known correlate of these sort of personality types that can be seen in kind of the structure of the brain i'm not even sure if so it's possible using fmri functional machine magnetic resonance imaging um a technique where you're basically looking at the amount of energy use in different parts of the brain using using an imaging device and so you can kind of get this map of what part of the brain is most active now and that's the thing where lots of people have done lots of experiments on you know when you are doing this particular task what part of the brain is most active when you are thinking about this particular kind of thing what part of the brain is most active and i don't know of correlates of that with these sort of personality types and so on so another question would be if you just try and you know ask people questions how do you respond in this situation are you one of these people who prefers to take charge of a group of people or you prefer to be the person who's told what to do are you a person who will uh uh prefers to sort of uh invent your own way of doing things versus some uh have um uh follow what people say one should do all these kinds of things you know there's these kind of surveys you can take and a question would be if you do those kinds of surveys and you ask the question if you for any given person they will answer certain things in those surveys probably they'll be fairly consistent i'm not sure how consistent i think fairly consistent probably over uh by the time they're by the time they're teenagers too much later that probably it's probably quite consistent what people answer um for any given person so you've answered all these things and that sort of puts you you say you're a person who answered this this this this for all these different questions and you can say in this sort of space of possible personality types where does that put you so in a sort of mathematical sense let's say there are a hundred questions and each one has a whole scale of answers from from zero to one let's say that's like you have uh coordinates in 100 dimensional space for where are you in this in this hundred dimensional space where one of the coordinate directions is introvert extrovert another one might be um uh i think one of these other ones is um oh my gosh uh judgmental versus something else i'd forgotten um but anyway so so is another one of these sort of personality uh parameters but but you could just almost take it from the direct answers to the questions um you could just say where do you where do you you know what were your answers to questions that puts you in one point in the sort of question answering space now the question would be let's say you take a billion people you ask them all these questions and you say you've got all these points for all these different people and how they answered and they're in different places in this kind of space of answers then the question would be does that cluster in some way is it the case that there are there are a whole bunch of people who answer more or less the same way over here and then there are a whole separate bunch of people who want some or else the same way over here with big gaps in between with big places where there aren't any people who answer the questions in this combination of ways it's always this combination or that combination so that's the sort of the question of what is the what's the structure of the space of personalities and i was curious about this at one point actually i i am i almost had a chance to to really study this because um it was a company that was a large dating site which had asked people oh and aggregate billions of questions and um this was quite a number of years ago when sort of machine learning was just getting started and we had been talking to them about using some machine learning technology that we were developing to just analyze this data and to me that would have been an interesting thing to look at because with these billions of answers you might have been able to answer this question is there a continuous range of sort of personality types based on question answering or are there clumps of personality types if there were clumps of personality types it makes it much more seem like oh there's you know people either have this feature of their brain or that feature of their brain not a continuous variation between them i mean it's like in our anatomy in general the people you know everybody is slightly different with different heights we're different uh you know we have different ratios of lengths of fingers we have all kinds of different um uh different characteristics and if you look kind of internally to us there'll be people who have you know oh there are two branches of this artery in this place oh there are three branches of this artery in this place and so on we all have certain differences but some of them are kind of discreet differences like you either got the two artery branch version or you've got the three artery branch version and they're definite different things and there isn't like the two and a half artery branch version and so one could imagine that some similar things would be true about brains brains are very very diverse and that you know one might think like fingerprints are a typical way of of having something where different people have somewhat different fingerprints fingerprints are formed by kind of a folding process during embryonic develop development uh similar to a little bit to how ears are formed and both of those things once you are folding differently you uh you end up with in the end very different looking fingerprints so to speak so the uh the question is how different are brains relative to fingerprints and the answer is they're very different when if you look at brains that they're just their structure of sort of the folds and ways that they're formed are just really different from person to person and so you know does that have an effect is that the new phrenology i don't know um people have been sort of collecting um collecting brains the collections aren't really huge yet um to to try and analyze things like differences of the of these types um i think that uh uh that's so uh there are there are still many mysteries here and i think um there's this question of of kind of are there distinctly different personality types or is it sort of a continuous range of different kinds of responses to things and also how much does it differ day to day so to speak you know when you when you fill out these tests of what would you do in this situation what would you in that situation well it may be different from day to day and i don't know how much variation there is along those lines so that's um it's a little bit of response to to that let's see um ah okay somebody's pointing out that um the entj classification the n stands for intuitive which intuitively starts with an n i suppose the t stands for thinking and the j stands for judging okay uh and uh paul mendes is commenting entj is the ceo personality type hmm interesting see that's another interesting question is uh given you know if you look at different activities that people have how does that map into personality type and which comes first the personality type or the activity i mean it could be that if for whatever reason you back into being a ceo you end up being sort of nurtured into the ntj personality type let's say or it could be that being the entj personality type is what leads you to even consider doing the ceo thing because otherwise you'd um uh you know if you if you're very um you know doing the ceo thing you know some part of that is you've got to kind of be in charge do leadership kinds of things and and there are definitely personality types where people are very uh don't like that at all um i mean i have to say i'm i tend to be of the personality type as revealed by the the entj nest i suppose that um the personality type where if i'm in the opposite situation where i'm being led i don't like that at all so it's you know different people have different um uh different characteristics in that regard so um let's see let's see so common question here from ic about is replication a necessary criterion for the validation of a scientific experiment um well let's explain what that means even so you know one of the things when you report some results in science like you say i don't know if i give you this vaccine it stops you getting this disease let's say or some more basic science result about i don't know um something about this particular um when when two black holes collide they uh merge in this or that way any of these results in science the the first sort of question is can one tell from can we do an experiment to answer the question of what happens when i don't know let's pick another example let's say if you've got a um uh a plant that's growing this is a classic experiment let me see if i can remember how this works um you've got something like a a um yeah okay let's do a really simple kind of thing let's say how does the horn of a rhinoceros grow okay one possible answer is the end of the horn just keeps getting longer another possibility is at the at the place where the horn attaches to the head of the rhinoceros more horn is added being added in that place and so the question is which of these things is actually true and so you might do an experiment you might put a mark on the horn of the rhinoceros and see how that mark moves as the horn grows that might be an example of an experiment you might do and so the question then is well you could do that experiment on one rhinoceros at one time and you might discover this thing happened and then you say but is it a repeatable is it a reproducible experiment if i do the same thing on another rhinoceros will i get the same result or is it that one rhinoceros grows at horn in this way and another rhinoceros grows its horn in that way and it's not a reproducible experiment um so a big thing and this is kind of a kind of an idea from the early days of of thinking about doing science experiments i mean it's worth saying that historically in the development of theoretical science the idea of doing experiments wasn't really well formalized for a very long time i mean people i think thought when if you say how does the world work and people were thinking about that a couple of thousand years ago in in ancient greece or something like this a lot of the answer to let's figure out how the world works is let's just think about it we can see various things happening in the world now that we've seen those things we've observed what happens now let's just sit and think and figure out why does that happen and that's how we do it now the idea that a rose particularly i would say by the by the 1500s and 1600s was this idea of no don't just sit back and look at how the natural world works and then start thinking through why it does what it does instead do experiments actually set things up and say if i do this thing then what will happen and if i do another thing then what will happen and deduce how the world works by doing experiments that you've set up where you say i want to see what happens if this or that happens okay so some experiments are very easy to make reproducible let me give an example that is in uh not about the natural world but about the kind of computational mathematical world let's say you say something like uh well very very s something like there's some this particular number is um uh has this particular characteristic you know you say 27 is three cubed okay well that's you know you might say that once but it's it's trivial to reproduce that experiment you can do the multiplication again you can do it again you can do it a different way you'll always get the same answer it's trivially reproducible the same is true if you set up some computational rule and you have a program and you now in physics there is uh up to a point a decent ability to have reproducible experiments just because one imagines one can set up the physical system has sort of has cut and dried enough characteristics you can really say i want to make it exactly the same this time versus the next time now there are some some caveats to that for example let's say you're dealing with something where it matters you're dealing with a gas the gas has molecules the molecules are bouncing around in this seemingly quite random way which is associated with kind of that and that the rate at which they're bouncing around is associated with temperature and heat and so on whenever you have kind of lots and lots of molecules bouncing around and and implementing heat basically you have a situation where you in a typical experimental setting you don't know where every molecule is you just know in the aggregate they're roughly like this and so that means when you do the experiment it can matter that the molecules in that particular time you did that experiment were arranged in exactly this or that way and so you get this result and if they arrange differently you get a different result but that's not something you're controlling in the setup of your experiment it gets even worse in quantum mechanics in the theory of sort of individual electrons and small particles and so on because in quantum mechanics there is in a sense a sort of fundamental randomness to what happens we think we know now from our fundamental theory of physics that we've developed in the last couple of years a lot more about exactly why this happens and how this works but the the observed feature is that in quantum mechanics there aren't definite things that happen there are many different possible things that can happen and we when we observe how the experiment came out with a certain probability we get this result versus that result so the idea of a reproducible experiment in quantum mechanics is just kind of out the window because in quantum mechanics the way we think it now works is actually our universe has many different possible branches of history many different things can be happening in the universe and what is going on it's a little bit of a brain twisting thing is the universe is kind of branching into all these possibilities so are our brains branching into all these possibilities and so the kind of issue is how does a branching brain perceive a branching universe and it can happen in different ways leading us to conclude that there's a different result from any particular measurement and so but that process of kind of how our brains sort of align with the universe so to speak in making a particular measurement that process is not something we can control for in our experiment and so there is in in some very theoretical sense there is no reproducibility in quantum mechanics because that is to get reproducibility you have to control all aspects of things and that aspect is something that with the way that we exist in the universe we are simply not in a position to control it is not self-evident that we are the only way you can exist in the universe and in fact although i think it's a fundamental feature of our kind of consciousness and our kind of perception of things like the fact that we have the notion of a definite thread of experience over time those features of us which are pretty central to the way that we work those features of us force you into this situation like quantum mechanics is at least for us not perceived as reproducible and so that means that the idea of reproducibility in experiments as soon as we're dealing with quantum mechanics basically the existence of consciousness and a definite threat of experience and so on forces us to not have access to the things we need to have access to in order to make quantum experiments reproducible now if you say well how can you conclude anything about quantum mechanics the answer is that although you can't say what the outcome of any particular quantum experiment will be you might be able to say the probability very accurately you might be able to say exactly 60 of the time it will do this and well what does that really mean that means if you ran the experiment a million times roughly six hundred thousand times it would do this but it isn't exactly six hundred thousand times whenever you say there's a certain probability of something it just means on the average in the long run that will be the frequency of time something happens it doesn't mean in any particular thing that you do that you'll get that result so a a characteristic example of that is coin tossing if you say you know if you have a an unbiased coin half the time it will come up heads half the time it will come up tails but if you say uh you do an actual experiment where you toss the coin a bunch of times and you say well what's the result you'd say well it should be 50 50. 50 heads 50 tails but any particular experiment you do it won't be exactly 50 heads and 50 tails it will have some sort of random walk fluctuation around that value and so sometimes let's say you actually observe that it was 70 heads after you did 100 experiments it came out 70 heads you would say so then the question would be and assume that this coin was sort of tossed randomly each time well you would say what does it mean that it came out 70 heads does that mean that actually this coin is biased does that mean we know for sure the coin is biased or what does it really mean or is that just bad luck and if we just kept on flipping the coin more it would even out and eventually it would be 50 50. that's a complicated question that's a sort of central question of statistics this question of if you see that it is 70 heads what does it mean what does that mean for for example the probability that the coin is a biased coin this all relates to a thing called beza's theorem which is sort of a way of backing out given what you saw what can you tell about what was going on in the in the model underneath and the end result of of kind of thinking about that is it's it's ultimately you can't get something for nothing and ultimately to answer the question i just answered about or just asked about what does that mean for the coin in the end you have to have an underlying model for what possible coins could be produced you can't answer the question i just posed without knowing something which is sort of a further back issue which is imagine that you have a lot of different coins what fraction of them are biased in what way if you can answer that the so-called prior a priori uh kind of um assumptions if you can answer that then you can go from the observation oh we saw 70 heads or something to what the chance that the coin is biased would be but you never get something for nothing you always have to have some sort of fundamental assumption down there but any case so so one reason that there can be science even when an indefinite quantitative results that are these kinds of results that say with probability x such and such will happen but that's not going to be a perfectly reproducible experiment that's going to be an experiment that is somehow statistically reproducible okay now here's where things get really difficult when you go into for example biology and you say can you make a reproducible experiment in biology well that gets quite difficult because you know every every time you do the experiment you know every mouse has a slightly different state you know even if and people do this a lot to breed genetically identical mice that they keep in these giant very automated mouse colonies for for in labs and so on they they breed these mice that are genetically identical but even genetically identical mice aren't absolutely identical it depends what did the mouse do that day did the mouse eat this did the mouse do that it's never precisely controllable how things work and so that makes it much more difficult to do biology experiments and to know you know you say well i did this on 10 mice and i got the same result each time and then i do it on 11th mouse and oh by it's different what happened well it if sometimes if you can figure it out you can say well the reason it was different is because that mouse was older than the other mice or the reason it was different is because that mouse had gone through this maze the day before and done this or something um but it's very hard usually to figure out what it was that was different and this fights one incredibly strongly in medicine because you know us humans are not all genetically identical and unless we happen to have identical twins or something and we have many differences both in our genetics and in our uh you know what have we eaten all of our life what exercise we take in all of our life what experiences we had all of our life we have lots of differences and so when you say we're going to test this particular medicine on people and say does it work does it not work well it might depend on a lot of things and teasing out for example the the the sort of the definitive science question does it work is a really hard thing to answer because the answer might be well it works with people who have this characteristic and that characteristic but not this characteristic and there are so many characteristics that people might or might not have that it's really difficult to do that experiment you might have to i mean i suppose that in things like the vaccines that are going on right now there's literally billions of people getting these vaccines and so those are have the capability to be in a sense as good by biology experiments as pretty much you'll ever get i mean you can do experiments on at the level of cells where you can probably sample even more cells than we have humans sampled in this particular case but the you know the question will be well what can you actually conclude you know how does this work and that work and so on and the answer tends to be it's really quite complicated and the notion of a pure outright it's a reproducible experiment is not so much of a thing and it's a difficult thing because for example let's give an example of how this impacts the world at large so if you're going to say there's a drug some drug company has produced it you're going to say is this a good drug or not so then what you do is the the basic standard of how that's figured out these days is to do a clinical trial you say we're going to give some people a drug we're going to give some people not knowing that they didn't get the drug we're going to give them a placebo where they they they might have got the drug but they didn't actually get the drug we're going to have these two different groups of people and we're going to see did the drug do really good things for people who got it and did nothing interesting happen for the people who didn't get it and if the answer is the drug did really well for the people who got it and uh the things that that did well for the people who got the drug didn't happen for the people who got the placebo and the people who got the drug didn't have some horrible side effect and so on then we say great we've got a drug that works now go feed it to another billion people okay so now the problem is what tends to happen is that well it works on some people it doesn't work on other people it has side effects for some people it doesn't have them for other people okay what do we conclude one thing we might conclude is as soon as anything goes wrong we might say forget it we're out of here this drug is a failure and unfortunately that happens a lot but the truth of the situation is a little bit more of the reproducibility of experiments question it is probably the case that the people if the drug worked on some people but failed on others gave side effects on others that really there's different characteristics of those people it's not just pure chance that oh well you know it so happened that this happens just chance it's it's undoubtedly not chance it's undoubtedly because the person who had the side effect is a person who has a different uh you know slightly different version of this enzyme that came from their genetics that blah blah blah blah blah whatever else but the question is can you tease that out can you figure out that there's that kind of thing and one of the things people hope is that in the future there will be a much more personalized version of medicine where perhaps this particular drug with this particular particular molecule with a particular shape and so on it'll be like well for you we need a modified version of that molecule that has this extra few atoms in this place because the place it's going to bind inside your body is you know is for genetic reasons has a slightly different configuration so we need to give you this sort of personalized changed molecule for this drug now the question is if you say well we're going to test you know for drugs to be okay we're just going to test them using this on off it works it doesn't work kind of approach that's a very difficult thing to do if everybody gets a custom version of a drug um and quite how that's going to be worked through because obviously one doesn't want to just say oh just do whatever you want it's you know because all sorts of terrible things will happen um but then the question is how do you essentially validate a a um a whole kind of algorithm for producing drugs or even more than that if the drugs themselves are molecules which when they go into your body actually start algorithmically as a molecule doing computations and reconfiguring themselves and so on it's like you know the the molecule goes in as a generic molecule but when it's inside you when it detects that there's some particular kind of cell it modifies in this way and so on and so on and so on how do you validate that that all the right things are going to happen and uh to some extent issues like that have been addressed in medical devices a bit but not completely um and i think that's a sort of a challenging thing and it would be a real real shame if a lot of things that would be super helpful medically just can't be done because nobody can figure out the kind of reproducibility of experiment story in a way that that everybody's happy with um for for those kinds of um uh for those kinds of things let's see um oh this is some coming back to um uh earlier question here about personality types from a toray wondering if personalities are clustered and wondering if their data sets available on this uh the data set that we uh sort of almost had access to from this uh dating site i think was sort of the biggest one at the time that was really the biggest realistic such data set um uh and i i think that's the kind of place such data sets would have been collected is um uh is those kinds of places um let's see ah oh boy um there's a question here about um from slayer i'm afraid uh about a a weird medical phenomenon a chap called phineas gage who had a very unfortunate accident that caused a giant rod to get um go through his his brain and um uh i don't know all the all the um characteristics of what happened you know one of the general principles about medicine is you know you think there's a kind of cut and dried answer to something you know if you destroy this part of the brain it's all over or something and then you find oh but there's an exception here's a person who survived this particular thing well here's a person where this characteristic just worked in a different way with respect to the brain and one of the big questions about the brain is if you sort of look at a brain it has a complicated structure you know you can just see it has a complicated structure but also each different part of the brain is quite specialized in what it does you know there's a piece that that is controlling you know there's a there's a definite area that controls the muscles in your you know left index finger or something there's a definite area that controls you know kind of the um uh what you see in this part of your visual field and how you do the first levels of a vision from this part of your visual field there's a definite part of your brain that is associated with for example language well one of the things that happens is some things are in pretty fixed locations uh like the things to do with movement the things to recite they're in pretty fixed locations the things like how you understand language that look at that moves around a bit between different people but there's sort of a big question of to what extent is the brain retargetable and what it does that is if you if some part of your brain fails in some way you know let's say you have a a stroke and that part of your brain is deprived of oxygen and those cells die or you have some nasty concussion and you kind of tear apart a bunch of nerve nerve cells and so on all definitely bad things to do but um uh you know when when that happens is it the case that some other part of the brain can kind of come in and take over the functions that were being done by the part of the brain that was damaged and the answer seems to be well to some extent yes uh particularly if it happens you know if the damage occurs when one's very young then it is it is often the case that one can sort of have other parts of the brain come in and uh take over those functions and all is good so to speak it's sort of interesting that that's possible because there's sort of the question of one might have thought a brain is a catholic sort of controlled machine where every piece does a definite thing and each piece is necessary for some aspect of how we operate but the kind of the big thing that was discovered that's sort of the foundation of the idea of software for computers is that you can have a device like a computer which when you just feed a different software it will do all these different kinds of computations it will do all these different kinds of things and so you might imagine that brains are a little bit the same idea that is that there's a fixed hardware all of those hundred billion neurons and so on but that depending on the software depending on what actual memories are in the brain and so on that you can have the brain do different kinds of things and i think that it tends to look as if uh in many respects the brain can be sort of rearranged and like this thing that was really most often does this for in people can be made to do that in people instead clearly when people have different senses like for example they you know they they don't have good vision for example they don't have good hearing those parts of the brain that were responsible for vision or hearing um end up being you can use those neurons for something else and so a quite common experience is people who have difficulty with one of those senses the other senses become more acute and probably more neurons are recruited to to make use of those things than them would be otherwise um so so that's um uh so i mean i think this question of just how retargetable at different parts of the brain is an interesting one and i don't think one really knows the full answer to that um there will be other aspects of that question which will come in when we have um if we have when we have sort of better direct connections to the brain you know the brain is an electrical device in effect where all these neurons are producing electrical signals that they're transmitting to the neurons that they're next to and all those kinds of things and so one of the questions is can we just connect into that electrical network and just wires things into our brains well that's something that's done as a sort of medical clinical thing particularly in the case of parkinson's disease um where the um i guess it's dopamine uh one of the neurotransmitters is does not get produced as much as it normally would by the cells that normally produce it and one of the ways to kind of fix things is to actually have a a stimulator that is a a an electrical device implanted in the brain that uh and and so that that's a rather coarse example and there are similar examples in things related to epilepsy where there's sort of a a let's recontrol what the brain does by feeding it electrical signals that will potentially uh divert what it was otherwise going to do now one can imagine a much finer version of that and people do lots of experiments on this trying to get something that's like like the technology microprocessors sort of directly with all those detailed little wires and things being able to directly connect to neurons in the brain and be able to sort of monitor for example have a raise of of electrodes that kind of monitor what's happening in many different neurons in the brain one day we might be able to kind of pick up the kind of internal communications between neurons and the brain uh it will be not easy to decode that i mean the thing that's probably the more immediate thing in which i think there's been some success on is uh things like in the spinal cord where one's using the bundle of nerves that control uh you know all of one's movements and and so on um a question of come one come on have an electrical device that is just sort of sees all of those nerve endings and can figure out which nerve is doing what and can produce its own version of that stimulation and there's some degree of success nowadays with that kind of thing but i think that the um the question is if we were able to pick up if we were to monitor a very fine level what's happening in all those electrical communications inside the brain could we figure out what those communications are about and if we were able to feed in new things could we like let our brains directly do arithmetic for us without not rather than arithmetic being done by our neurons say well i'm just thinking about what is uh 13 times 39 or something and just by thinking about it it would kind of the the electrical connections would be you those thoughts would be decodable and then you could just go off and ask your computer to do that computation and feedback the results that's sort of a challenging question to can we sort of can we can we detect that internal communication in the brain understand it and go back and and and put something else in now now the thing to understand is that brains the communication between brains doesn't seem to be such a simple matter human language is the main way we have to communicate between brains and in a sense what's happening is you know as i'm yakking away here no doubt you know a billion neurons in my brain are doing all kinds of nerve firings and sending all kinds of electrical signals and so on and the result is that i say fool or whatever and the then for you to hear that word then that causes a bunch of nerve firings in your brain and then that causes you to have certain thoughts within your brain that are based on other thoughts you've had and other things you remember and so on and it's kind of in the in the computer sense it's sort of the api that we have between brains the way of communicating brain to brain is human language and we're sort of putting it into that we're taking the inner thoughts that we have which involve nerve firings of billions of neurons and so on and we're arranging those into this thing with is this essentially api this communication channel that we're able to communicate to another brain and the question is is there a direct kind of brain to brain or or brain to machine kind of interface where you're not turning everything into this whole elaborate language you're just looking at the level of kind of the individual nerve firings and so on a slightly cautionary tale is what would happen with computers if i opened up my computer and i put little wires and probes in different places in its cpu and i said what's going on here it'd be pretty hard to tell you know back in the day i have to say i haven't tried this in years but back a long time ago like like uh 30 40 years ago if you took a radio a standard kind of am radio and you put it next to a computer and the computer was doing things the radio would go crazy because there are all these electromagnetic emissions from the electrical activity in the computer and you'd be essentially hearing some version of what the computer is doing but what you'd heard you know with all the squawks and squeaks and this and that and the other it's very hard to decode it's very much and if you get down to the level of there's individual wires and there's individual values in this register in the cpu et cetera et cetera et cetera decoding what's going on reverse engineering what's going on from that very low level representation is is pretty difficult and that's what we're kind of signing up for if we say well we're going to take what happens in the brain and look at these individual nerve firings and so on and see what they mean now it turns out machine learning which is kind of an idealized version of how brains work is probably one of the best shots we've got at going in and looking at these detailed oh there's this particular set of no firings what does it mean um that machine learning is basically well here are some examples of what it means then machine learning can kind of extrapolate from that and say so in this different case that i'm seeing now this is what it means here so that's probably our best hope for being able to sort of decode the inner language of of of brains even though there might not be we don't know to what extent there is in a language we don't know whether there's a level in between the me talking to you in english type thing and the level of individual neurons firing and doing what they do we don't know if there's some way that there is a sort of an intermediate language where it's like not as beautifully formed as english but not as just write down the level of individual bits of the individual nerve firings it's a it's a sort of an interesting and fundamental challenge for neuroscience is to figure out is there such an intermediate language we don't know the answer yet let's see uh well it's a question here from prath prathamesh asking about ai politics and decisions gosh i'm not sure there's many different aspects to that that um uh that those words one of the things that um we might wonder about is when we set up machines and the machines kind of in some sense decide what to do how do we feel about that so at some level we we have certain devices we build and sometimes we can readily understand we built this device we move this lever in this way the device is going to engage this wheel in this way and it's going to move in this way and we can kind of see what's going to happen and we kind of know we set this device up we pull this lever we know what's going to happen if you pull the lever you know what's going to happen you're responsible for what happens if you pull that lever all quite straightforward now let's imagine that in that loop is this complicated ai system that has all these things where it's a it's a self-driving car and it's figuring out based on some very complicated algorithm and this and that and the other is it going to swerve when it sees a a deer jump into the road or not and is it going to swerve and if it swerves does it swerve into another car and how does that all work and who is responsible for what goes on there now one of the things you might say is well whoever wrote the program they're responsible for what goes on there but that's not a realistic thing to say because we know that i mean my own work actually in this area is is kind of makes it clear that there's this phenomenon of computational irreducibility that is you write a program the program specifies rules but when you run those rules and you run the millions billions of times or whatever to know what consequences that will have is not something that you can foresee in advance you just have to run it and see what happens so you can't just sort of say well i'm going to make sure that my device never does the wrong thing look i'm going to tweak this piece of code to make sure it never does the wrong thing that's not realistic because knowing what it will do there's this irreducible amount of computation that is involved in figuring out what it does in any particular case and that's not something where you can just say i'm going to guarantee it's going to do nothing bad the next question is what do you even mean by nothing bad what do you mean in a situation where you say um you know is this thing going to successfully shut down this uh giant motor or something well that's a very well-defined objective and you can say did you succeed did you not succeed if the question is did this uh self-driving car ever do the wrong thing ever do something you didn't want it to do that's a much more complicated question let's take an example that's a very fraught question which is autonomous weapons so you've got some drone running around and the drone has a gun and the gun can shoot at things and the question is under what circumstances should the the drone shoot at something and it's like you say well i've got certain rules i don't want to do it in this case i do want to do it in this case but it's it's impossible to specify with all different eventualities of what could happen what if the um i don't know you know what if you say uh leave the animals alone this is a problem between humans but yet you have the trained killer pigeon or something you know what do you do in that case the um it's uh uh and all these different things where a human might say well based on my whole story of judgment of this and that and the other i decide to do this but this ai system has been set up with certain principles and so on and it's but it's not uh well one of the worst cases is let's say the ai system is just learnt by from examples let's say it had a thousand examples of what to do and then it's out in the wild and it's told just go do the right thing based on those thousand examples the problem is the thing that happens out in the wild may be so different from any of those examples it has no idea what to do and sometimes you don't even know it has no idea what to do because it just learnt from those examples and it built up some whole sort of machine learning sort of idea of what to do and it didn't um and you don't know that it doesn't know and he it doesn't know that it doesn't know um because it just follows the principles that it's been operating according to and it just it ends up that well actually if there'd been this critical extra example given it would have done the right thing but there wasn't and it didn't know that there was a hole there and you didn't know there was a hole there now i tend to think that a better approach is computational contracts effectively for for devices what does that mean that means something where just like you would like to write a legal contract where you write out in english or some other language or some kind of legalese this is what i want to have happen in a way where where both sides both parties can read the document and say yes we think we know what this means we think we understand it and therefore we can follow it in the case of machine learning for example where you just say let's learn by example you don't really have that situation at least not as directly you are much better shape if you say well i'm going to write this computational language that is a thing that i can read and i understand what it says and okay now ai go follow this principle that's a better situation to be in now of course you have to have a language that can express the kinds of concepts that are relevant for telling the ai do the right thing that's something that actually in in our technology development over the last uh 35 years or so on um we've we've come a long way to having kind of a computer full-scale computational language that can kind of express all these different kinds of things in the world and ideas about what should happen in the world and so on and so i think that's a sort of a good direction for that um but uh um it's uh let's see i think that um um uh well let's see parmenides is commenting how crazy would it be if politics becomes like hedging where it's all run by algorithms vote for algorithm one for posterity algorithm three is tough on crime etc etc etc yeah well you know that's kind of what i'm sure that's basically what's going to happen i mean in a sense that's the story of democracy you know you say you might say uh you know vote for party x you know but in a sense you're voting for the algorithm of party x you're voting for whatever you know or you might say well i'm voting for the character of of party x but in a sense it's not as cut-and-dried as there's this computational contract and this is what you're voting for it's more you you're you're kind of prox you're saying i want things to be run according to the procedure that is associated with party x or person y or whatever else and so in a sense that's what's happening except it's not implemented it's implemented with a human system rather than a computational system now the question is what happens if it is a computational system what happens if for example you say i'm voting for this computational contract okay here it is it's uh you know it's 50 pages long it is the description of what i want to have happen in the world how i want the country to be governed or whatever else it is and it is like a constitution like the one that a country like the us and many countries in the world have a constitution which is sort of supposed to be the foundation for kind of how the country is is governed um but in this case rather than it being just a foundation it's interpreted and it's got millions of pages of legal opinions and things that are have been added to that sort of underlying set of principles that instead you say well no actually it's just going to be this piece of code and that's it and that's what decides what happens and you know people will say well i don't like that piece of code and maybe they have some democratic process where they say we're voting out that piece of code we want this piece of code instead and people would argue you know they would draft a new piece of code and people would spend a long time kind of developing you know what what is that let's put an addendum to this piece of code that deals with this complicated situation and what happens if this etc etc cetera and then there might be three variants of it and people might vote we want this variant versus this variant and so on but it's all in the sense code um is that a good situation well is that is that a situation that can and will happen i'm sure it will happen um and whether it happens in that rather transparent form or whether it happens in a much less transparent form where it's like i vote for this and where this is a human system and that human system writes the code and then you're stuck with what the code does but the the point about code making decisions that's inevitable and that's already happening to a large extent i mean basically what you know when decisions have to be made quickly or in bulk whatever else it makes it is completely impractical for humans to make those decisions whether it's in let's say in financial trading kinds of things you know should i buy this should i sell this oh i'm doing it a thousand times a second well no human is going to be able to do that or i i'm taking in all of these signals for what i might buy and sell just like one might imagine a lot of the kind of controls that governments have are you know that they're comparatively coarse it's like what are we going to do with interest rates what are we going to do with money supply these are levers that you turn they don't get turned you know it's not like it's not like an airplane flying where you see those little um uh things at the back of the wings that flap up and down automatically controlled by a control system to try to keep the airplane flying in a level flight or whatever you know it's not the case that in the financial system at least in the government financial system that there's a similar kind of rapid control system where something is kind of flapping you know every day doing this or that at least not that i can think of that certainly happens in the private sector of finance but it does not happen in the public sector of finance i believe um and one can imagine a situation where where people say well actually we want that to happen install you know at the central bank install this algorithm basically install this computational contract that's going to determine what the interest rates are reset every second or something well uh you know in those kinds of things you can have a local version of that or you can have something more global where you say this is the overall set of things we're trying to achieve now optimize these thousand parameters we'd never specify all those thousand parameters separately but you can say you know you you're setting up oh i don't know tariffs on this these goods and and this particular tax on this thing and this particular whatever those are the current controls of government um and uh you know i i happen to be a the less controls there are from government the better but that's just a personal preference um the uh uh i think that um uh it's something and and um no doubt um but but uh uh it's something where where um you can you can kind of ask um maybe that goes with the entj personality type i don't know um but uh uh that's another interesting issue is let's say you have a situation where you uh that's a that's a bizarre possibility um you know you've got something where you sort of are interacting between you know people with all their complicated preferences and all of their um the um uh um the uh um um uh with them uh yeah it's um it's it's it's complicated i'm getting a note here that reminds me i'm late for something and it notes that it's a it's a meeting with an isp fp which is apparently the exact opposite of an entj so uh so so it goes um all right maybe maybe um uh um maybe i'll i'll um do just there's there's lots more to say about um uh about kind of um the the interaction of ai with politics with this and that um i i'm not a person who knows much about politics as such but but um the um uh the thing that i think is kind of the the challenge is in a situation where it is possible to have a finer delegation of things than just saying i vote for party x where you can specify things in more detail with computational contracts and so on what does that mean and do you end up with does every person effectively have their own political party where they say i i declare that i want the country to be run according to this computational contract i wrote it i started writing it in grade school i kept on adjusting it my whole life and this is my contract this is what i want the world to be like and now at the age of 40 or something my contract is 8 000 lines long and it's got all this complicated cases for this and then the other this is what i want the world to be like and you then imagine the situation where okay you collect everybody's you know preferences and you put them all into some big machine and you say okay now figure out what to do because unfortunately it is the nature of society in the world and the physicality of the world and so on that uh you know it's not possible to give everybody everything that they want um and because sort of everybody has to live in the same space so to speak and interact and so on and so then the question is how do you optimize given all these different preferences that came in how do you come up with the optimum and it's a little bit like what i was saying before with statistics that in the end there has to be a base base base base based set of assumptions without that you really can't resolve that question and and even with that it's like for example a very obvious question is if there are things that will make 80 of the world really happy and 20 really unhappy and whereas there are other things that will make uh 60 you know 50 percent of the no will make everybody reasonably happy what is the trade-off between making 80 really really happy and 20 really unhappy versus everybody's kind of happy but not as happy as they would have been and not as unhappy as they would have been which of those is the better world which of those do you want to have happen now people might put that very statement into their computational contracts this is what we want for for that balancing of things but that's a very complicated infinite regress and and there won't be a resolution unless you somehow at the bottom of it say this is what globally we decide wants to have happen and you know i think that's a it's a it's a feature i suppose i'm sure that there are echoes of this throughout the history of political philosophy and in a sense we're pretty lucky that the world is set up so that it has for example multiple countries where different things can be done different sort of core assumptions can be made and i think it's some so i mean and then there's the whole question which is which again turns into all kinds of complicated politics about you know in in the world of the future let's imagine that we're all uploaded to some giant digital thing and we don't have to worry about the physicality of do we move from country x to country y physically in the physical world it's all digital and just by merely sort of a a puff of electrons we could be transported into a different into a different environment um you know how then would uh you know would you just say to people well just go to where you want to be of course that doesn't work either because it could be the case that um uh you know you end up in a situation where where the world has to be maintained in some sense and everybody wants to do this everybody wants to be a um i don't know everybody wants to be a painter and nobody wants to make the trains run well that's not good now again in some sort of sense of some sort of digital utopian something or other where where there is no physicality there are no trains to be running maybe you can maybe there's a way to deal with that or even then i suspect it isn't the case i suspect it's sort of inevitably the case that if everybody wants to paint and nobody wants to make the trains run then uh that's not a good situation and then what do you do with that and and you're you're thrust back into the same kind of complicated morass that um that is the sort of the long-term history of political philosophy right well that gets me into um uh um um yeah there are all kinds of interesting questions here which i haven't had a chance to address this time but that's okay because there'll be another time next week um and uh should wrap up here for now and i look forward to um uh seeing you again well i wish i could see you all but uh at least um um in some sense being with you all same time next week all right thanks very much and uh bye for now

Channel: Wolfram

Views: 1,230

Rating: 5 out of 5

Keywords:

Id: Tju1oq1wqZ4

Length: 69min 45sec (4185 seconds)

Published: Fri Sep 03 2021