Mark Twain summed up what I take to be one of the fundamental problems of cognitive science with a single witticism. He said, "There's something fascinating about science. One gets such wholesale returns of conjecture out of such a trifling investment in fact." (Laughter)
Mark Twain resumiu o que eu considero que é un dos problemas fundamentais da ciencia cognitiva cunha sinxela ocorrencia. Dixo, "A ciencia é fascinante. Conséguense cantidades masivas de conxecturas a partir dun investimento tan insignificante en feitos.” (Risas)
Twain meant it as a joke, of course, but he's right: There's something fascinating about science. From a few bones, we infer the existence of dinosuars. From spectral lines, the composition of nebulae. From fruit flies, the mechanisms of heredity, and from reconstructed images of blood flowing through the brain, or in my case, from the behavior of very young children, we try to say something about the fundamental mechanisms of human cognition. In particular, in my lab in the Department of Brain and Cognitive Sciences at MIT, I have spent the past decade trying to understand the mystery of how children learn so much from so little so quickly. Because, it turns out that the fascinating thing about science is also a fascinating thing about children, which, to put a gentler spin on Mark Twain, is precisely their ability to draw rich, abstract inferences rapidly and accurately from sparse, noisy data. I'm going to give you just two examples today. One is about a problem of generalization, and the other is about a problem of causal reasoning. And although I'm going to talk about work in my lab, this work is inspired by and indebted to a field. I'm grateful to mentors, colleagues, and collaborators around the world.
Twain quería facer unha broma, claro, pero ten razón: A ciencia é fascinante. A partir duns cantos ósos, inferimos a existencia dos dinosauros. Das liñas espectrais, a composición das nebulosas. A partir das moscas da froita, os mecanismos da herdanza, e de imaxes reconstruídas de sangue fluíndo a través do cerebro, ou no meu caso, do comportamento de nenos moi pequenos, intentamos dicir algo sobre os mecanismos fundamentais da cognición humana. En concreto, no meu laboratorio no Dpto. de Cerebro e Ciencias Cognitivas, no MIT, pasei a última década intentando entender o misterio de por que os nenos aprenden tanto, a partir de tan pouco, e tan rápido. Porque resulta que o que a ciencia ten de fascinante téñeno tamén de fascinante os nenos, e é, dicíndoo de forma máis suave ca Mark Twain, precisamente a súa capacidade de extraer inferencias ricas e abstractas de forma rápida e precisa a partir de datos dispersos e confusos. Vou dar só dous exemplos hoxe. Un deles aborda un problema de xeneralización, e o outro un de razoamento causal. E aínda que vou falar do que facemos no meu laboratorio, este traballo está inspirado por un campo e en débeda con el. Estoulles agradecida a mentores, colegas e colaboradores de todo o mundo.
Let me start with the problem of generalization. Generalizing from small samples of data is the bread and butter of science. We poll a tiny fraction of the electorate and we predict the outcome of national elections. We see how a handful of patients responds to treatment in a clinical trial, and we bring drugs to a national market. But this only works if our sample is randomly drawn from the population. If our sample is cherry-picked in some way -- say, we poll only urban voters, or say, in our clinical trials for treatments for heart disease, we include only men -- the results may not generalize to the broader population.
Quero comezar co problema de xeneralización. Xeneralizar a partir de pequenas mostras de datos é o pan de cada día da ciencia. Entrevistamos unha fracción mínima do electorado e predicimos o resultado das eleccións nacionais. Vemos como un puñado de pacientes responde a tratamento nun ensaio clínico, e incorporamos fármacos ao mercado nacional. Pero isto soamente funciona se a mostra se extrae aleatoriamente da poboación. Se a nosa mostra ten algunha manipulación --por exemplo, entrevistamos só votantes urbanos, ou nos nosos ensaios clínicos de tratamentos para doenzas cardíacas incluímos só homes-- os resultados poden non ser xeneralizables a toda a poboación.
So scientists care whether evidence is randomly sampled or not, but what does that have to do with babies? Well, babies have to generalize from small samples of data all the time. They see a few rubber ducks and learn that they float, or a few balls and learn that they bounce. And they develop expectations about ducks and balls that they're going to extend to rubber ducks and balls for the rest of their lives. And the kinds of generalizations babies have to make about ducks and balls they have to make about almost everything: shoes and ships and sealing wax and cabbages and kings.
Por tanto aos científicos impórtalles se a mostra se recolleu ou non ao chou, pero que ten iso que ver cos bebés? Os bebés teñen que xeneralizar seguido a partir de pequenas mostras de datos. Ven uns poucos parrulos de goma e aprenden que flotan, ou algunhas pelotas e aprenden que botan. E desenvolven expectativas sobre os parrulos e as pelotas que aplicarán a uns e outras o resto das súas vidas. E os tipos de xeneralizacións que deben facer sobre parrulos e pelotas, deben facelos para case todo: zapatos e barcos e lacre e verzas e reis.
So do babies care whether the tiny bit of evidence they see is plausibly representative of a larger population? Let's find out. I'm going to show you two movies, one from each of two conditions of an experiment, and because you're going to see just two movies, you're going to see just two babies, and any two babies differ from each other in innumerable ways. But these babies, of course, here stand in for groups of babies, and the differences you're going to see represent average group differences in babies' behavior across conditions. In each movie, you're going to see a baby doing maybe just exactly what you might expect a baby to do, and we can hardly make babies more magical than they already are. But to my mind the magical thing, and what I want you to pay attention to, is the contrast between these two conditions, because the only thing that differs between these two movies is the statistical evidence the babies are going to observe. We're going to show babies a box of blue and yellow balls, and my then-graduate student, now colleague at Stanford, Hyowon Gweon, is going to pull three blue balls in a row out of this box, and when she pulls those balls out, she's going to squeeze them, and the balls are going to squeak. And if you're a baby, that's like a TED Talk. It doesn't get better than that. (Laughter) But the important point is it's really easy to pull three blue balls in a row out of a box of mostly blue balls. You could do that with your eyes closed. It's plausibly a random sample from this population. And if you can reach into a box at random and pull out things that squeak, then maybe everything in the box squeaks. So maybe babies should expect those yellow balls to squeak as well. Now, those yellow balls have funny sticks on the end, so babies could do other things with them if they wanted to. They could pound them or whack them. But let's see what the baby does.
Entón aos bebés impórtalles se o pequeno anaco de proba que ven representa de forma plausíbel unha poboación maior? Descubrámolo. Vou amosar dous vídeos, un por cada suposto dun experimento, e como só se verán dous vídeos, só se verán dous bebés, e un par calquera de bebés difire de calquera outro de innumerábeis formas. Pero estes bebés, por suposto, representan aquí a grupos de bebés, e as diferenzas que se van ver representan as diferenzas grupais medias no comportamento dos bebés en cada suposto. En cada vídeo verase un bebé facendo tal vez xusto o que se agardaría que fixese, e dificilmente podemos volver os bebés máis máxicos do que xa son. Pero para min o máxico, e ao que quero que se lle preste atención, é o contraste entre estes dous supostos, porque o único que difire entre os dous vídeos son os datos estatísticos que os bebés van observar. Imos ensinarlles unha caixa de bólas azuis e amarelas, e a que era a miña estudante graduada, hoxe compañeira en Stanford, Hyowon Gweon, vai sacar tres bólas azuis seguidas desta caixa, e despois de sacalas, vainas apertar, e as bólas van chiar. E se es un bebé, iso é como un charla TED. Non pode haber nada mellor. (Risas) Pero o importante é que é moi sinxelo sacar tres bólas azuis seguidas dunha caixa que ten sobre todo bólas azuis. Poderíase facer cos ollos pechados. Pódese admitir que é unha mostra aleatoria desta poboación. E se podes meter a man aleatoriamente nunha caixa e sacar cousas que chían, ao mellor todo o que hai na caixa chía. Así que tal vez os bebés deberían esperar que as bólas amarelas chíen tamén. As bólas amarelas teñen divertidos paus nun extremo, que permiten facer con elas outras cousas se se quere. Poderían axitalas ou bater con elas. Pero vexamos qué fai o bebé.
(Video) Hyowon Gweon: See this? (Ball squeaks) Did you see that? (Ball squeaks) Cool. See this one? (Ball squeaks) Wow.
(Vídeo) Ves isto? (A bóla chía) Viches iso? (A bóla chía) Xenial. Ves estoutra? (A bóla chía) Uaau.
Laura Schulz: Told you. (Laughs)
Díxenvolo. (Ri)
(Video) HG: See this one? (Ball squeaks) Hey Clara, this one's for you. You can go ahead and play. (Laughter)
Viches esta? (A bóla chía) Clara, agora esta é para ti. Veña, podes collela e xogar. (Barullo) (Risas)
LS: I don't even have to talk, right? All right, it's nice that babies will generalize properties of blue balls to yellow balls, and it's impressive that babies can learn from imitating us, but we've known those things about babies for a very long time. The really interesting question is what happens when we show babies exactly the same thing, and we can ensure it's exactly the same because we have a secret compartment and we actually pull the balls from there, but this time, all we change is the apparent population from which that evidence was drawn. This time, we're going to show babies three blue balls pulled out of a box of mostly yellow balls, and guess what? You [probably won't] randomly draw three blue balls in a row out of a box of mostly yellow balls. That is not plausibly randomly sampled evidence. That evidence suggests that maybe Hyowon was deliberately sampling the blue balls. Maybe there's something special about the blue balls. Maybe only the blue balls squeak. Let's see what the baby does.
LS: Non teño nin que dicir nada, verdade? Vale, está ben que os bebés xeneralicen propiedades das bólas azuis ás bolas amarelas. E é impresionante que poidan aprender imitándonos. Pero sabemos iso dos bebés dende hai moito tempo. A pregunta realmente interesante é que ocorre cando lles amosamos aos bebés exactamente a mesma cousa, podemos asegurar que é a mesma porque temos un compartimento secreto e en realidade sacamos as bólas del, pero esta vez o que cambiamos foi a poboación aparente da que extraemos as mostras. Esta vez amosarémoslles aos bebés tres bólas azuis sacadas dunha caixa que ten sobre todo bólas amarelas, e saben que? Non se poden sacar aleatoriamente tres bólas azuis seguidas dunha caixa que ten sobre todo bólas amarelas. Esa non é unha mostra aleatoria. Esa proba suxire que ao mellor Hyowon estivo amosando deliberadamente as azuis. Tal vez as bólas azuis teñen algo especial Tal vez soamente as bólas azuis chían. Vexamos o que fai o bebé.
(Video) HG: See this? (Ball squeaks) See this toy? (Ball squeaks) Oh, that was cool. See? (Ball squeaks) Now this one's for you to play. You can go ahead and play.
(Vídeo) Ves isto? (A bóla chía) Ves este xoguete? (A bóla chía) Oh, que xenial. Ves? (A bóla chía) Agora esta é para que xogues ti. Veña, podes xogar.
(Fussing) (Laughter)
(Barullo) (Risas)
LS: So you just saw two 15-month-old babies do entirely different things based only on the probability of the sample they observed. Let me show you the experimental results. On the vertical axis, you'll see the percentage of babies who squeezed the ball in each condition, and as you'll see, babies are much more likely to generalize the evidence when it's plausibly representative of the population than when the evidence is clearly cherry-picked. And this leads to a fun prediction: Suppose you pulled just one blue ball out of the mostly yellow box. You [probably won't] pull three blue balls in a row at random out of a yellow box, but you could randomly sample just one blue ball. That's not an improbable sample. And if you could reach into a box at random and pull out something that squeaks, maybe everything in the box squeaks. So even though babies are going to see much less evidence for squeaking, and have many fewer actions to imitate in this one ball condition than in the condition you just saw, we predicted that babies themselves would squeeze more, and that's exactly what we found. So 15-month-old babies, in this respect, like scientists, care whether evidence is randomly sampled or not, and they use this to develop expectations about the world: what squeaks and what doesn't, what to explore and what to ignore.
LS: Acabades de ver dous bebés de 15 meses facendo dúas cousas totalmente diferentes baseadas só na probabilidade da mostra que observaron. Quero ensinar os resultados experimentais. No eixe vertical, pódese ver a porcentaxe de bebés que apertaron a bóla en cada suposto, e como se ve, os bebés tenden moito máis a xeneralizar a mostra cando é representativa da poboación ca cando está claramente manipulada. E isto lévanos a unha predición curiosa: supoñamos que sacamos só unha bóla azul da caixa que ten sobre todo bólas amarelas. Non se poderían sacar aleatoriamente 3 bólas azuis seguidas dunha caixa amarela pero poderíase sacar soamente unha. Non é unha mostra improbable. E se se puidese meter a man ao chou nunha caixa e sacar algo que chía, tal vez todo o da caixa chíe. Entón, aínda que os bebés van observar moita menos probas para chíos, e contan con moitas menos accións que imitar neste suposto dunha única bóla ca no que vimos antes, predicimos que os bebés por si sós apertarían a bóla máis veces, e iso é exactamente o que atopamos. Así que aos bebés de 15 meses, neste sentido, como científicos, impórtalles se a proba é unha mostra representativa ou non, e usan isto para desenvolver expectativas sobre o mundo: qué chía e qué non, qué explorar e qué ignorar.
Let me show you another example now, this time about a problem of causal reasoning. And it starts with a problem of confounded evidence that all of us have, which is that we are part of the world. And this might not seem like a problem to you, but like most problems, it's only a problem when things go wrong. Take this baby, for instance. Things are going wrong for him. He would like to make this toy go, and he can't. I'll show you a few-second clip. And there's two possibilities, broadly: Maybe he's doing something wrong, or maybe there's something wrong with the toy. So in this next experiment, we're going to give babies just a tiny bit of statistical data supporting one hypothesis over the other, and we're going to see if babies can use that to make different decisions about what to do.
Agora quero amosar outro exemplo, esta vez sobre un problema de razoamento causal. E comeza cun problema de proba confusa que todos temos: o feito de que formamos parte do mundo. Isto pode non parecer un problema, pero como a maior parte deles, maniféstase só cando as cousas van mal. Velaquí este bebé, por exemplo. As cousas están indo mal para el. Gustaríalle facer funcionar o seu xoguete, e non pode. Amosarei un vídeo duns poucos segundos. En xeral, hai dúas posibilidades: ou el está facendo algo mal, ou algo non funciona no xoguete. Así que no seguinte experimento, darémoslles aos bebés só unha mínima porción de datos estatísticos que apoian unha das hipóteses sobre a outra, e veremos se os bebés poden usar iso para tomar decisións diferentes sobre qué facer.
Here's the setup. Hyowon is going to try to make the toy go and succeed. I am then going to try twice and fail both times, and then Hyowon is going to try again and succeed, and this roughly sums up my relationship to my graduate students in technology across the board. But the important point here is it provides a little bit of evidence that the problem isn't with the toy, it's with the person. Some people can make this toy go, and some can't. Now, when the baby gets the toy, he's going to have a choice. His mom is right there, so he can go ahead and hand off the toy and change the person, but there's also going to be another toy at the end of that cloth, and he can pull the cloth towards him and change the toy. So let's see what the baby does.
Velaquí o plan. Hyowon vai intentar que o xoguete funcione, e conségueo. Entón eu vou intentalo dúas veces e fracasar as dúas, despois Hyowon vai intentalo outra vez e conseguilo, o que resume en xeral a miña relación cos meus estudantes de posgrao no que ten que ver coa tecnoloxía. Pero o importante aquí é que proporciona algunha proba de que o problema non é o xoguete, senón a persoa. Algunhas poden facer que o xoguete funcione, e outras non. Agora, cando o bebé consegue o xoguete, vai ter unha elección. Súa nai está xusto alí, polo que pode ir e darlle o xoguete e cambiar a persoa, pero tamén vai haber outro xoguete no bordo desa tea, así que pode tirar da tea cara a el e cambiar o xoguete. Vexamos logo qué fai o bebé.
(Video) HG: Two, three. Go! (Music) LS: One, two, three, go! Arthur, I'm going to try again. One, two, three, go! YG: Arthur, let me try again, okay? One, two, three, go! (Music) Look at that. Remember these toys? See these toys? Yeah, I'm going to put this one over here, and I'm going to give this one to you. You can go ahead and play. LS: Okay, Laura, but of course, babies love their mommies. Of course babies give toys to their mommies when they can't make them work. So again, the really important question is what happens when we change the statistical data ever so slightly. This time, babies are going to see the toy work and fail in exactly the same order, but we're changing the distribution of evidence. This time, Hyowon is going to succeed once and fail once, and so am I. And this suggests it doesn't matter who tries this toy, the toy is broken. It doesn't work all the time. Again, the baby's going to have a choice. Her mom is right next to her, so she can change the person, and there's going to be another toy at the end of the cloth. Let's watch what she does.
(Vídeo) HG: Dous, tres. Xa! (Música) LS: Un, dous, tres. Xa! Arthur, vou intentalo outra vez. Un, dous, tres. Xa! HG: Arthur, déixame probar outra vez, si? Un, dous, tres. Xa! (Música) Mira. Acórdaste destes xoguetes? Ves estes xoguetes? Si, vou poñer este por aquí, e a ti vouche dar este. Veña, xa podes xogar. LS: Vale, Laura, pero claro, os bebés quérenlles ás súas mamás. Normal que lles dean os xoguetes a ela cando non conseguen que funcionen. De novo, a pregunta realmente importante é que ocorre cando cambiamos os datos estatísticos só levemente. Agora, os bebés van ver o xoguete funcionar e fallar xusto na mesma orde, pero imos cambiar a distribución da proba. Agora, Hyowon vai conseguilo unha vez e fracasar outra, e eu tamén. O que suxire que non importa quen proba este xoguete, está roto. Non funciona nunca. De novo, o bebé vai ter que tomar unha decisión. A súa nai está xusto ao lado, así que pode cambiar a persoa, e haberá outro xoguete ao final da tea. Vexamos que fai.
(Video) HG: Two, three, go! (Music) Let me try one more time. One, two, three, go! Hmm.
HG: Dous, tres, xa! (Música) Déixame probar outra vez. Un, dous, tres, xa! Umm.
LS: Let me try, Clara. One, two, three, go! Hmm, let me try again. One, two, three, go! (Music) HG: I'm going to put this one over here, and I'm going to give this one to you. You can go ahead and play. (Applause)
LS: Déixame probar a min, Clara. Un, dous, tres, xa! Umm, déixame probar outra vez. Un, dos, tres, xa! (Música) HG: Vou poñer este por aquí, e vouche dar este a ti. Veña, xa podes xogar. (Aplausos)
LS: Let me show you the experimental results. On the vertical axis, you'll see the distribution of children's choices in each condition, and you'll see that the distribution of the choices children make depends on the evidence they observe. So in the second year of life, babies can use a tiny bit of statistical data to decide between two fundamentally different strategies for acting in the world: asking for help and exploring. I've just shown you two laboratory experiments out of literally hundreds in the field that make similar points, because the really critical point is that children's ability to make rich inferences from sparse data underlies all the species-specific cultural learning that we do. Children learn about new tools from just a few examples. They learn new causal relationships from just a few examples. They even learn new words, in this case in American Sign Language.
LS: Amosarei agora os resultados experimentais. No eixe vertical, vese a distribución das eleccións dos nenos baixo cada suposto, e vese que a distribución das eleccións que fan depende da proba que observan. No segundo ano de idade, os bebés poden usar unha fracción mínima de datos estatísticos para decidir entre dúas estratexias fundamentalmente diferentes para actuar no mundo: pedir axuda e explorar. Acabo de amosar dous experimentos de laboratorio dos literalmente centos neste campo que chegan a conclusións similares, porque o auténtico punto clave é que a capacidade dos nenos para facer ricas inferencias partindo de datos dispersos serve de base a toda a nosa aprendizaxe cultural específica como especie. Os nenos aprenden sobre novas ferramentas a partir duns poucos exemplos. Aprenden novas relacións causais a partir duns poucos exemplos. Incluso aprenden palabras novas , neste caso en lingua de signos americana.
I want to close with just two points. If you've been following my world, the field of brain and cognitive sciences, for the past few years, three big ideas will have come to your attention. The first is that this is the era of the brain. And indeed, there have been staggering discoveries in neuroscience: localizing functionally specialized regions of cortex, turning mouse brains transparent, activating neurons with light. A second big idea is that this is the era of big data and machine learning, and machine learning promises to revolutionize our understanding of everything from social networks to epidemiology. And maybe, as it tackles problems of scene understanding and natural language processing, to tell us something about human cognition. And the final big idea you'll have heard is that maybe it's a good idea we're going to know so much about brains and have so much access to big data, because left to our own devices, humans are fallible, we take shortcuts, we err, we make mistakes, we're biased, and in innumerable ways, we get the world wrong. I think these are all important stories, and they have a lot to tell us about what it means to be human, but I want you to note that today I told you a very different story. It's a story about minds and not brains, and in particular, it's a story about the kinds of computations that uniquely human minds can perform, which involve rich, structured knowledge and the ability to learn from small amounts of data, the evidence of just a few examples. And fundamentally, it's a story about how starting as very small children and continuing out all the way to the greatest accomplishments of our culture, we get the world right.
Quero concluír con só dúas cousas. A quen seguise o meu campo (o do cerebro e as ciencias cognitivas) durante os últimos anos, chamaríanlle a atención tres grandes ideas. A primeira é que esta é a era do cerebro. E por suposto, houbo descubrimentos impresionantes en neurociencia: localizar rexións do córtex funcionalmente especializadas, facer transparentes os cerebros de ratos, activar neuronas con luz. Unha segunda grande idea é que esta é a era dos datos masivos e da aprendizaxe automática, e a aprendizaxe automática promete revolucionar a nosa comprensión de todo, dende as redes sociais ata a epidemioloxía. E tal vez, á vez que afronta problemas de comprensión do contexto e de procesamento da linguaxe natural, poida desvelarnos algo sobre a cognición humana. E a gran idea final que escoitarían é que pode ser boa idea saber tanto sobre os cerebros e ter tanto acceso a datos masivos, porque pola nosa conta, os humanos somos falíbeis, buscamos atallos, erramos, temos fallos, non somos neutrais, e de formas innumerables, chegamos a ideas falsas sobre o mundo. Eu creo que todas estas son noticias importantes, e que teñen moito que contarnos sobre qué significa ser humano, pero gustaríame destacar que hoxe tratei unha noticia moi distinta. Unha noticia sobre mentes, non sobre cerebros, e en particular, sobre o tipo de computación que só as mentes humanas poden realizar, que implican coñecementos ricos e estruturados e capacidade de aprender a partir de pequenas cantidades de datos, coa proba de só uns poucos exemplos. E fundamentalmente, é unha noticia sobre como dende meniños e continuando todo o camiño ata os máis grandes logros da nosa cultura, conseguimos entender ben o mundo.
Folks, human minds do not only learn from small amounts of data. Human minds think of altogether new ideas. Human minds generate research and discovery, and human minds generate art and literature and poetry and theater, and human minds take care of other humans: our old, our young, our sick. We even heal them. In the years to come, we're going to see technological innovations beyond anything I can even envision, but we are very unlikely to see anything even approximating the computational power of a human child in my lifetime or in yours. If we invest in these most powerful learners and their development, in babies and children and mothers and fathers and caregivers and teachers the ways we invest in our other most powerful and elegant forms of technology, engineering and design, we will not just be dreaming of a better future, we will be planning for one.
Amigos, as mentes humanas non aprenden só a partir de pequenas cantidades de datos As mentes humanas pensan ideas totalmente novas. As mentes humanas xeran investigación e descubrimento, e as mentes humanas xeran arte e literatura e poesía e teatro, e as mentes humanas coidan doutros seres humanos: os nosos maiores, a nosa mocidade, os nosos enfermos. Incluso os curamos. Nos próximos anos, imos ver innovacións tecnolóxicas máis alá do que podo concibir, pero hai moi poucas probabilidades de que vexamos algo que se aproxime sequera ao poder computacional dun neno humano, no resto da miña vida ou da vosa. Se investimos nestes potentísimos aprendices e no seu desenvolvemento, en bebés e cativos, e nais e pais e coidadores e profesores do xeito que investimos nas nosas outras poderosísimas e elegantes formas de tecnoloxía, enxeñaría e deseño, non estaremos simplemente soñando cun mellor futuro, estaremos planificándoo.
Thank you very much.
Moitísimas grazas.
(Applause)
(Aplausos)
Chris Anderson: Laura, thank you. I do actually have a question for you. First of all, the research is insane. I mean, who would design an experiment like that? (Laughter) I've seen that a couple of times, and I still don't honestly believe that that can truly be happening, but other people have done similar experiments; it checks out. The babies really are that genius.
Chris Anderson: Grazas, Laura. Quería facerche unha pregunta. Antes de nada, esta investigación é de tolos. Quen deseñaría un experimento coma ese? (Risas) Vino unhas cantas veces, e sigo sen acabar de crer que poida estar ocorrendo de verdade, pero outras persoas fixeron experimentos similares; está comprobado. Os bebés son realmente xenios.
LS: You know, they look really impressive in our experiments, but think about what they look like in real life, right? It starts out as a baby. Eighteen months later, it's talking to you, and babies' first words aren't just things like balls and ducks, they're things like "all gone," which refer to disappearance, or "uh-oh," which refer to unintentional actions. It has to be that powerful. It has to be much more powerful than anything I showed you. They're figuring out the entire world. A four-year-old can talk to you about almost anything. (Applause)
LS: Parecen realmente impresionantes nos nosos experimentos, pero pensa no que fan na vida real, non? Todo comeza cun bebé. Dezaoito meses despois, estache falando, e as primeiras palabras dos bebés non van de pelotas e parrulos, son cousas como “non ta” que se refire á desaparición, ou “uh oh”, para referirse a accións involuntarias. Ten que ser así de poderoso. Ten que ser moito máis poderoso que o que ensinei. Están descifrando o mundo enteiro. Un neno de catro anos pode falarche sobre case todo. (Aplausos)
CA: And if I understand you right, the other key point you're making is, we've been through these years where there's all this talk of how quirky and buggy our minds are, that behavioral economics and the whole theories behind that that we're not rational agents. You're really saying that the bigger story is how extraordinary, and there really is genius there that is underappreciated.
CA: E se entendo ben, o outro punto clave que destacas é que durante estes anos tivemos todo este debate sobre o peculiares e confusas que son as nosas mentes, coa economía condutual e teorías enteiras detrás de que non somos axentes racionais. E ti estás a dicir que este fenómeno é extraordinario, e que en realidade hai xenialidade que está subestimada.
LS: One of my favorite quotes in psychology comes from the social psychologist Solomon Asch, and he said the fundamental task of psychology is to remove the veil of self-evidence from things. There are orders of magnitude more decisions you make every day that get the world right. You know about objects and their properties. You know them when they're occluded. You know them in the dark. You can walk through rooms. You can figure out what other people are thinking. You can talk to them. You can navigate space. You know about numbers. You know causal relationships. You know about moral reasoning. You do this effortlessly, so we don't see it, but that is how we get the world right, and it's a remarkable and very difficult-to-understand accomplishment.
Unha das miñas citas favoritas en psicoloxía é do psicólogo social Solomon Asch, que dixo que “o cometido fundamental da psicoloxía é eliminar o veo de autoevidencia das cousas”. Hai millóns de decisións que se toman a diario que interpretan ben o mundo. Coñecemos os obxectos e as súas propiedades. Recoñecémolos cando están ocultos. Recoñecémolos na escuridade. Camiñamos por cuartos. Podemos percibir o que pensan outros. Podemos falarlles. Podemos navegar no espazo. Coñecemos os números. Entendemos as relacións causais. Entendemos o razoamento moral. E todo isto sen esforzo ningún, por iso non nos decatamos, pero así interpretamos ben o mundo, e moi difícil de entender.
CA: I suspect there are people in the audience who have this view of accelerating technological power who might dispute your statement that never in our lifetimes will a computer do what a three-year-old child can do, but what's clear is that in any scenario, our machines have so much to learn from our toddlers. LS: I think so. You'll have some machine learning folks up here. I mean, you should never bet against babies or chimpanzees or technology as a matter of practice, but it's not just a difference in quantity, it's a difference in kind. We have incredibly powerful computers, and they do do amazingly sophisticated things, often with very big amounts of data. Human minds do, I think, something quite different, and I think it's the structured, hierarchical nature of human knowledge that remains a real challenge.
CA: Imaxino que hai persoas no público que comparten esa visión do crecente poder tecnolóxico que poderían cuestionar a túa afirmación de que nunca nas nosas vidas un ordenador fará o que un neno de tres anos pode facer, pero está claro que en calquera situación, as nosas máquinas teñen moito que aprender dos nosos cativos. LS: Eu tamén o creo. Aquí haberá partidarios da aprendizaxe automática. Nunca deberías apostar contra os bebés ou os chimpancés ou da tecnoloxía, en principio. pero non se trata só dunha diferenza de cantidade, é unha diferenza cualitativa. Temos ordenadores incriblemente potentes, que fan cousas incriblemente sofisticadas, por veces con enormes cantidades de datos. As mentes humanas fan, para min, algo bastante diferente, e creo que é a natureza estruturada e xerarquizada do coñecemento humano o que permanece como un verdadeiro desafío.
CA: Laura Schulz, wonderful food for thought. Thank you so much.
CA: Laura Schulz, un gran tema para reflexionar. Moitas grazas.
LS: Thank you. (Applause)
Grazas (Aplausos)