{"id":473232,"date":"2018-03-21T07:43:55","date_gmt":"2018-03-21T14:43:55","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?p=473232"},"modified":"2018-05-23T14:54:34","modified_gmt":"2018-05-23T21:54:34","slug":"building-literate-machines-with-dr-adam-trischler","status":"publish","type":"post","link":"https:\/\/www.microsoft.com\/en-us\/research\/podcast\/building-literate-machines-with-dr-adam-trischler\/","title":{"rendered":"Building Literate Machines with Dr. Adam Trischler"},"content":{"rendered":"<div id=\"attachment_473238\" style=\"width: 1010px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-473238\" class=\"wp-image-473238 size-full\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/03\/MicrosoftResearch_Podcast_Adam-Treschler_MCR-Blog-Header_1000x400.jpg\" alt=\"\" width=\"1000\" height=\"400\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/03\/MicrosoftResearch_Podcast_Adam-Treschler_MCR-Blog-Header_1000x400.jpg 1000w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/03\/MicrosoftResearch_Podcast_Adam-Treschler_MCR-Blog-Header_1000x400-300x120.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/03\/MicrosoftResearch_Podcast_Adam-Treschler_MCR-Blog-Header_1000x400-768x307.jpg 768w\" sizes=\"(max-width: 1000px) 100vw, 1000px\" \/><p id=\"caption-attachment-473238\" class=\"wp-caption-text\">Dr. Adam Trischler &#8211; Microsoft Research Manager<\/p><\/div>\n<p><strong>Episode 16, March 21, 2018<\/strong><\/p>\n<p>Learning to read, think and communicate effectively is part of the curriculum for every young student. But <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/adtrisch\/\">Dr. Adam Trischler<\/a>, Research Manager and leader of the Machine Comprehension team at <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/lab\/microsoft-research-montreal\/\">Microsoft Research Montreal<\/a>, would like to make it part of the curriculum for your computer as well. And he\u2019s working on that, using methods from machine learning, deep neural networks, and other branches of AI to close the communication gap between humans and computers.<\/p>\n<p>Today, Dr. Trischler talks about his dream of making literate machines, his efforts to design meta-learning algorithms that can actually learn to learn, the importance of what he calls \u201cfew-shot learning\u201d in that meta-learning process, and how, through a process of one-to-many mapping in machine learning, our computers not may not only be answering our questions, but asking them as well.<\/p>\n<p><b>Relate<\/b><b>d:<\/b><\/p>\n<ul type=\"disc\">\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/podcast\">Microsoft Research Podcast<\/a>: Visit our podcast page on Microsoft.com<\/li>\n<li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/itunes.apple.com\/us\/podcast\/microsoft-research-a-podcast\/id1318021537?mt=2\">iTunes<\/a>: Subscribe and listen to new podcasts each week on iTunes<\/li>\n<li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/subscribebyemail.com\/www.blubrry.com\/feeds\/microsoftresearch.xml\">Email<\/a>: Subscribe and listen by email<\/li>\n<li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/subscribeonandroid.com\/www.blubrry.com\/feeds\/microsoftresearch.xml\">Android<\/a>: Subscribe and listen on Android<\/li>\n<li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/open.spotify.com\/show\/4ndjUXyL0hH1FXHgwIiTWU\">Spotify<\/a>: Listen on Spotify<\/li>\n<li><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.blubrry.com\/feeds\/microsoftresearch.xml\">RSS feed<\/a><\/li>\n<\/ul>\n<hr \/>\n<p><strong>Transcript<\/strong><\/p>\n<p>Adam Trischler: The problem right now is algorithms require so much data to learn even simple things like recognizing cats and dogs. And that brings us back to the meta-learning aspect, is we really want to build systems that learn on-the-fly, and continually, rather than just once and doing their task forever and ever. And we want those systems to be able to pick things up rapidly, really data-efficiently. So, from just a few examples, I can learn a new task.<\/p>\n<p><strong>Host: You\u2019re listening to the Microsoft Research Podcast. A show that brings you closer to the cutting edge of technology research and the scientists behind it. I\u2019m your host, Gretchen Huizinga.<\/strong><\/p>\n<p><strong>Learning to read, think, and communicate effectively is part of the curriculum for every young student. But Dr. Adam Trischler, Research Manager and leader of the Machine Comprehension team at Microsoft Research Montreal would like to make it part of the curriculum for your computer as well. And he\u2019s working on that using methods from machine learning, deep neural networks, and other branches of AI, to close the communication gap between humans and computers.<\/strong><\/p>\n<p><strong>Today, Dr. Trischler talks about his dream of making literate machines, his efforts to design meta-learning algorithms that can actually learn to learn, the importance of what he calls \u201cfew-shot learning\u201d in that meta-learning process, and how, through the method of one-to-many mapping in machine learning, our computers may not only be answering our questions, but asking them as well.<\/strong><\/p>\n<p><strong>That and much more on this episode of the Microsoft Research Podcast.<\/strong><\/p>\n<p><strong>(music plays)<\/strong><\/p>\n<p><strong>Host: Welcome, Adam Trischler, to the podcast this morning. Great to talk to you.<\/strong><\/p>\n<p>Adam Trischler: Thank you, Gretchen. I\u2019m glad to be here.<\/p>\n<p><strong>Host: All the way from Montreal, Quebec, from the&#8230;<\/strong><\/p>\n<p>Adam Trischler: Yes.<\/p>\n<p><strong>Host: &#8230;Microsoft Research Montreal lab. Tell me what it\u2019s like working at the research lab in Montreal. It\u2019s become a global hotbed for AI research.<\/strong><\/p>\n<p>Adam Trischler: Yeah, it\u2019s been incredible. It\u2019s like, when Maluuba first moved here in December 2015, it was small. I mean, Montreal had this big lab through Yoshua at MILA, but in terms of the sort of corporate interest and presence, it was nothing. And our lab was one of the first, and we started with about 4 or 5 people. And since then, it\u2019s been like watching a skyscraper go up around you. It\u2019s really, really cool.<\/p>\n<p><strong>Host: So, you lead the Machine Comprehension team at Microsoft Research Montreal. In broad strokes, because I\u2019m going to go specific, tell me what gets you up in the morning. What do you do, what do you study, why is it important?<\/strong><\/p>\n<p>Adam Trischler: Okay. So, broadly speaking, the goal of my team is to build literate machines. So, what that means, a bit more specifically, is machines that learn from and understand the world through language like people do. So, I\u2019m really inspired by the prospect of using machines to unlock all of the human knowledge that\u2019s recorded in text. We have textbooks, we have Wikipedia, so many instructions of how to do things, skills that we can gain, recipes we can make. Whether, like, literal recipes for cooking or a recipe for how to do something or build something.<\/p>\n<p><strong>Host: Like an algorithm?<\/strong><\/p>\n<p>Adam Trischler: Exactly, like an algorithm! So, yeah, we could get the machines to be sort of self-aware and adjust their own algorithms, in some sense. But anyway, what really drew me to AI in general, and this field in particular, is first of all, the prospect of using AI as a window onto human intelligence. So, I\u2019ve always been fascinated by thought. As a kid, I definitely tended to spend a lot of time in my own mind, just, you know, thinking about thinking, talking to myself even. And I\u2019ve always loved language as well. So, I\u2019m just generally fascinated by how language shapes and facilitates thoughts even, like grandly, philosophically speaking, yes, of course. But even at the smaller, like more personal scale as well. For example, how writing an idea down or explaining it to another person can clarify and crystalize that idea for you, yourself, and help you to understand it better. So, like I said, our goal is to build literate machines that learn from and understand the world through language which is so useful for us. I don\u2019t want to make machines that, sort of, make people read less, but I want machines that can augment human reading, for example, by taking care of some of the more mundane parts, you know, slogging through your insurance policy or an HR manual or maybe filtering the massive stream of text, this massive stream that we have coming at us through Twitter and everywhere else.<\/p>\n<p><strong>Host: Or maybe even helping me to understand all the legalese that I hit \u201cagree\u201d to when I\u2019m getting an app.<\/strong><\/p>\n<p>Adam Trischler: All those terms of service, exactly.<\/p>\n<p><strong>Host: Right, I agree.<\/strong><\/p>\n<p>Adam Trischler: The things that everyone is supposed to read, but no one does. You know, I think one of the things you hinted at there is what I\u2019m most excited about is, like, seeing a literate machine as a kind of librarian, or a tutor, who could guide, like, a human student or just somebody who has an interest in something through new books, new materials, new ideas, like stoking their natural curiosity and feeding them new information as the student would ask questions.<\/p>\n<p><strong>Host: You know, my mind is racing already. I have a list of questions that I want to ask. And some of them are jumping up to the front of the line, raising their hands going, \u201cAsk me! Ask me!\u201d<\/strong><\/p>\n<p>Adam Trischler: All right. Go ahead.<\/p>\n<p><strong>Host: I know, right? Let\u2019s go off on that tangent a little bit about HOW your research is teaching these machines to be literate; read, think and communicate like humans. Because I\u2019m trying to wrap my brain around what it looks like if the machine is doing some of the heavy lifting we could call it. For me, how does that transfer to my brain so that I could say, understand something more quickly. My daughter, in college, how could she use a machine to help her be more successful in school?<\/strong><\/p>\n<p>Adam Trischler: I guess where we really started in this, sort of, you know, quest for literate machines was a lot more concrete, pretty straightforward, in the field of question answering. So, the idea here is, we simply want to build machines that, if given a document, let\u2019s say an essay or a news article, you could ask the machine a question about that document, and it could provide you with a reasonable, hopefully ideally correct, answer to your question. So, why we were interested in this is because I think it\u2019s a nice proxy for test and comprehension. So, understanding, comprehending language \u2013 obviously, this is a sort of ephemeral concept. We don\u2019t have a good way of measuring something like comprehension or understanding directly. But we can use proxies like question answering. So, in building a literate machine, for example, one of the tests we can imagine is a comprehension test, like a human student would receive at school. You\u2019re given these test questions. What happened here? Why? What were the motivations? And what followed?<\/p>\n<p><strong>Host: Let\u2019s talk about what it actually looks like. What you\u2019ve just described is so fascinating, and I\u2019ve talked to several researchers here at Microsoft Research who all use this idea of the delicate balance, the dance between human and machine, augment versus replace. What does that look like in my life? How could that play out? Right now, I have a tablet where if I hold my finger on a word I don&#8217;t know, I can look it up, right? Is there some more advanced version of that? I mean, what do you envision here?<\/strong><\/p>\n<p>Adam Trischler: Yeah, I mean, question answering definitely has a passive nature to it, right? The machine is just kind of sitting there waiting for you, as the user, to highlight the word you don\u2019t understand. Another thing we\u2019ve worked on fairly recently, and which is perhaps even more exciting, is the idea of question asking. So, you know, just the other side of the coin. A machine that, rather than just waiting for you to pose a question and answering it for you, can do the sort of curiosity-driven question asking, just sort of guide you along through knowledge or act as a tutor for you. So, we\u2019re just getting started on this. Obviously, it\u2019s a complex task. In some ways, asking questions is more complex than answering them, because you can imagine if I give you a document and a question, if it\u2019s well-posed, it probably leads to a single answer. Whereas, if I gave you a document, even if I provided you with a set of terms or snippets from that document that, I said, \u201cAsk questions about these,\u201d even if you\u2019re just looking at the information in the document, you can probably formulate several questions that lead to the same answer. So, it\u2019s this one-to-many mapping, rather than a one-to-one mapping, that we see more typically in the question answering case. So, it\u2019s really difficult. As I said, we\u2019re just getting started. But already, we\u2019ve seen some adoption of this. I think it could be super useful in things like MOOCs, massively online open courses. But ultimately, you can see this as really, kind of, driving people, hopefully, to learn more and to improve their understanding.<\/p>\n<p><strong>Host: When you\u2019re talking about all the interaction there, I start thinking about user interface. We\u2019ll talk in a bit about the technology behind everything, but ultimately, it\u2019s going to be \u201cAI has an interface.\u201d And I imagine you\u2019re already thinking about what user interface these kinds of machines can have that anticipate and generate questions for me, or answer them, or communicate with me?<\/strong><\/p>\n<p>Adam Trischler: We\u2019ve certainly thought about it. I think we\u2019re sufficiently far away from making this technology work nicely outside of, you know, kind of, trivial, literally trivial settings, like trivia on search. But we have thought about it. You know, you could imagine, if you\u2019re in this sort of interaction on your phone, the camera could be watching your face for those sort of visual cues. We can pick up on vocal cues as well. I mean, the bigger picture is SO big that, you know, it\u2019s not something that one team is going to tackle. It\u2019s these different teams coming together, breaking the problem down into its parts, and then, hopefully, bringing them all together into, you know, a really compelling product or assistant or use-case in the end. But there\u2019s just so much for us in the language itself that we\u2019re not even close to that yet. And so thankfully, we do have, you know, in MSR, these other amazing teams who are working on these other really challenging aspects of these problems.<\/p>\n<p><strong>(music plays)<\/strong><\/p>\n<p><strong>Host: Let\u2019s focus on the language for a bit since that\u2019s your work. When I think of talking to machines, being able to communicate with me, I think pretty well ethno-centrically.<\/strong><\/p>\n<p>Adam Trischler: Of course. We all tend to, right?<\/p>\n<p><strong>Host: Yeah. And I know there\u2019s a lot of work going on in machine translation as well. Are we heading for an AI future where language makes no difference to any of us, people or machines?<\/strong><\/p>\n<p>Adam Trischler: I think we\u2019re definitely working in that direction. One of the really interesting things about this new wave of AI through deep learning, is that we get a lot of stuff, like bilingualism or trilingualism et cetera, almost for free. It\u2019s not totally free. But on the algorithmic side, we can do things that are very general. We can build systems that are agnostic to the particular language they\u2019re operating on. And so, you know, there\u2019s generalities to language. Obviously, there are different specifics, and you can\u2019t classify all languages the same. There are structural differences, morphological differences. But from the algorithmic side, there\u2019s a lot of generality. And so, we can build something on our end that can really operate in a whole variety of languages. And all that matters for us is that we have the training data to tailor it to each of those individual languages. So, I can build the same recurrent neural network that processes English or French, or both. Whether it will do those things, and do them well, is really just a factor of the data that I use to train my system.<\/p>\n<p><strong>Host: Well, and we\u2019ve talked about data on this podcast before, and how important it is to have, not just lots, but quality, of the right kind of data to learn and train machines. So, let\u2019s talk about machine-learning for a second. Um, there\u2019s several lines of research in that deep learning, supervised, unsupervised reinforcement learning. And until fairly recently, the models have been pretty task-specific, but you\u2019re doing work&#8230;<\/strong><\/p>\n<p>Adam Trischler: Absolutely.<\/p>\n<p><strong>Host: You\u2019re doing work on what we call meta-learning algorithms. Can you tell us about that, and particularly your work on, like, rapid adaptation and conditionally-shifted neurons?<\/strong><\/p>\n<p>Adam Trischler: Yeah, meta-learning is something that we\u2019re really excited about here in the group. And in general, the field is really picking up this new sort of paradigm, I guess. So meta-learning really refers to learning to learn. The goal of a meta-learning algorithm is the ability to learn new tasks efficiently, given little training data for each individual task. So, you know, these systems that we\u2019re training right now, these task-specific systems, require so much data to perform really, really well. And they do, and that\u2019s great, but we don\u2019t always have a ton of data. So, the problem we\u2019re trying to address with meta-learning is that, like, right now, neural networks, they need to see, generally-speaking, hundreds or thousands of examples of a class to be able to recognize it. On the other hand, you have people. If you showed me one or two pictures of some hypothetical, brand-new car model, I\u2019d probably be able to recognize it on the road, in different colors, in different lighting and weather conditions, right away, from one or two pictures. So, this is something we call \u201cfew-shot learning.\u201d It just takes a few-shots of this thing I want to learn to be able to recognize it. And yeah, humans are really, really good at it. So, the standard way we\u2019ve trained machine-learning models, in particular, deep neural networks up until now, it really doesn\u2019t encourage this ability for few-shot learning. ML systems, you know, they\u2019re typically trained through one optimization phase after which that\u2019s it, learning is over. So, we build these systems in this \u201ctrain and then test\u201d manner. And they don\u2019t really scale to complex environments, and they don\u2019t have the capability to pick up topics on-the-fly. So, one of the things we do in meta-learning, first of all, is just change the training setup. Rather than showing a model how to do one big task, with lots of data, we\u2019ll show it a set of smaller, related tasks that sort of have a few things in common, but they\u2019re not exactly the same. So, to give you an example, let\u2019s say, instead of learning to recognize, like, fifty breeds of dogs all at once, we\u2019ll give a model the smaller task of recognizing, let\u2019s say just Chihuahuas versus huskies, and then a different task, which is just poodles versus bulldogs, and so on. So, when you have this kind of setup, there are these general features of all dogs that will remain constant across all these smaller tasks, and the model can learn these gradually, picking them up across tasks as it seems them in sequence and over many examples. But there are also these specific features of the specific breeds that the model has to pick up rapidly from just a few examples while it\u2019s doing each individual task. And so, there are these, sort of, two levels of learning, the slower like, \u201cWhat do dogs look like in general?\u201d and then there\u2019s the faster, \u201cWhat do these specific dogs look like, and what are the features that discriminate them from each other?\u201d<\/p>\n<p><strong>Host: So, is that the rapid adaptation that you\u2019re talking about?<\/strong><\/p>\n<p>Adam Trischler: Exactly. Yeah, so that second level that I mentioned is the rapid adaptation.<\/p>\n<p><strong>Host: Right. So, as a side note, if you\u2019re training a machine on recognizing dogs, and they all have four legs, and then suddenly you\u2019ve got a dog that lost a leg and it sees three-legged dog, can it still say that it\u2019s a dog?<\/strong><\/p>\n<p>Adam Trischler: Yeah, so that would definitely be a problem. That\u2019s another place where human-learning&#8230;<\/p>\n<p><strong>Host: Humans are good.<\/strong><\/p>\n<p>Adam Trischler: &#8230;they\u2019re like\u2026 just vastly diverges from machine-learning right now. An algorithm that learns to recognize dogs\u2026 ahhh, I mean, it\u2019s tough to say, because we can\u2019t peek inside them and we can\u2019t ask them questions. But I feel pretty confident&#8230;<\/p>\n<p><strong>Host: Not yet.<\/strong><\/p>\n<p>Adam Trischler: Yeah, not yet. We\u2019re getting there. But I feel pretty confident in saying that the algorithm that\u2019s learning to recognize dogs and seeing all these dogs with four legs, it doesn\u2019t really understand what legs are. It doesn\u2019t know what legs are for. It doesn\u2019t know that legs can be lost without fundamentally altering, you know, the \u201cdogness\u201d of this thing, right? It sees an object composed of all these parts, and so it\u2019s going to infer that all of these parts are necessary to the definition of that thing. As humans, we know, because of the way we decompose things, that that\u2019s not always the case.<\/p>\n<p><strong>Host: Yeah, so that\u2019ll be a big challenge, I think, um, for the next round of research to say three-legged dogs are still dogs.<\/strong><\/p>\n<p>Adam Trischler: Yeah, so that relates to an idea that Yoshua Bengio has been interested in for a long time. Of course, Yoshua is the luminary here for us in Montreal. But he\u2019s interested in this idea of the factors of variation and the controllable factors of variation that really define classes of things, for example, and what we can do in the world.<\/p>\n<p><strong>Host: So, let\u2019s go on that topic a little more, a little deeper, Adam. Because \u201cunderstanding\u201d is a tough concept, even when we\u2019re talking about, do humans understand other humans? Can machines ever really understand? I\u2019m going to go out on a limb and say no.<\/strong><\/p>\n<p>Adam Trischler: Sure.<\/p>\n<p><strong>Host: But that\u2019s in the context of traditional historical, even metaphysical, our understanding of understanding. Since this is what you\u2019re working on, how would you argue that they eventually can? Or would you?<\/strong><\/p>\n<p>Adam Trischler: I absolutely think that machines are capable of understanding in the way that we use the word to refer to humans. I do believe that. So, for me, I think that one of the fundamental aspects of understanding is just the way we relate, and map, different things to each other. So, one example that sort of indicates the importance of relations and mappings for understanding, is there\u2019s this thing you can do where you repeat a word to yourself over and over again, and it just loses all meaning. And you\u2019re like, what the hell is this word? Like, why does this refer to the thing that it refers to? I\u2019m sure you\u2019ve done that before.<\/p>\n<p><strong>Host: I\u2019ve done it.<\/strong><\/p>\n<p>Adam Trischler: Of course.<\/p>\n<p><strong>Host: I still do it.<\/strong><\/p>\n<p>Adam Trischler: Yeah, it\u2019s fun, and it\u2019s weird.<\/p>\n<p><strong>Host: Well, we start\u2026 sometimes when we start saying our names over and over, it\u2019s like, what??<\/strong><\/p>\n<p>Adam Trischler: Exactly. The name. Yeah, that\u2019s the first thing I can remember doing that with, is my own name. And I think that we repeat this thing over and over again to ourselves, and in that process, it becomes disconnected. It loses its meaning, because you\u2019ve cut it off from the things it maps to. So, one of the things that we talk about as very important in language research for AI is the concept of grounding. So, language is grounded in the real world. It reflects the real world. The simplest, sort of, example of this is nouns. Like, nouns are essentially labels for things we see in the world. The point is that language is grounded, and it reflects the world, and it\u2019s fundamentally connected to the world, and there\u2019s only so much understanding you can glean from language without that connection. And that\u2019s one of the things that we\u2019re missing right now in AI and machine-learning, is that typically we teach machines language just within the world of language. So, they learn how to use words by looking at words, by reading documents. But they don\u2019t learn how to use words by interacting with people in a conversation or asking for things to be given to them, or seeing that, you know, a chair is this thing that I can sit on and push around. They\u2019re missing that mapping and that grounding to the real world, and that\u2019s where I think understanding stems from. And because I do think it is possible for machines to get that other aspect of things, to ground language, I do think they can understand.<\/p>\n<p><strong>Host: So that leads to a question, obvious in my mind, is how? How are you going to do that?<\/strong><\/p>\n<p>Adam Trischler: How?<\/p>\n<p><strong>Host: Yeah, I mean, like you guys have said, \u201cWe want to close the communication gap between humans and machines.\u201d So yeah, how? Because what you described to me is the reverse engineering. It\u2019s the human disengaging from that grounding and becoming the machine, saying, \u201cAdam, Adam, Adam, that doesn\u2019t make any sense.\u201d<\/strong><\/p>\n<p>Adam Trischler: Right.<\/p>\n<p><strong>Host: So, let\u2019s engineer the \u201cbackwards\u201d you know? So, from Adam meaning nothing to the machine, to Adam meaning this cool guy at Montreal, you know, Microsoft Research.<\/strong><\/p>\n<p>Adam Trischler: Please, go on. No, this &#8211; actually, this is a \u2013 this podcast is very challenging. This is like a philosophical test.<\/p>\n<p><strong>Host: Well, yeah, and probably if you tell me, you\u2019ll have to shoot me, because it\u2019s like proprietary information on how we\u2019re doing that.<\/strong><\/p>\n<p>Adam Trischler: Well, you\u2019re really asking the tough questions here.<\/p>\n<p><strong>Host: No, I know.<\/strong><\/p>\n<p>Adam Trischler: No, this is good. Um, so one of the things we have to do is move away from this paradigm we have right now of training, let\u2019s say, agents that use language, these literate machines I\u2019m trying to build\u2026 we have to move away from training them just on language. As I said, they have to learn how to map those words to things that are not words. So, that could be images. That could be actions. So, learning that the words \u201csitting down\u201d refer to an action that you can observe, let\u2019s say, in the real world or from videos, that is making the connection between the words and something else.<\/p>\n<p><strong>Host: Maybe this goes back to what we talked about earlier on the fact that multiple teams that are working on multiple hard problems, and you\u2019ve got people doing computer vision and emotive computing. And you\u2019re not working on that, but they are. And then suddenly, your algorithms get together and have this love child algorithm that, you know, I mean, I hear what you\u2019re saying, and it\u2019s\u2026 but what it begs also is that the computer is going to have to be watching a lot of stuff, reading a lot of stuff, experiencing, to use another human term, a lot of stuff.<\/strong><\/p>\n<p>Adam Trischler: For sure. Yeah, I don\u2019t think you can get around that. The way we know, right now, to create an intelligent system is to have a baby, and then raise it through eight, ten, twenty years. It\u2019s a long process to get, you know, a fully-formed, functional adult. Obviously, even a child is way, way smarter than the systems we have right now. But the point is, intelligence isn\u2019t easy, right? Even in humans, you know, we don\u2019t just come out being able to speak. We pick it up really quickly. And we\u2019re tailored to do things like few-shot learning really effectively. But still, it doesn\u2019t just happen.<\/p>\n<p><strong>Host: That\u2019s why research. I mean&#8230;<\/strong><\/p>\n<p>Adam Trischler: Exactly, Exactly.<\/p>\n<p><strong>Host: That\u2019s why your lab is doing what you\u2019re doing.<\/strong><\/p>\n<p>Adam Trischler: Exactly, and the systems \u2013 to pick up these mappings that I\u2019m talking about, you know, they need to experience the world or at least experience proxies or recordings of the world through books, a lot of text, images, videos, all that sort of stuff which we call multimodal learning. And, you know, the problem right now is we can\u2019t possibly have an algorithm do that, because algorithms require so much data to learn even simple things like recognizing cats and dogs. And that brings us back to the meta-learning aspect, is we really want to build systems that learn on-the-fly, and continually, rather than just once and doing their task forever and ever. And we want those systems to be able to pick things up rapidly, really data-efficiently. So, from just a few examples, I can learn a new task.<\/p>\n<p><strong>Host: Yeah, I think you called that sample efficiency?<\/strong><\/p>\n<p>Adam Trischler: Exactly, yeah. Sample efficiency, data efficiency\u2026<\/p>\n<p><strong>Host: And then being able to transfer what it learns to other scenarios that aren\u2019t exactly the same.<\/strong><\/p>\n<p>Adam Trischler: Exactly. So right now, in meta-learning, the way we set things up is that the different tasks that the model is undertaking, they are fairly similar to each other, but ultimately, we\u2019d like to start breaking that and saying, okay, now you\u2019ve learned cats and dogs, but let\u2019s take that to something very different like elephants and horses.<\/p>\n<p><strong>(music plays)<\/strong><\/p>\n<p><strong>Host: I interviewed one of your colleagues over there, Harm van Seijen, about how they used reinforcement learning to beat Ms. Pacman at its own game.<\/strong><\/p>\n<p>Adam Trischler: Yup.<\/p>\n<p><strong>Host: And, um, he used the phrase, \u201cislands of tractability,\u201d which is where you focus your efforts because you know you can, you know, have some semblance of success there. So, what are the biggest challenges right now that might be offshore from the islands of tractability that you see are most exciting or promising areas of research, especially for people that might be interested in getting into this?<\/strong><\/p>\n<p>Adam Trischler: One of the big ones for me, because of my focus on language \u2013 and I think a lot of people here at the Montreal lab will echo this \u2013 is evaluation of language. You know, in machine-learning and many, many other fields, you only get what you measure. And it is so hard to measure the quality of language. Language is slippery, and it\u2019s really hard to measure. So, one of the things we focused on in this group here is natural language generation. So obviously, this factors into the earlier work I was talking about on question generation. Like, we have to build a question, in natural language, that flows, that makes sense to people, and even more importantly, that asks about real information and is well-posed, and leads you to the answer that you\u2019re looking for. And it\u2019s so incredibly hard to measure and evaluate language. This comes up in machine translation as well. And part of the reason for this is that there\u2019s so many different ways to say the same thing. And so, even training a language-generation system on these massive corpora of language data that we have now, they\u2019re still missing out on very many plausible and reasonable ways to say things. They\u2019ll never see those hidden-away examples.<\/p>\n<p><strong>Host: I could help you, but I don\u2019t write algorithms, but\u2026 You know what I mean?<\/strong><\/p>\n<p>Adam Trischler: Well, see, that\u2019s one of the really interesting things, is that \u2013 so in the lab here in MSR Montreal right now, we are all, with a few exceptions, we\u2019re computer scientists. And we\u2019re the ones tackling this language problem and trying to measure the quality of language outputs. But we\u2019re not necessarily the best-suited to that job. I really think that this is a problem that could really, really benefit from an interdisciplinary effort. There\u2019s so much that goes into language which is beyond algorithm and computation, I think, that we really need to take into account.<\/p>\n<p><strong>Host: And I could help you\u2026<\/strong><\/p>\n<p>Adam Trischler: I would love that. We need help. Like, honestly, evaluating language is so, so hard. Like, let me tell you about this metric we have called Blue, which is used in machine translation. So, the way you measure the quality of outputs in machine translation is you have one or more, sort of, \u201cgold\u201d reference translations. It\u2019s one sentence that says maybe, \u201cmachine-learning is hard.\u201d So, your system is going to say all these different things like \u201cmachine-learning is a tough problem,\u201d \u201cartificial intelligence is not very easy,\u201d \u2013 you know, you can imagine all these different ways. But the way that Blue measures these outputs is it just looks at how many words overlap or how many pairs of words overlap between the two candidate translations. And so, you can obviously imagine, there\u2019s a translation which has zero or very, very low overlap with my reference. But it could still be completely valid. And so, in this case, my algorithm is being told, \u201cNo, this thing that you tried to say is complete garbage because it has zero overlap,\u201d and it\u2019s being punished very, very unfairly for saying something totally reasonable. But just because we have this very limited ability to measure what really is reasonable, the whole thing\u2019s breaking down.<\/p>\n<p><strong>Host: Hmmm. I think AI\u2019s changing our world in ways that \u2013 well, of course ways that we don\u2019t understand \u2013 but one of them is this, you know, having my daughter in college right now, when anyone says, what are you studying? If she says anything besides a STEM subject, people look at her and say, \u201cOh, what are you going to ask with the degree? Do you want fries with that?\u201d However, I keep hearing from computer scientists like yourself, especially researchers, that other things are necessary, you know? Computation is \u201cnecessary, but not sufficient\u201d for AI.<\/strong><\/p>\n<p>Adam Trischler: Yeah, I mean, the thing is, AI \u2013 intelligence \u2013 this word refers to human behavior. And so, if you want to build a system that exhibits intelligence, and intelligence is this human thing, it should be intelligent in ways similar to the ways we are. And so, we need an understanding of human behavior, and that\u2019s something that we hoodie-clad STEM guys don\u2019t necessarily have.<\/p>\n<p><strong>Host: I hear you. Hey, so let\u2019s talk about, um, you know, I asked you at the beginning of the program what gets you up in the morning. And I sort of want to find out at the end of the program what keeps you up at night. And I read an article that you wrote in Fast Company called, Who Will Protect AI from Humanity? Why do machines need to be protected from us?<\/strong><\/p>\n<p>Adam Trischler: Right now, I don\u2019t think that they do. This really goes back to what I said before, when you asked me about, do I think that machines can one day understand in a human sense or have true comprehension? You know, I said the answer is yes. And, if I\u2019m right, if the answer is yes, then one day \u2013 not necessarily tomorrow or even five years from now \u2013 but one day, we\u2019re going to have a system that understands, that has memory of experiences that it has had. And if such a system exists, obviously we\u2019ll have played a significant hand in its creation, but I don\u2019t think we can consider ourselves to be the \u201cowners\u201d of such as system. To some degree, it will be an individual, because it has its own memories and its own understanding. And so that\u2019s where I think that we have to start realizing that, you know, just because you built it and you trained it and you even wrote the code for it, it\u2019s not necessarily yours or property of some corporation that fed it all of its data. It\u2019s not a huge concern. We\u2019re not even close to that. But it\u2019s good to think about these things probably far in advance.<\/p>\n<p><strong>Host: Exactly, because, sort of, the follow up question on that is, yeah, I believe that if we don\u2019t ask the questions now, the unintended consequences will hit. And so, having read the article, I should say, you bring up the issues of rights and ethics. And we\u2019ve all seen the Boston Dynamics robots both get pushed over and you kind of feel sorry for the robot.<\/strong><\/p>\n<p>Adam Trischler: For sure.<\/p>\n<p><strong>Host: And now you see it open a door and fighting off a stick, and you\u2019re going, \u201cI don\u2019t feel sorry for you anymore. I\u2019m scared of you.\u201d But these are issues that we, as humans, we understand rights and ethics and compassion and things like that. And so, I guess my better question would be, what questions do we need to be asking, and what issues do we need to be addressing while we are still upstream from pretty fundamental changes in our relationship with technology?<\/strong><\/p>\n<p>Adam Trischler: One of the things we mentioned before is this idea of measurement. I think I talk about this in the article to some degree is, you know, we have to be able to\u2026 (laughs)\u2026 you know, how can you measure individuality? It doesn\u2019t really make sense. How can you measure the memory capacity of something? Obviously, we can measure in megabytes and gigabytes, but I mean more in an experiential sense. We don\u2019t know how to do that. But if we\u2019re going to consider artificial intelligence as, you know, something on the level of people, then we have to start thinking about yes, first, how do we measure consciousness or sentience, or memory, or understanding? Because only then can you start to say, you know, this thing is pretty close to being human; I don\u2019t think we should, you know, kick it as it\u2019s trying to walk along or wipe out its memory that it built up over thousands of simulated hours, or even real hours of experience. So, measurement is a big thing, for sure. It\u2019s so philosophical, and I love thinking about those things, but they\u2019re definitely outside of my purview, really.<\/p>\n<p><strong>Host: Expertise, and&#8230;<\/strong><\/p>\n<p>Adam Trischler: Absolutely.<\/p>\n<p><strong>Host: &#8230;even task. You\u2019re not actually paid at the Montreal lab to come up with these deep philosophical answers. It\u2019s like, \u201cGet the machine-learning done, baby!\u201d<\/strong><\/p>\n<p>Adam Trischler: Yes.<\/p>\n<p><strong>Host: Oh. Adam Trischler, it\u2019s been fantastic talking to you this morning. I suspect we\u2019ll be seeing the fruits of your labors in ways that we might not even expect, but I\u2019m looking forward to watching where you\u2019re going and what\u2019s going on in Montreal. It looks fantastic.<\/strong><\/p>\n<p><strong>(music plays)<\/strong><\/p>\n<p>Adam Trischler: Thank you.<\/p>\n<p><strong>Host: And I look forward to seeing the machine that\u2019s going to understand me and talk to me in my old age.<\/strong><\/p>\n<p>Adam Trischler: This has been a lot of fun, thanks.<\/p>\n<p><strong>Host: To learn more about Dr. Adam Trischler, and the quest for literate machines, visit <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/\">Microsoft.com\/research<\/a>.<\/strong><\/p>\n<p>&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Episode 16, March 21, 2018 &#8211; Dr. Trischler talks about his dream of making literate machines, his efforts to design meta-learning algorithms that can actually learn to learn, the importance of what he calls \u201cfew-shot learning\u201d in that meta-learning process, and how, through a process of one-to-many mapping in machine learning, our computers not may not only be answering our questions, but asking them as well.<\/p>\n","protected":false},"author":37074,"featured_media":473241,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"msr-url-field":"https:\/\/player.blubrry.com\/id\/32518391","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"footnotes":""},"categories":[240054],"tags":[],"research-area":[13556,13545],"msr-region":[],"msr-event-type":[],"msr-locale":[268875],"msr-post-option":[],"msr-impact-theme":[],"msr-promo-type":[],"msr-podcast-series":[],"class_list":["post-473232","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-msr-podcast","msr-research-area-artificial-intelligence","msr-research-area-human-language-technologies","msr-locale-en_us"],"msr_event_details":{"start":"","end":"","location":""},"podcast_url":"https:\/\/player.blubrry.com\/id\/32518391","podcast_episode":"","msr_research_lab":[437514],"msr_impact_theme":[],"related-publications":[],"related-downloads":[],"related-videos":[],"related-academic-programs":[],"related-groups":[629145],"related-projects":[],"related-events":[],"related-researchers":[],"msr_type":"Post","featured_image_thumbnail":"<img width=\"480\" height=\"280\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/03\/MicrosoftResearch_Podcast_Adam-Treschler_MCR-Academic-Carousel_480x280.jpg\" class=\"img-object-cover\" alt=\"a man wearing glasses and smiling at the camera\" decoding=\"async\" loading=\"lazy\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/03\/MicrosoftResearch_Podcast_Adam-Treschler_MCR-Academic-Carousel_480x280.jpg 480w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2018\/03\/MicrosoftResearch_Podcast_Adam-Treschler_MCR-Academic-Carousel_480x280-300x175.jpg 300w\" sizes=\"(max-width: 480px) 100vw, 480px\" \/>","byline":"","formattedDate":"March 21, 2018","formattedExcerpt":"Episode 16, March 21, 2018 - Dr. Trischler talks about his dream of making literate machines, his efforts to design meta-learning algorithms that can actually learn to learn, the importance of what he calls \u201cfew-shot learning\u201d in that meta-learning process, and how, through a process&hellip;","locale":{"slug":"en_us","name":"English","native":"","english":"English"},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/473232"}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/users\/37074"}],"replies":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/comments?post=473232"}],"version-history":[{"count":11,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/473232\/revisions"}],"predecessor-version":[{"id":487652,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/473232\/revisions\/487652"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/473241"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=473232"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/categories?post=473232"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/tags?post=473232"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=473232"},{"taxonomy":"msr-region","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-region?post=473232"},{"taxonomy":"msr-event-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-event-type?post=473232"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=473232"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=473232"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=473232"},{"taxonomy":"msr-promo-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-promo-type?post=473232"},{"taxonomy":"msr-podcast-series","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-podcast-series?post=473232"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}