YouSaid · the spoken record

Tom Mitchell

lines on the record: 60
first: 2019-06-19
most recent: 2019-06-19
sittings or episodes: 1
sources: podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

“We're already seeing an increasing number of embedded machine learning systems in continuous use. And as we see more and more of those in the Internet of Things and elsewhere, the opportunity for learning continuously for days and weeks and months and years and decades is increasingly there. We ought to be developing the ideas, the concepts of how to organize those systems to take advantage of that.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“It's long running and it has many different learning tasks. It's building up a knowledge base about the world. But to keep it short, I'll just say we've learned so much from that project about how to organize the architecture of a system so that it can invent new learning tasks as it goes, so that it can get synergy once it learns one thing to become better at learning another thing. How, in fact, very importantly, it can use unlabeled data to train itself instead of requiring an army of data labelers. So I just think this is an area that's relatively untouched in the machine learning field. But looking forward,”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Oh, fascinating. And there's something about sort of the longitudinal, right? We've been, we started in 2010 and just keeps on going. So it's not just like transfer learning from one model to another. It's like long running.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. You can add numbers, you can play music, all kinds of things. And a lot of those things we learn over time in a kind of synergistic way, in a staged sequence. First, you learn to crawl, then you learn to walk, then you learn to run, then you learn to ride a bike. And it wouldn't make any sense to do them in the other sequence because you're actually learning to learn. when you acquire one skill it puts you in a position that you now are capable of learning the next skill so i'm very interested in what it would mean to give a computer that kind of capability to do learning for days and weeks and years and decades And so we have a project we call our never ending language learner, which started in 2010 running 24 hours a day trying to learn to read the web.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Sure. Again, I just go back to what do humans do, the computers don't yet. And computers are very good at, say, learning to diagnose skin cancer. You give it some very specific task and some data. But if you look at people, people learn so many things.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“The computer is a fixed functionality. What if in that team the human could just teach the computer how they want the computer to help them do their job? It would be a completely different dynamic. It would change the future of work.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Exactly. Think about what it means for the future of jobs. Right now, if you have a computer introduced as your teammate, you, the human, and the computer or a team, well, the computer is frozen and the only, the teammate who gets to do the adapting is the human.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Exactly, and if this, then that is a wonderful thing, it has a huge library of these apps that you can download. But as you say, you still have to learn the language of the computer to create those. We're trying to have the computer learn the language of the person. If that line of research plays out, and I believe it will this decade. What we'll be in will be in a very different world because we'll be in a world where instead of the elite few, less than 1% of phone users being able to program, it'll be 99% of phone users who can do this. Now think about what that does for the whole conception of how we think about human computer interaction.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“I love that. It's sort of a natural language front end to we have an investment, a company called IFT, If this, then that, which is you can program those things, but you have to be a little sophisticated. You'd like to just be able to talk to your phone and have it figured out, how do I fill the slots into IFT that ift wants?”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So, with a combination of showing, demonstrating, and telling voice, we're trying to give users the opportunity to create their own apps, their own programs, with the same kind of instruction, voice, and demonstration, that you would use if you're trying to teach me. How to do it”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“But only far less than 1% of phone users can actually have taken the time to learn the language of the computer. We're giving the phone the chance to learn the language of the person. So with our phone prototype, if you say whenever snows at night, wake me up 30 minutes earlier, it says, I don't understand, do you want to teach me? And you can say, yes, here's how you find out if it's snowing at night. Open up this weather amp right here where it says current conditions, if that says SNOW, it's snowing. Here's how you wake me up 30 minutes earlier. You open up that alarm app and this number you subtract 30 from it.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Have the device statistically learn it. And if you go down that road, here's a really interesting angle on it. It becomes kind of like replacing computer programming with natural language instruction. So I'll give you an example of a prototype system that we've been working on together with Brad Myers, one of our faculty in HCI. It allows you to say to your phone something like, Whenever it snows at night, I want you to wake me up 30 minutes earlier. If you live in Pittsburgh, this is a useful app. And none of the California engineers have created that app. And today I could create that app if I took the trouble of learning the computer language of the phone. I could program it.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. So she doesn't learn that way. She learns by having a conversation with me. I go into the office and I say, Hey, this semester, I'm team teaching a course with Katerina on deep reinforcement learning. Here's what I want you to do. Whenever this happens, you do this. Whenever we're preparing to hand out a homework assignment, if it hasn't been pre-tested by the teaching assistants two days before handout, you send a note saying get that thing pre-tested. So what I do is I teach her and we have a conversation she clarifies. So one of the... New paradigm for machine learning that I predict we will see in the coming decade is what I'll call conversational learning. Use the kind of conversational interfaces that we have, say, with our phones. To allow people to literally teach their devices what they want them to do instead of”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So, going forward, I think a lot about how I want to invest my own research time. I'm interested still in machine learning. I'm very proud of the field of machine learning. It's come a long way. But I'm also somebody who thinks we're only at the beginning. Think if you want to know the future of machine learning. All you need to do is look at how humans learn and computers don't yet. So we learn, for example, We do learn statistically like computers do. My phone watches me over time and statistically it eventually learns where it thinks my house is and where it thinks my work is, it statistically learns what my preferences are. But I also have a human assistant. And if she tried to figure out what I wanted her to do, By statistically watching me do things a thousand times, I would have fired her so long ago.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“And now think about what are the kinds of intelligence that you could exhibit if you were deaf and blind. Well, you could do game playing and inventory control. You could do things that don't involve perception. But once you can perceive the world and converse in the world, there's an explosion of new applications you can do. So we're going to have garage door openers that open for you because they recognize that your car coming down the driveway. We're going to have many, many things that we haven't even thought about that just leverage off this very recent progress in perceptual AI.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Better than trades. So it's hard to remember that it's only been 10 years. And that's the thing about progress in AI. You forget because it becomes so familiar just how dramatic the improvement has been. Now think about what that means. That means we're really in the first five years. Of having computers that are not deaf and blind.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“When iPhone came out, you could not talk to your iPhone. Right. That's like such a weird idea, but you could not talk to your iPhone because speech recognition didn't work. And you know how, well, and now computers can transcribe voice to text just as well as people. Right. Similarly, when you pointed your camera at a scene, it couldn't recognize with any accuracy the things that were on the table in the scene.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“It really is different because we have real working stuff to point to. And over the next 10 years, we'll have a whole lot more real working stuff that influences our lives daily. So as a university researcher, I look at this and I say, where is this going and what should we be doing in the university”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. And so in the US, there was great concern that this would have a big impact. Japan would take over the economy. So there are some parallels here. Now, again, AI is very popular. People have great expectations. And there's a great amount of fear, I have to say, about what China and other countries might be doing. In AI. But one really, really important difference is that Unlike in the 1980s, right now there's a huge record of accomplishment over the last 10 years. We already have AI being and machine learning being used across many, many different Really economically valuable tasks. And therefore, I think really there's very little chance that we'll have a crash. Although I completely agree with my friends who say, but isn't AI overhyped? Absolutely it's overhyped. But there's enough reality there to...”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“I think we're absolutely at the best time ever for the field of artificial intelligence. And there have been, as you say, ups and downs over the year. And for example, in the late 80s, AI was very hot and there was great expectation. Of the things it would be able to do. There was also great fear, by the way, of what Japan was going to do. Yeah, this was the”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So, the key insight is to think of a job as a bundle of tasks, and that bundle might change over time as AI enters and says, well, look, this specific task I'm very good at in algorithm land. And so let's get humans to focus on other things. We just need to think of them as differently bundled Well, the last topic I wanted to talk with you about, Tom, is around whether this is the best time ever for AI research, right? So we started the grand campaign. You know, some would argue summer of 1956 with the Dartmouth Conference. And we've had several winters and summers. Where are we now? And then what are you most excited about looking into the future?”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. And so what you find if you look into this, and Eric and I recently had a paper in science with a more detailed study of this. But what you find is that the majority of jobs are not like toll booth operators where there's just one task. And if that gets automated, that's the end of the job. The majority of jobs like podcast interviewer are god. Or professor or doctor really are a bundle of tasks. And so what's going to happen is that According to our study, the majority, more than half of jobs are going to be influenced, impacted by automation, but the impact won't be elimination. It'll be a redistribution of the time that you spend on different tasks. And we even conjecture that successful businesses in the future will to some degree be redefining what the collection of jobs is that they're hiring for. Right.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. So, what does this mean about the future of doctors? Well, I think what it means is automation happens at the level of the individual tasks, not at the job level. If a job is a bundle of tasks like diagnosis, therapy, heart-to-heart chat, what's going to happen is computers will provide future doctors with more assistance, in some degrees hopefully automating billing, but some amount of automation or advice giving, but for other tasks like having that heart-to-heart chat, we're very, very far from when computers are going to be able to do anything close to that.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“But in other kinds of jobs, instead of the job going away, there'll be a shift, a redistribution of the tasks. So take, for example, doctor. A doctor has multiple tasks. For instance, they have to diagnose the patient. They have to generate some possible therapies. They have to have a heart-to-heart discussion with the patient about which of those therapies the patient elects to follow. They have to build the patient. Now, computers are getting, computers are pretty good at billing, but they're getting better at diagnosis and they're getting better at suggesting therapies. For example, you know, just in the last couple years, we've seen computers that are at the same level, if not a little better than doctors at things like diagnosing skin cancer and other kinds of diseases.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Well, this actually started with Eric and I co-chairing a National Academy study on automation in the workforce, which was a two-year affair with a committee of about 15 experts from around the country who were economists, social scientists, labor experts, technologists. And in that study, I think we learned so much. It turns out when you really dig into the question of what's going to be the impact of AI and automation on jobs. You can't escape noticing that there are many different forces that automation and technology is exerting on the workforce. One of them, of course, is automation toll booth operators are going away, do not sign up to be a toll booth operator.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, that's fascinating, right? Who is the master? Who is the student? The intuitive thing is let's make the humans the models that we train our systems to approach like human competence being the goal. The other way to think about it is no, we can actually introduce constraints like equal number of men and women or equal number of this ethnicity versus another ethnicity. And our algorithms as a result of those constraints could be more fair than humans. And so we invert it, right? Let's get humans up to that level of impartiality.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“True. And I mean, who knows which way it will go in the future if we continue to have human loan officers and some computer loan officers will we Up the constraints on the humans so that they pass the same qualifications? Or will we drop the constraints on the computers so that there are no more titrated than the people?”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. Yeah, it's really interesting that we sort of hold computers to a different standard because we're programming them. We can be explicit and we can have them sort of hit goals or not, right? And those are design decisions rather than sort of bundled into a brain of a person And so I think of, you know, look, banks historically have hired loan officers. Those loan officers may or may not be fair, right, according to the definitions that we're sort of talking about now. But we kind of hold those humans, those human loan officers, to a different standard than we would hold the algorithms.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“And so, even if you just think about self driving cars, we have when I was 16, I took a test and I was approved to be a human driver. They never asked me questions about whether I would swerve to hit the old lady or swerve to hit the baby carriage. Right.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“We don't have an objective measure. And there's a lot of activity right now to discussing that, including people like our philosophy professor David Danks, who is very much part of this discussion. And social scientists, technology people all getting together. And in fact, there are now a couple conferences centered around how to introduce fairness and explainability and trust in AI systems. This is a very important issue, but it's not only technical. It's partly getting our philosophical, social, trying to get our heads around what it is that we really want. That's the beautiful thing about AI and about computers in general. It forces you to be way more precise when you are getting a computer to do it about what you want.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, that's super interesting. We're sort of envisioning the world we want rather than the data of the world that we came from, right? Because we might not be happy with the representation of the representativeness, I guess, of the data that we came from.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“But if you're training examples have this kind of bias that maybe females where you see fewer loan approvals than males, there's some new work where people say, well, let's change that objective that we're trying to optimize. In addition to fitting the decisions that are in the training data as well as possible, let's put another constraint that the probability of a female being approved for a loan has to be equal to the probability of a male being approved. And then subject to that constraint, we'll try to match as many decisions as possible. So there's a lot of work right now in really technical work trying to understand if there are ways of thinking more creatively, more imaginatively, about how to even frame the machine learning problem so that we can take what might be.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“There's some interesting research going on now. So, for example, typically when we train a machine learning system, say to do loan approval, a typical thing would be you want to, you can think of these machine learning algorithms as optimization algorithms. They're going to tune the maybe the parameters of your deep network so that they maximize the number of decisions that they make that agree with the training examples.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Decision making that you want the program to make if you're going to train the program. And that's kind of the common sense notion of bias that most people talk about. But there's a lot of confusion in the field right now because bias is also used in statistical and machine learning to really with a very different meaning. We'll say that an algorithm is unbiased if the patterns that it learns, the decision rules that it learns for approving loans, for example, reflect correctly the patterns that are in the data. So that notion of statistically unbiased just means the algorithms doing its job of recapitulating the decisions that are in the data. The notion of the data itself being biased is really an orthogonal notion.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So that's Yeah, sure. And this is really a very important issue right now because now that machine learning is being used in practice in many different ways, the issue of bias really is very important to deal with. You gave an example, another example would be, for instance, you have some historical loan applications in which ones were approved, but maybe there are some bias that say people of one gender receive fewer loan approvals just because of their gender and if that's inherent in the data and you train a machine learning system that's successful, well it's probably going to learn the patterns that are in that data. So the notion of what I'll call social bias, socially unacceptable bias, is really this idea that you want the data set to reflect the kind of”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“We're a ways away from that. But I'm with you. Yeah. Awesome. So, another area that interests you is finding biases. And why don't we start by distinguishing sort of two types of biases? Because, you know, when you hear the word bias today in machine learning, you're mostly thinking about things like, gee, let me make sure my data set is representative so I don't draw the wrong conclusion from that, right? So the classic example being here that I don't do good recognition on people with darker skin because I didn't have enough of those samples in my dataset. And so the bias here is you've selected a very small subset of the target data set that you want to cover and make predictions on.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, that is fascinating. What's so exciting to think that the brain structures are identical across people because what everybody wants is sort of that, remember that scene in the matrix where you sort of like, you know, you're jacked straight into your brain and you're like, oh, now I know Kung Fu, right? Like this is what we want, right? We want to learn new skills and sort of new facts and new inferences just like loading an SD card, right? And so the fact that we are sort of converging to the same structures in the brain at least makes that theoretically possible.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Fascinating. What other people were thinking about? And we also found that the representations themselves are grounded in parts of the brain that are associated with perception. So if I give you a word like peach, the parts of your brain that code the meaning of that are the ones associated with the sense of taste and manipulation because sometimes you pick up a peach and visual color.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right. And so then we had a model that we trained with machine learning that captured something about representations in the brain. We used that to discover that. Representations are almost identical in your brain and mine. We could train on one set of people and decode,”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“By showing it, the two held out nouns and having it predict the images. Then we'd show it two images and we say, well, which of those is strawberry and which of those is airplane? And it was right 80% of the time. Wow.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“I put it on the screen in front of you. What happens during that 400 milliseconds? How do these representations evolve and come to be? And one of the most interesting things we found, we studied this question by training a machine learning system to take as input an arbitrary noun and to predict the brain image that we will see if a person reads that noun. Now we only had data for 60 nouns at that time, so we didn't train it on every noun in the world. We only trained it on 60. In fact, what we did was we trained it only on 58 so we could hold out two now today and scene. And then we would test how well it could extrapolate to new nouns it had never seen.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, it's an amazing thing. In fact, now that you mention it, I have to tell you, half of my research life these days is in studying how the human brain represents meaning of language. Use brain imaging methods to do this. And in one set of studies, we put people in an ephMRI scanner and we showed them just common nouns like automobile, airplane, a knife, a chair, and so forth. And we would get a picture, literally, with about three millimeter resolution of the three-dimensional neural activity in their brain as they think about these different words. And we're interested in the question of all kinds of fundamental questions, like what do these representations look like? Are they the same in your brain and my brain, given that they don't appear instantaneously? By the way, it takes you about 400 milliseconds to understand a word.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, and one of my favorite things about the brain, which is otherwise this very sort of slow computer, right? If you just look at neuron speeds, is that not only can they do this, but they can actually use this the representation they're deriving to actually inform our actions and our plans and our goals, right? So not only is it like this picture has a chair in it, but like I can sit in that chair, I can simulate sitting in that chair. I think like that chair is going to support my weight and all of these things happen in like milliseconds despite the fact that the basic components of the brain are very slow.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Better understanding, although we still don't fully understand, of how these artificial neural networks can learn very, very useful representations. And for me, a simple example of that, that in a sentence summarizes it, is we have neural networks now that can take as input an image of photograph and output a text caption for that photograph. What kind of representation must be in the middle of that neural network in order to actually capture the meaning well enough that you can go from a visual stimulus to the equivalent textual content, that it's really, it must be capturing a very basic core representation of the meaning of that photograph.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“So it's such a completely different notion of what it even means to represent knowledge. And really, one of the most exciting things that has come out of the last decade of research in deep networks is”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Definitely. And the level of abstraction at which the computational neural nets met up with the real biological neural nets was not a very detailed level, but where they Kind of became the same was this idea of distributed representations. That, in fact, it might be a collection of hundreds or thousands or millions of neurons that simultaneously were firing that represent your mother. Instead of a symbol”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Got it. And so neuroscientists at this time were making progress in understanding the structure of neurons and how they connected to each other and how they form connections and those connections could strength over time, right? All mediated by chemical interactions in the computer science community was inspired by this.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Right, fascinating really a computational efficiency argument. And therefore, Jerry would say There must be a lot of stuff happening in parallel. It must be a very wide chain of inference if it's only 10 layers deep. And then he says, look at the brain.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“Definitely, definitely the kinds of arguments Jerry Feldman was one of the people who gave some of these arguments. He said, The Q recognize your mother in about 100 milliseconds. Your neurons can't switch state in faster than a few milliseconds. And so it looks like at most the chain of inference that you're doing to go from your retina to recognize your mother can only be about 10 deep. Just from the timing.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source
“That we really are living in a golden age here in deep learning and neural network land. But let's go back to the original sort of rebel group, right? This is Jeff Hinton hanging out in the shadow of sort of first order logic and saying, no, this is going to work.”
2019-06-19 · a16z Podcast · a16z Podcast: The History and Future of Machine Learning · IDENTIFIED FROM THE TRANSCRIPT · source