[Script Info]
Title: 
[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:17.82,0:00:21.32,Default,,0000,0000,0000,,Our world is changing in many ways
Dialogue: 0,0:00:21.32,0:00:25.98,Default,,0000,0000,0000,,and one of the things which is going\Nto have a huge impact on our future
Dialogue: 0,0:00:25.98,0:00:29.36,Default,,0000,0000,0000,,is artificial intelligence - AI,
Dialogue: 0,0:00:29.36,0:00:32.95,Default,,0000,0000,0000,,bringing another industrial revolution.
Dialogue: 0,0:00:33.63,0:00:39.50,Default,,0000,0000,0000,,Previous industrial revolutions\Nexpanded human's mechanical power.
Dialogue: 0,0:00:40.01,0:00:45.57,Default,,0000,0000,0000,,This new revolution,\Nthis second machine age
Dialogue: 0,0:00:45.57,0:00:50.12,Default,,0000,0000,0000,,is going to expand\Nour cognitive abilities,
Dialogue: 0,0:00:50.12,0:00:52.10,Default,,0000,0000,0000,,our mental power.
Dialogue: 0,0:00:52.78,0:00:57.18,Default,,0000,0000,0000,,Computers are not just going\Nto replace manual labor,
Dialogue: 0,0:00:57.60,0:00:59.90,Default,,0000,0000,0000,,but also mental labor.
Dialogue: 0,0:01:00.50,0:01:03.45,Default,,0000,0000,0000,,So, where do we stand today?
Dialogue: 0,0:01:04.03,0:01:07.72,Default,,0000,0000,0000,,You may have heard\Nabout what happened last March
Dialogue: 0,0:01:07.72,0:01:11.78,Default,,0000,0000,0000,,when a machine learning system\Ncalled AlphaGo
Dialogue: 0,0:01:11.78,0:01:17.71,Default,,0000,0000,0000,,used deep learning to beat\Nthe world champion at the game of Go.
Dialogue: 0,0:01:18.28,0:01:20.68,Default,,0000,0000,0000,,Go is an ancient Chinese game
Dialogue: 0,0:01:20.68,0:01:24.16,Default,,0000,0000,0000,,which had been much more difficult\Nfor computers to master
Dialogue: 0,0:01:24.16,0:01:25.98,Default,,0000,0000,0000,,than the game of chess.
Dialogue: 0,0:01:26.89,0:01:32.09,Default,,0000,0000,0000,,How did we succeed,\Nnow, after decades of AI research?
Dialogue: 0,0:01:33.07,0:01:36.70,Default,,0000,0000,0000,,AlphaGo was trained to play Go.
Dialogue: 0,0:01:37.68,0:01:41.30,Default,,0000,0000,0000,,First, by watching over and over
Dialogue: 0,0:01:41.81,0:01:46.89,Default,,0000,0000,0000,,tens of millions of moves made\Nby very strong human players.
Dialogue: 0,0:01:47.75,0:01:52.50,Default,,0000,0000,0000,,Then, by playing against itself,\Nmillions of games.
Dialogue: 0,0:01:54.22,0:01:59.94,Default,,0000,0000,0000,,Machine Learning allows computers\Nto learn from examples.
Dialogue: 0,0:02:00.46,0:02:02.58,Default,,0000,0000,0000,,To learn from data.
Dialogue: 0,0:02:03.88,0:02:07.24,Default,,0000,0000,0000,,Machine learning\Nhas turned out to be a key
Dialogue: 0,0:02:07.24,0:02:11.64,Default,,0000,0000,0000,,to cram knowledge into computers.
Dialogue: 0,0:02:12.17,0:02:14.07,Default,,0000,0000,0000,,And this is important
Dialogue: 0,0:02:14.07,0:02:19.30,Default,,0000,0000,0000,,because knowledge\Nis what enables intelligence.
Dialogue: 0,0:02:20.44,0:02:26.77,Default,,0000,0000,0000,,Putting knowledge into computers had been\Na challenge for previous approaches to AI.
Dialogue: 0,0:02:27.52,0:02:28.74,Default,,0000,0000,0000,,Why?
Dialogue: 0,0:02:29.06,0:02:33.86,Default,,0000,0000,0000,,There are many things\Nwhich we know intuitively.
Dialogue: 0,0:02:34.60,0:02:38.08,Default,,0000,0000,0000,,So we cannot communicate them verbally.
Dialogue: 0,0:02:38.62,0:02:42.78,Default,,0000,0000,0000,,We do not have conscious access\Nto that intuitive knowledge.
Dialogue: 0,0:02:43.27,0:02:46.69,Default,,0000,0000,0000,,How can we program computers\Nwithout knowledge?
Dialogue: 0,0:02:47.66,0:02:49.11,Default,,0000,0000,0000,,What's the solution?
Dialogue: 0,0:02:49.31,0:02:55.34,Default,,0000,0000,0000,,The solution is for machines to learn\Nthat knowledge by themselves,
Dialogue: 0,0:02:55.34,0:02:56.44,Default,,0000,0000,0000,,just as we do.
Dialogue: 0,0:02:56.44,0:03:03.19,Default,,0000,0000,0000,,And this is important because knowledge\Nis what enables intelligence.
Dialogue: 0,0:03:03.19,0:03:06.97,Default,,0000,0000,0000,,My mission has been\Nto contribute to discover
Dialogue: 0,0:03:06.97,0:03:12.68,Default,,0000,0000,0000,,and understand principles\Nof intelligence through learning.
Dialogue: 0,0:03:13.17,0:03:18.12,Default,,0000,0000,0000,,Whether animal, human or machine learning.
Dialogue: 0,0:03:19.45,0:03:25.07,Default,,0000,0000,0000,,I and others believe that there are\Na few key principles,
Dialogue: 0,0:03:25.07,0:03:27.30,Default,,0000,0000,0000,,just like the law of physics.
Dialogue: 0,0:03:27.88,0:03:32.74,Default,,0000,0000,0000,,Simple principles which could explain\Nour own intelligence
Dialogue: 0,0:03:32.74,0:03:36.74,Default,,0000,0000,0000,,and help us build intelligent machines.
Dialogue: 0,0:03:37.88,0:03:41.60,Default,,0000,0000,0000,,For example, think about the laws\Nof aerodynamics
Dialogue: 0,0:03:41.60,0:03:48.04,Default,,0000,0000,0000,,which are general enough to explain\Nthe flight of both, birds and planes.
Dialogue: 0,0:03:49.15,0:03:55.38,Default,,0000,0000,0000,,Wouldn't it be amazing to discover\Nsuch simple but powerful principles
Dialogue: 0,0:03:55.38,0:03:59.19,Default,,0000,0000,0000,,that would explain intelligence itself?
Dialogue: 0,0:04:00.03,0:04:03.39,Default,,0000,0000,0000,,Well, we've made some progress.
Dialogue: 0,0:04:04.38,0:04:10.86,Default,,0000,0000,0000,,My collaborators and I have contributed\Nin recent years in a revolution in AI
Dialogue: 0,0:04:11.78,0:04:16.40,Default,,0000,0000,0000,,with our research on neural networks\Nand deep learning,
Dialogue: 0,0:04:16.40,0:04:20.95,Default,,0000,0000,0000,,an approach to machine learning\Nwhich is inspired by the brain.
Dialogue: 0,0:04:22.04,0:04:25.24,Default,,0000,0000,0000,,It started with speech recognition
Dialogue: 0,0:04:25.24,0:04:29.96,Default,,0000,0000,0000,,on your phones,\Nwith neural networks since 2012.
Dialogue: 0,0:04:30.98,0:04:35.65,Default,,0000,0000,0000,,Shortly after, came a breakthrough\Nin computer vision.
Dialogue: 0,0:04:36.68,0:04:43.09,Default,,0000,0000,0000,,Computers can now do a pretty good job\Nof recognizing the content of images.
Dialogue: 0,0:04:43.67,0:04:50.05,Default,,0000,0000,0000,,In fact, they approach human performance\Non some benchmarks over the last 5 years.
Dialogue: 0,0:04:50.71,0:04:54.72,Default,,0000,0000,0000,,A computer can now get\Nan intuitive understanding
Dialogue: 0,0:04:54.72,0:04:58.19,Default,,0000,0000,0000,,of the visual appearance of a Go-board
Dialogue: 0,0:04:58.19,0:05:01.76,Default,,0000,0000,0000,,that is comparable to that\Nof the best human players.
Dialogue: 0,0:05:01.76,0:05:03.45,Default,,0000,0000,0000,,More recently,
Dialogue: 0,0:05:03.45,0:05:06.58,Default,,0000,0000,0000,,following some discoveries made in my lab,
Dialogue: 0,0:05:06.58,0:05:11.41,Default,,0000,0000,0000,,deep learning has been used to translate\Nfrom one language to another
Dialogue: 0,0:05:11.41,0:05:14.44,Default,,0000,0000,0000,,and you are going to start seeing\Nthis in Google translate.
Dialogue: 0,0:05:15.19,0:05:18.19,Default,,0000,0000,0000,,This is expanding the computer's ability
Dialogue: 0,0:05:18.19,0:05:22.53,Default,,0000,0000,0000,,to understand and generate\Nnatural language.
Dialogue: 0,0:05:23.55,0:05:25.52,Default,,0000,0000,0000,,But don't be fooled.
Dialogue: 0,0:05:25.52,0:05:30.05,Default,,0000,0000,0000,,We are still very, very far from a machine
Dialogue: 0,0:05:30.05,0:05:34.03,Default,,0000,0000,0000,,that would be as able as humans
Dialogue: 0,0:05:34.03,0:05:37.59,Default,,0000,0000,0000,,to learn to master\Nmany aspects of our world.
Dialogue: 0,0:05:38.54,0:05:41.24,Default,,0000,0000,0000,,So, let's take an example.
Dialogue: 0,0:05:41.64,0:05:46.79,Default,,0000,0000,0000,,Even a two year old child\Nis able to learn things
Dialogue: 0,0:05:46.79,0:05:50.66,Default,,0000,0000,0000,,in a way that computers\Nare not able to do right now.
Dialogue: 0,0:05:51.77,0:05:56.17,Default,,0000,0000,0000,,A two year old child actually\Nmasters intuitive physics.
Dialogue: 0,0:05:56.97,0:06:01.91,Default,,0000,0000,0000,,She knows when she drops a ball\Nthat it is going to fall down.
Dialogue: 0,0:06:02.49,0:06:06.09,Default,,0000,0000,0000,,When she spills some liquids\Nshe expects the resulting mess.
Dialogue: 0,0:06:06.59,0:06:09.52,Default,,0000,0000,0000,,Her parents do not need to teach her
Dialogue: 0,0:06:09.52,0:06:12.98,Default,,0000,0000,0000,,about Newton's laws\Nor differential equations.
Dialogue: 0,0:06:13.84,0:06:20.20,Default,,0000,0000,0000,,She discovers all these things by herself\Nin a unsupervised way.
Dialogue: 0,0:06:21.35,0:06:27.71,Default,,0000,0000,0000,,Unsupervised learning actually remains\None of the key challenges for AI.
Dialogue: 0,0:06:28.18,0:06:33.01,Default,,0000,0000,0000,,And it may take several more decades\Nof fundamental research
Dialogue: 0,0:06:33.01,0:06:34.67,Default,,0000,0000,0000,,to crack that knot.
Dialogue: 0,0:06:34.67,0:06:40.90,Default,,0000,0000,0000,,Unsupervised learning is actually trying\Nto discover representations of the data.
Dialogue: 0,0:06:41.73,0:06:43.78,Default,,0000,0000,0000,,Let me show you and example.
Dialogue: 0,0:06:44.36,0:06:49.35,Default,,0000,0000,0000,,Consider a page on the screen\Nthat you're seeing with your eyes
Dialogue: 0,0:06:49.35,0:06:54.20,Default,,0000,0000,0000,,or that the computer is seeing\Nas an image, a bunch of pixels.
Dialogue: 0,0:06:54.99,0:07:00.11,Default,,0000,0000,0000,,In order to answer a question\Nabout the content of the image
Dialogue: 0,0:07:00.86,0:07:05.21,Default,,0000,0000,0000,,you need to understand\Nits high-level meaning.
Dialogue: 0,0:07:05.67,0:07:10.82,Default,,0000,0000,0000,,This high level meaning corresponds\Nto the highest level of representation
Dialogue: 0,0:07:10.82,0:07:12.32,Default,,0000,0000,0000,,in your brain.
Dialogue: 0,0:07:12.91,0:07:18.31,Default,,0000,0000,0000,,Low down, you have\Nthe individual meaning of words
Dialogue: 0,0:07:19.19,0:07:23.80,Default,,0000,0000,0000,,and even lower down, you have characters\Nwhich make up the words.
Dialogue: 0,0:07:24.81,0:07:27.68,Default,,0000,0000,0000,,Those characters could be\Nrendered in different ways
Dialogue: 0,0:07:27.68,0:07:30.88,Default,,0000,0000,0000,,with different strokes\Nthat make up the characters.
Dialogue: 0,0:07:31.56,0:07:34.84,Default,,0000,0000,0000,,And those strokes are made up of edges\N
Dialogue: 0,0:07:34.84,0:07:37.28,Default,,0000,0000,0000,,and those edges are made up of pixels.
Dialogue: 0,0:07:37.28,0:07:40.45,Default,,0000,0000,0000,,So these are different\Nlevels of representation.
Dialogue: 0,0:07:41.08,0:07:44.24,Default,,0000,0000,0000,,But the pixels are not\Nsufficient by themselves
Dialogue: 0,0:07:44.24,0:07:46.58,Default,,0000,0000,0000,,to make sense of the image,
Dialogue: 0,0:07:46.58,0:07:51.90,Default,,0000,0000,0000,,to answer a high level question\Nabout the content of the page.
Dialogue: 0,0:07:52.93,0:07:57.59,Default,,0000,0000,0000,,Your brain actually has\Nthese different levels of representation
Dialogue: 0,0:07:57.59,0:08:02.29,Default,,0000,0000,0000,,starting with neurons\Nin the first visual area of cortex - V1,
Dialogue: 0,0:08:02.29,0:08:04.60,Default,,0000,0000,0000,,which recognizes edges.
Dialogue: 0,0:08:04.60,0:08:09.33,Default,,0000,0000,0000,,And then, neurons in the second\Nvisual area of cortex - V2,
Dialogue: 0,0:08:09.33,0:08:12.80,Default,,0000,0000,0000,,which recognizes strokes and small shapes.
Dialogue: 0,0:08:12.80,0:08:17.06,Default,,0000,0000,0000,,Higher up, you have neurons\Nwhich detect parts of objects
Dialogue: 0,0:08:17.06,0:08:19.99,Default,,0000,0000,0000,,and then objects and full scenes.
Dialogue: 0,0:08:21.18,0:08:24.76,Default,,0000,0000,0000,,Neural networks,\Nwhen they're trained with images,
Dialogue: 0,0:08:24.76,0:08:28.86,Default,,0000,0000,0000,,can actually discover these types\Nof levels of representation
Dialogue: 0,0:08:28.86,0:08:32.78,Default,,0000,0000,0000,,that match pretty well\Nwhat we observe in the brain.
Dialogue: 0,0:08:33.64,0:08:38.80,Default,,0000,0000,0000,,Both, biological neural networks,\Nwhich are what you have in your brain
Dialogue: 0,0:08:38.80,0:08:42.83,Default,,0000,0000,0000,,and the deep neural networks\Nthat we train on our machines
Dialogue: 0,0:08:42.84,0:08:48.08,Default,,0000,0000,0000,,can learn to transform from one level\Nof representation to the next,
Dialogue: 0,0:08:48.37,0:08:53.30,Default,,0000,0000,0000,,with the high levels corresponding\Nto more abstract notions.
Dialogue: 0,0:08:53.30,0:08:57.56,Default,,0000,0000,0000,,For example the abstract notion\Nof the character A
Dialogue: 0,0:08:57.56,0:09:00.89,Default,,0000,0000,0000,,can be rendered in many different ways\Nat the lowest levels
Dialogue: 0,0:09:00.89,0:09:03.89,Default,,0000,0000,0000,,as many different configurations of pixels
Dialogue: 0,0:09:03.89,0:09:09.10,Default,,0000,0000,0000,,depending on the position,\Nrotation, font and so on.
Dialogue: 0,0:09:10.44,0:09:15.82,Default,,0000,0000,0000,,So, how do we learn\Nthese high levels of representations?
Dialogue: 0,0:09:16.96,0:09:20.68,Default,,0000,0000,0000,,One thing that has been\Nvery successful up to now
Dialogue: 0,0:09:20.68,0:09:22.85,Default,,0000,0000,0000,,in the applications of deep learning,
Dialogue: 0,0:09:22.86,0:09:25.98,Default,,0000,0000,0000,,is what we call supervised learning.
Dialogue: 0,0:09:26.30,0:09:31.59,Default,,0000,0000,0000,,With supervised learning, the computer\Nneeds to be taken by the hand
Dialogue: 0,0:09:31.59,0:09:35.47,Default,,0000,0000,0000,,and humans have to tell the computer\Nthe answer to many questions.
Dialogue: 0,0:09:35.47,0:09:41.42,Default,,0000,0000,0000,,For example, on millions and millions\Nof images, humans have to tell the machine
Dialogue: 0,0:09:41.42,0:09:44.27,Default,,0000,0000,0000,,well... for this image, it is a cat.
Dialogue: 0,0:09:44.27,0:09:47.10,Default,,0000,0000,0000,,For this image, it is a dog.
Dialogue: 0,0:09:47.10,0:09:49.58,Default,,0000,0000,0000,,For this image, it is a laptop.
Dialogue: 0,0:09:49.60,0:09:55.60,Default,,0000,0000,0000,,For this image, it is a keyboard,\NAnd so on, and so on millions of times.
Dialogue: 0,0:09:56.07,0:10:01.03,Default,,0000,0000,0000,,This is very painful and we use\Ncrowdsourcing to manage to do that.
Dialogue: 0,0:10:01.46,0:10:03.40,Default,,0000,0000,0000,,Although, this is very powerful
Dialogue: 0,0:10:03.42,0:10:06.27,Default,,0000,0000,0000,,and we are able to solve\Nmany interesting problems,
Dialogue: 0,0:10:06.27,0:10:08.31,Default,,0000,0000,0000,,humans are much stronger
Dialogue: 0,0:10:08.31,0:10:12.08,Default,,0000,0000,0000,,and they can learn over many more\Ndifferent aspects of the world
Dialogue: 0,0:10:12.08,0:10:13.81,Default,,0000,0000,0000,,in a much more autonomous way,
Dialogue: 0,0:10:13.81,0:10:17.61,Default,,0000,0000,0000,,just as we've seen with the child\Nlearning about intuitive physics.
Dialogue: 0,0:10:17.62,0:10:23.74,Default,,0000,0000,0000,,Unsupervised learning could also help us\Ndeal with self-driving cars.
Dialogue: 0,0:10:24.57,0:10:26.10,Default,,0000,0000,0000,,Let me explain what I mean:
Dialogue: 0,0:10:26.10,0:10:31.84,Default,,0000,0000,0000,,Unsupervised learning allows computers\Nto project themselves into the future
Dialogue: 0,0:10:31.84,0:10:37.20,Default,,0000,0000,0000,,to generate plausible futures\Nconditioned on the current situation.
Dialogue: 0,0:10:38.37,0:10:42.90,Default,,0000,0000,0000,,And that allows computers to reason\Nand to plan ahead.
Dialogue: 0,0:10:43.45,0:10:47.98,Default,,0000,0000,0000,,Even for circumstances\Nthey have not been trained on.
Dialogue: 0,0:10:48.75,0:10:50.44,Default,,0000,0000,0000,,This is important
Dialogue: 0,0:10:50.44,0:10:53.95,Default,,0000,0000,0000,,because if we use supervised learning\Nwe would have to tell the computers
Dialogue: 0,0:10:53.95,0:10:57.40,Default,,0000,0000,0000,,about all the circumstances\Nwhere the car could be
Dialogue: 0,0:10:57.40,0:11:01.38,Default,,0000,0000,0000,,and how humans\Nwould react in that situation.
Dialogue: 0,0:11:02.45,0:11:06.19,Default,,0000,0000,0000,,How did I learn to avoid\Ndangerous driving behavior?
Dialogue: 0,0:11:07.28,0:11:10.79,Default,,0000,0000,0000,,Did I have to die\Na thousand times in an accident?
Dialogue: 0,0:11:10.79,0:11:12.11,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:11:12.11,0:11:14.61,Default,,0000,0000,0000,,Well, that's the way we train\Nmachines right now.
Dialogue: 0,0:11:15.18,0:11:18.34,Default,,0000,0000,0000,,So, it's not going to fly\Nor at least not to drive.
Dialogue: 0,0:11:18.34,0:11:19.93,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:11:21.29,0:11:25.66,Default,,0000,0000,0000,,So, what we need is to train our models
Dialogue: 0,0:11:25.66,0:11:31.29,Default,,0000,0000,0000,,to be able to generate plausible images\Nor plausible futures, be creative.
Dialogue: 0,0:11:31.29,0:11:33.93,Default,,0000,0000,0000,,And we are making progress with that.
Dialogue: 0,0:11:33.93,0:11:37.46,Default,,0000,0000,0000,,So, we're training\Nthese deep neural networks
Dialogue: 0,0:11:37.46,0:11:40.82,Default,,0000,0000,0000,,to go from high-level meaning to pixels
Dialogue: 0,0:11:40.82,0:11:43.30,Default,,0000,0000,0000,,rather than from pixels\Nto high level meaning,
Dialogue: 0,0:11:43.31,0:11:46.79,Default,,0000,0000,0000,,going into the other direction\Nthrough the levels of representation.
Dialogue: 0,0:11:46.79,0:11:50.46,Default,,0000,0000,0000,,And this way, the computer\Ncan generate images
Dialogue: 0,0:11:51.19,0:11:55.07,Default,,0000,0000,0000,,that are new images different\Nfrom what the computer has seen
Dialogue: 0,0:11:55.07,0:11:56.49,Default,,0000,0000,0000,,while it was trained,
Dialogue: 0,0:11:57.02,0:12:00.37,Default,,0000,0000,0000,,but are plausible and look like natural images.
Dialogue: 0,0:12:01.89,0:12:06.33,Default,,0000,0000,0000,,We can also use these models\Nto dream up strange,
Dialogue: 0,0:12:06.34,0:12:09.49,Default,,0000,0000,0000,,sometimes scary images,
Dialogue: 0,0:12:09.49,0:12:11.80,Default,,0000,0000,0000,,just like our dreams and nightmares.
Dialogue: 0,0:12:12.68,0:12:16.85,Default,,0000,0000,0000,,Here's some images\Nthat were synthesized by the computer
Dialogue: 0,0:12:16.85,0:12:19.83,Default,,0000,0000,0000,,using these deep charted models.
Dialogue: 0,0:12:19.83,0:12:21.65,Default,,0000,0000,0000,,They look like natural images
Dialogue: 0,0:12:21.65,0:12:24.55,Default,,0000,0000,0000,,but if you look closely,\Nyou will see they are different
Dialogue: 0,0:12:25.46,0:12:28.70,Default,,0000,0000,0000,,and they're still missing\Nsome of the important details
Dialogue: 0,0:12:28.70,0:12:31.06,Default,,0000,0000,0000,,that we would recognize as natural.
Dialogue: 0,0:12:31.100,0:12:33.95,Default,,0000,0000,0000,,About 10 years ago,
Dialogue: 0,0:12:33.95,0:12:38.92,Default,,0000,0000,0000,,unsupervised learning has been\Na key to the breakthrough
Dialogue: 0,0:12:38.92,0:12:42.44,Default,,0000,0000,0000,,that we obtained\Ndiscovering deep learning.
Dialogue: 0,0:12:44.14,0:12:48.06,Default,,0000,0000,0000,,This was happening in just few labs,\Nincluding mine at the time
Dialogue: 0,0:12:48.06,0:12:51.46,Default,,0000,0000,0000,,at a time when neural networks\Nwere not popular.
Dialogue: 0,0:12:51.46,0:12:55.22,Default,,0000,0000,0000,,They were almost abandoned\Nby the scientific community.
Dialogue: 0,0:12:56.39,0:12:58.94,Default,,0000,0000,0000,,Now, things have changed a lot.
Dialogue: 0,0:12:58.94,0:13:01.38,Default,,0000,0000,0000,,It has become a very hard field.
Dialogue: 0,0:13:01.38,0:13:06.93,Default,,0000,0000,0000,,There are now hundreds of students\Nevery year applying for graduate studies
Dialogue: 0,0:13:06.95,0:13:09.78,Default,,0000,0000,0000,,at my lab with my collaborators.
Dialogue: 0,0:13:11.01,0:13:16.63,Default,,0000,0000,0000,,Montreal has become\Nthe largest academic concentration
Dialogue: 0,0:13:16.64,0:13:19.39,Default,,0000,0000,0000,,of deep learning researchers in the world.
Dialogue: 0,0:13:20.18,0:13:26.12,Default,,0000,0000,0000,,We just received a huge\Nresearch grant of 94 million dollars
Dialogue: 0,0:13:26.13,0:13:29.80,Default,,0000,0000,0000,,to push the boundaries\Nof AI and data science
Dialogue: 0,0:13:29.80,0:13:36.07,Default,,0000,0000,0000,,and also to transfer technology of deep\Nlearning and data science to the industry.
Dialogue: 0,0:13:37.25,0:13:43.79,Default,,0000,0000,0000,,Business people stimulated by all this\Nare creating start-ups, industrial labs,
Dialogue: 0,0:13:43.79,0:13:46.91,Default,,0000,0000,0000,,many of which near the universities.
Dialogue: 0,0:13:48.54,0:13:49.62,Default,,0000,0000,0000,,For example,
Dialogue: 0,0:13:49.62,0:13:54.73,Default,,0000,0000,0000,,just a few weeks ago, we announced\Nthe launch of a start-up factory
Dialogue: 0,0:13:54.73,0:13:56.51,Default,,0000,0000,0000,,called 'Element AI'
Dialogue: 0,0:13:56.51,0:13:59.60,Default,,0000,0000,0000,,which is going to focus\Non the deep learning applications.
Dialogue: 0,0:14:01.56,0:14:05.72,Default,,0000,0000,0000,,There are just not enough\Ndeep learning experts.
Dialogue: 0,0:14:06.36,0:14:10.68,Default,,0000,0000,0000,,So, they are getting paid crazy salaries,
Dialogue: 0,0:14:11.03,0:14:17.21,Default,,0000,0000,0000,,and many of my former academic colleagues\Nhave accepted generous deals
Dialogue: 0,0:14:17.23,0:14:20.52,Default,,0000,0000,0000,,from companies to work in industrial labs.
Dialogue: 0,0:14:21.08,0:14:25.01,Default,,0000,0000,0000,,I, for myself, have chosen\Nto stay in university,
Dialogue: 0,0:14:25.01,0:14:27.17,Default,,0000,0000,0000,,to work for the public good,
Dialogue: 0,0:14:27.17,0:14:28.89,Default,,0000,0000,0000,,to work with students,
Dialogue: 0,0:14:28.90,0:14:30.59,Default,,0000,0000,0000,,to remain independent.
Dialogue: 0,0:14:30.60,0:14:34.84,Default,,0000,0000,0000,,To guide the next generation\Nof deep learning experts.
Dialogue: 0,0:14:35.29,0:14:41.02,Default,,0000,0000,0000,,One thing that we are doing\Nbeyond commercial value
Dialogue: 0,0:14:41.02,0:14:44.65,Default,,0000,0000,0000,,is thinking about the social\Nimplications of AI.
Dialogue: 0,0:14:45.88,0:14:50.03,Default,,0000,0000,0000,,Many of us are now starting\Nto turn our eyes
Dialogue: 0,0:14:50.03,0:14:55.99,Default,,0000,0000,0000,,towards social value added\Napplications, like health.
Dialogue: 0,0:14:56.46,0:14:58.96,Default,,0000,0000,0000,,We think that we can use deep learning
Dialogue: 0,0:14:58.96,0:15:02.70,Default,,0000,0000,0000,,to improve treatment\Nwith personalized medicine.
Dialogue: 0,0:15:03.96,0:15:05.67,Default,,0000,0000,0000,,I believe that in the future,
Dialogue: 0,0:15:05.67,0:15:10.36,Default,,0000,0000,0000,,as we collect more data from millions\Nand billions people around the earth,
Dialogue: 0,0:15:10.36,0:15:13.86,Default,,0000,0000,0000,,we will be able to provide medical advice
Dialogue: 0,0:15:13.86,0:15:17.25,Default,,0000,0000,0000,,to billions of people\Nwho don't have access to it right now.
Dialogue: 0,0:15:17.60,0:15:22.72,Default,,0000,0000,0000,,And we can imagine many other\Napplications for social value of AI.
Dialogue: 0,0:15:23.14,0:15:26.24,Default,,0000,0000,0000,,For example, something\Nthat will come out of our research
Dialogue: 0,0:15:26.24,0:15:28.58,Default,,0000,0000,0000,,on natural language understanding
Dialogue: 0,0:15:29.33,0:15:31.20,Default,,0000,0000,0000,,is providing all kinds of services
Dialogue: 0,0:15:31.20,0:15:34.06,Default,,0000,0000,0000,,like legal services,\Nto those who can't afford them.
Dialogue: 0,0:15:34.51,0:15:37.34,Default,,0000,0000,0000,,We are now turning our eyes
Dialogue: 0,0:15:37.34,0:15:41.13,Default,,0000,0000,0000,,also towards the social implications\Nof AI in my community.
Dialogue: 0,0:15:41.69,0:15:44.80,Default,,0000,0000,0000,,But it's not just for experts\Nto think about this.
Dialogue: 0,0:15:46.03,0:15:49.94,Default,,0000,0000,0000,,I believe that beyond the math\Nand the jargon,
Dialogue: 0,0:15:49.94,0:15:53.10,Default,,0000,0000,0000,,ordinary people can get the sense
Dialogue: 0,0:15:53.14,0:15:55.90,Default,,0000,0000,0000,,of what goes on under the hood
Dialogue: 0,0:15:55.90,0:16:01.19,Default,,0000,0000,0000,,enough to participate\Nin the important decisions
Dialogue: 0,0:16:01.19,0:16:06.55,Default,,0000,0000,0000,,that will take place, in the next\Nfew years and decades about AI.
Dialogue: 0,0:16:07.58,0:16:09.28,Default,,0000,0000,0000,,So please,
Dialogue: 0,0:16:09.93,0:16:16.23,Default,,0000,0000,0000,,set aside your fees and give yourself\Nsome space to learn about it.
Dialogue: 0,0:16:17.84,0:16:22.53,Default,,0000,0000,0000,,My collaborators and I have written\Nseveral introductory papers
Dialogue: 0,0:16:22.54,0:16:25.38,Default,,0000,0000,0000,,and a book entitled "Deep Learning"
Dialogue: 0,0:16:25.38,0:16:29.62,Default,,0000,0000,0000,,to help students and engineers\Njump into this exciting field.
Dialogue: 0,0:16:30.66,0:16:35.78,Default,,0000,0000,0000,,There are also many online resources:\Nsoftwares, tutorials, videos..
Dialogue: 0,0:16:36.31,0:16:41.21,Default,,0000,0000,0000,,and many undergraduate students\Nare learning a lot of this
Dialogue: 0,0:16:41.21,0:16:44.55,Default,,0000,0000,0000,,about research in deep learning\Nby themselves,
Dialogue: 0,0:16:44.55,0:16:47.84,Default,,0000,0000,0000,,to later join the ranks of labs like mine.
Dialogue: 0,0:16:49.37,0:16:55.17,Default,,0000,0000,0000,,Ai is going to have a profound\Nimpact on our society.
Dialogue: 0,0:16:56.65,0:17:01.67,Default,,0000,0000,0000,,So, it's important to ask:\NHow are we going to use it?
Dialogue: 0,0:17:03.37,0:17:07.90,Default,,0000,0000,0000,,Immense positives may come\Nalong with negatives
Dialogue: 0,0:17:07.90,0:17:10.17,Default,,0000,0000,0000,,such as military use
Dialogue: 0,0:17:10.80,0:17:15.36,Default,,0000,0000,0000,,or rapid disruptive changes\Nin the job market.
Dialogue: 0,0:17:15.95,0:17:21.63,Default,,0000,0000,0000,,To make sure the collective choices\Nthat will be made about AI
Dialogue: 0,0:17:21.63,0:17:23.07,Default,,0000,0000,0000,,in the next few years,
Dialogue: 0,0:17:23.07,0:17:25.14,Default,,0000,0000,0000,,will be for the benefit of all,
Dialogue: 0,0:17:25.14,0:17:28.56,Default,,0000,0000,0000,,every citizen should take an active role
Dialogue: 0,0:17:28.56,0:17:32.91,Default,,0000,0000,0000,,in defining how AI will shape our future.
Dialogue: 0,0:17:33.87,0:17:34.89,Default,,0000,0000,0000,,Thank you.
Dialogue: 0,0:17:35.06,0:17:39.40,Default,,0000,0000,0000,,(Applause)