Moore's Law at 40: Part 1
- 2005-May-12
These captions and transcript were generated by a computer and may contain errors. If there are significant errors that should be corrected, please let us know by emailing digital@sciencehistory.org.
Transcript
00:00:01 Good evening, ladies and gentlemen, and welcome to CHF.
00:00:06 And thank you so much for joining us.
00:00:10 We have in front of us, I think, 24 hours of intellectual feasting
00:00:16 as we undertake an investigation into how science and technology
00:00:21 have and still are continuing to shape and transform our worlds,
00:00:26 and of course the chemical sciences and technologies are right at the center of what we're talking about.
00:00:34 We'll hear from three different sort of groups of people over the next 24 hours.
00:00:41 There's those like Gordon Moore, Carver Mead, and Harry Sello,
00:00:46 who quite literally were present at the creation of the new world connecting chemistry and the electronic revolution.
00:00:54 There are those like Raj Gupta, Patrick Gelsinger, and Elsa Reich-Mannis,
00:01:00 who are continuing to shape that world and the emerging realities of tomorrow.
00:01:07 And there are those who reflect on and analyze the meaning of what is happening,
00:01:13 like Anna Lee Saxenian and Rodney Brooks, from whom we'll hear in just a few minutes now.
00:01:20 I want to give you my own one thought in relation to the strands that are the interpretation,
00:01:28 the meaning of what is happening.
00:01:31 And being right here in this building, I can scarcely do otherwise than begin by pointing to Ben Franklin,
00:01:41 who of course put the science of electricity very much on the public map.
00:01:47 And as you all surely know, this was Ben's stamping ground.
00:01:52 Literally his home was in our backyard right there, so we are where Ben was.
00:02:00 And of course Ben, who was a master publicist among other things,
00:02:04 made static electricity the most shocking thing of his age.
00:02:11 But what isn't quite so well known is that Ben was responsible for recruiting into science
00:02:19 the first individual who was really a world-class scientist here in the United States.
00:02:26 And I'm referring of course to Joseph Priestley, who Ben Franklin recruited into science.
00:02:33 And you can see that very clearly. I have here Joseph Priestley's first book.
00:02:39 We're delighted that we have it in our collections.
00:02:43 And its title is no less than The History and Present State of Electricity,
00:02:49 published just 238 years ago to analyze electricity.
00:02:56 And you can see that Priestley, who of course we usually think of in connection with the discovery of oxygen
00:03:01 and the development of chemistry as a science in itself,
00:03:06 here are his words 238 years ago.
00:03:10 Chemistry is the great field for the extension of electrical knowledge.
00:03:16 For chemistry and electricity are both conversant about the latent and less obvious properties of bodies.
00:03:26 And yet the relation to each other has been but little considered.
00:03:30 Well, here we are almost 250 years later, and we can surely say, well, we've been considering this relationship.
00:03:38 And Joseph, you were right. There's a lot to be discovered here.
00:03:44 And staying with the Philadelphia theme, of course, that mode of discovery takes us with a hot step and jump
00:03:52 to the creation of the Electrochemical Society in 1902.
00:03:57 And the Electrochemical Society was created here in Philadelphia in 1902.
00:04:03 And then we hop, step, and jump on forwards to the subjects we'll be looking at tomorrow.
00:04:10 But in that connection with the connection with ECS,
00:04:15 I'm delighted at this moment that ECS are the co-sponsors of this symposium and of this event.
00:04:23 And at this moment, I want to introduce Dennis Hess.
00:04:26 Dennis, who has a name chair in chemical engineering at Georgia Tech,
00:04:31 has enjoyed a distinguished career in the territory of electronic materials.
00:04:36 He's the editor of Electrochemical and Solid State Letters.
00:04:40 He was the 1996-97 president of ECS.
00:04:44 He holds numerous medals and awards in connection to his own work in microelectronics processing and electronic materials.
00:04:52 And Dennis will introduce our speaker this evening.
00:04:56 Dennis.
00:05:06 Thank you very much, Arnold.
00:05:07 I appreciate it.
00:05:09 I'm glad to see such a nice turnout this evening.
00:05:13 I thought I might say a few words about the Electrochemical Society
00:05:18 simply because it appears that we suffer from the same problem that Arnold alluded to,
00:05:23 and that is what's the connection between chemistry and electronics?
00:05:28 We have the question, what's the connection between electronics and electrochemistry?
00:05:32 So to do that, I have a few slides here that will go very quickly just to introduce you to this.
00:05:43 I'll start with a little background on the society.
00:05:51 As Arnold said, the society was founded in 1902, and in fact, we just held our centennial here three years ago.
00:06:00 So this was a very nice event, and we took advantage of some of the capabilities
00:06:06 and help from the Chemical Heritage Foundation to put all this information together.
00:06:12 We're an international society with more than 8,000 members.
00:06:15 These are split between electrochemistry areas and microelectronics or solid-state areas.
00:06:23 We hold two meetings a year, May and October, and in fact, our 207th meeting is coming up next week in Quebec City.
00:06:34 We publish two archival journals, the Journal of the Electrochemical Society
00:06:38 and Electrochemical and Solid-State Letters.
00:06:41 Both of these are split among the papers that are published there between electrochemical areas
00:06:47 and solid-state materials areas.
00:06:54 We have a rather long history in semiconductor materials and semiconductor processing.
00:07:01 1921, an electrothermics division was formed, and there, at that time at least,
00:07:09 the emphasis was on carbon electrodes, electric furnaces, even copper smelting.
00:07:18 That particular division became the high-temperature materials division
00:07:24 and holds symposia now in areas like chemical vapor deposition and solid oxide fuel cells.
00:07:33 In 1931, an electronics division was formed, and again, most of the materials covered in symposia there
00:07:39 were general electronics and fluorescence.
00:07:42 After the discovery of the transistor in 1947, the emphasis changed,
00:07:47 and much of the programming now is in electronic materials and microelectronics processing.
00:07:54 Dielectrics and insulation division was formed in 1945 with an emphasis on dielectrics used in capacitors and wire coatings.
00:08:05 The emphasis changed, as did electronics, after the discovery of the transistor,
00:08:11 and so starting in the 50s, there were a number of symposia
00:08:15 and continue to be a number of symposia in microelectronics processing and dielectrics.
00:08:20 The name change to dielectric science and technology took place in 1990.
00:08:25 The first sessions on semiconductor materials and processing really occurred in 1953
00:08:31 at the New York meeting of the Electrochemical Society.
00:08:35 About the same time, coincidentally, Biondi and Burns at Bell Laboratories
00:08:40 started encouraging scientists and engineers to get involved in the Electrochemical Society
00:08:47 with their areas of materials and microelectronics processing.
00:08:53 The primary home for that particular area started to come about at that time.
00:09:00 We've gone through a number of different symposia that we run and quite an extensive programming session.
00:09:07 We're constantly doing some new things, and one of the newer ones was starting in 2001
00:09:13 to have an international semiconductor technology conference that's held in China.
00:09:18 We actually had Jack Kilby be the plenary speaker for the first meeting of that one.
00:09:25 You'll notice in the lower right-hand corner our acronym.
00:09:29 We seldom go by Electrochemical Society anymore.
00:09:32 We're now really ECS, and you notice the tagline on there that says
00:09:37 the Society for Solid State and Electrochemical Science and Technology.
00:09:41 That pretty much describes what it is we do.
00:09:53 These are a number of symposia that we hold on a regular basis.
00:10:00 This is a short list compared to the actual list of symposia that are held at ECS meetings,
00:10:07 but you'll notice that these are very heavily incorporated into the microelectronics area.
00:10:13 We've been very fortunate to have participation in our society by pioneers in the microelectronics area.
00:10:22 I've selected just a few people here that you no doubt have heard of and met and know about.
00:10:29 Bruce Hene, Bruce Steele, and Jerry Woodall were all past presidents of the society.
00:10:34 A number of individuals who have been presidents of the society
00:10:38 have been from the solid state microelectronics area,
00:10:41 but these individuals are ones that I think are known to most.
00:10:47 We've also had a number of individuals serve as plenary lecturers.
00:10:51 On Monday morning of meeting week, we have a plenary lecture, and Bruce Hene did that.
00:10:57 Interestingly, Gordon Moore was plenary speaker twice in 1981 and again in 1997,
00:11:04 where we celebrated the 50th anniversary of the discovery of the transistor.
00:11:09 Les Hogan was also a speaker when he was at Motorola, and David Hodges from UC Berkeley.
00:11:19 Clearly, we have a strong connection and not exactly a passing interest in microelectronics
00:11:26 through electrochemistry and through the chemical end of the process procedures.
00:11:34 With that as an introduction to what we're about,
00:11:39 it's my pleasure to introduce our speaker for this evening, Dr. Rodney Brooks.
00:11:45 Dr. Brooks built his first artificially intelligent machine to play tic-tac-toe at the age of 12.
00:11:54 It's not exactly a surprise that he would be co-founder and chief technical officer of iRobot Corporation.
00:12:02 One of his books, entitled Flesh and Machines, How Robots Will Change Us,
00:12:07 explores the similarities and differences between humans and robots
00:12:12 and contemplates where those differences might lead.
00:12:15 In particular, as continued integration of robots into society takes place,
00:12:22 the boundary between flesh, as Dr. Brooks claims, and machines starts to blur.
00:12:30 I think that's already begun to happen.
00:12:32 Dr. Brooks received a degree in mathematics from Flinders University in South Australia
00:12:37 in computer science and a PhD in computer science from Stanford.
00:12:41 After he held research positions at Carnegie Mellon and at MIT and a faculty position at Stanford,
00:12:48 he joined the faculty at MIT in 1984.
00:12:51 He's currently the Fujitsu Professor of Computer Science.
00:12:56 He also is the director of the MIT Computer Science and Artificial Intelligence Laboratory.
00:13:01 Dr. Brooks has received numerous awards and accolades, and in the interest of time I won't list those,
00:13:07 but just to hit the highlights, he's a fellow of the American Association for Artificial Intelligence,
00:13:13 a fellow of the American Association for the Advancement of Science,
00:13:17 and a member of the National Academy of Engineering.
00:13:20 I'm very pleased to present Dr. Rodney Brooks, whose lecture this evening is entitled
00:13:25 The Age of Moore's Law, Cultural Perspectives, Societal Effects.
00:13:38 Thank you, and I must say I'm very honored to be here to celebrate the 40th anniversary
00:13:45 of the publication of Dr. Moore's paper.
00:13:48 And rather, besides being intimidated to be talking about the effects of this progress over the years,
00:13:55 I also notice many other people in the audience who sort of know what I'm talking about, which really scares me.
00:14:00 So I hope I don't make too many faux pas here.
00:14:09 By the way, just to intimidate me, someone said upstairs,
00:14:11 I hope you're not going to give a PowerPoint presentation.
00:14:16 The title of my talk says the cultural effects, and I thought, what's happening in culture these days?
00:14:23 Has anyone noticed how prevalent Texas Hold'em is these days?
00:14:27 So I renamed my talk, I'll See Your Quadratic and I'll Raise You an Exponential.
00:14:34 And the theme of my talk really is how exponentials are pervasive.
00:14:38 Exponentials change the world.
00:14:41 And they change us, they change what we're going to do.
00:14:45 We're here to celebrate an insight, of course, and that was Dr. Moore's paper.
00:14:50 Now, one of the things that really scares me is, these days, when you want to find something, you go on the web.
00:14:56 So I got a copy of Dr. Moore's paper, and I assume that this is an accurate copy that I read carefully,
00:15:02 but we'll see whether it is.
00:15:05 But the insight that Dr. Moore had, back in 1965, were quite radical.
00:15:14 In the first two paragraphs, this is the first two paragraphs.
00:15:17 The first one says about the future of electronics.
00:15:19 In the second paragraph, he says there'll be such wonders as home computers, automatic controls for automobiles,
00:15:27 and personal portable communications equipment.
00:15:31 This was, I think, a remarkable foresight to see the impact of how these integrated circuits that he was starting to build
00:15:40 were going to change the world.
00:15:42 And in the copy, at least, that I got from the web, there was a cartoon.
00:15:47 Is this cartoon the original?
00:15:49 And if you look in this cartoon, the person in the center is selling the handy home computer, small thing,
00:15:56 and it's right next to the cosmetics.
00:15:59 And the cartoonist here, and I don't know whether the cartoonist was Dr. Moore or not,
00:16:03 but the fact that seeing 40 years ahead that computers would be a commodity product in line with cosmetics,
00:16:10 which is indeed what has happened, to me is just very, very insightful.
00:16:17 The key graph in that paper was about the number of components in a single integrated function,
00:16:27 in the log coordinates going up, versus the year.
00:16:30 Now, this was at 1965, and the paper says there are about 50 components.
00:16:34 And in this log scale, then, it projects up, and the paper talks about out to 1975,
00:16:40 with 2 to the 16th, or roughly 65,000 components in a single component.
00:16:46 And in the paper, also, Dr. Moore says the size of the components at that time, back in 1965,
00:16:53 2 mils, 2 thousandths, would fit all the way up to 1975.
00:17:00 So it didn't really need to get smaller, but there was yield issues.
00:17:04 As I read the paper last night, carefully, but maybe I misinterpreted.
00:17:11 This was the first description of this exponential growth in the number of components in an integrated circuit.
00:17:21 Now, also, someone said, I hope that the talk is not going to be technical.
00:17:25 This is the only equation.
00:17:28 What defines an exponential?
00:17:30 Well, the differential equation.
00:17:31 This differential equation says the rate of change of stuff is proportional to the instantaneous amount of stuff that is around already.
00:17:40 And that gives you an exponential, something that doubles with some regular time period.
00:17:45 So, one question was, is this the explanation that the amount of stuff, the amount of computers around,
00:17:52 made for how quickly we were designing better computers?
00:17:56 And that's certainly one of the things that people are going to talk about tomorrow.
00:18:02 Does the presence of a computer of power P make it easier to build a computer of power WP, where W is greater than 1?
00:18:10 Well, here's an unverifiable anecdote.
00:18:13 And this is also from the web, but most of the sources say it's unverifiable.
00:18:19 Verified part, Apple Corporation bought a whole bunch of Cray supercomputers over the years to design their Macintoshes.
00:18:27 And the unverified part is supposedly, at least, that Seymour Cray, on hearing this, that Steve Jobs had bought a Cray supercomputer, said,
00:18:37 that's funny, I just bought a Macintosh to design the next Cray.
00:18:45 I hope this story is true, but I just don't know if it's true.
00:18:49 It's too good not to be true.
00:18:51 But there you see the idea of the computers, having more powerful computers leads you to more powerful computers,
00:18:57 although this is in a co-routining sort of strategy here.
00:19:01 But that certainly wasn't in play back in 1965.
00:19:05 These circuits were not making better computers at that time.
00:19:09 Soon afterwards they started to, but they weren't there right at the beginning.
00:19:14 So, exponentials happen in lots of places.
00:19:19 And Neil deGrasse Tyson, who's the director of the Hayden Planetarium, gives a great talk
00:19:24 where he explains looking at the issues of the astrophysical journal at the Princeton Library.
00:19:31 And halfway along the wall, this whole wall of journals, goes from 1895 to 15 years ago.
00:19:37 And the second half is the last 15 years.
00:19:41 So half of all research ever published in astrophysics have been published in the last 15 years.
00:19:46 But 15 years before that was true too.
00:19:48 So the wall keeps halving down the way.
00:19:52 And 15 before that, and 15 before that, and 15 before that.
00:19:56 The amount of knowledge, or the amount of published knowledge, in astrophysics has been an exponential growth.
00:20:02 And we look across many aspects of science and technology and see those exponentials.
00:20:07 So what makes these exponentials?
00:20:09 And really what makes exponentials is a number of factors coming together.
00:20:13 One is the existing level of adoption.
00:20:16 If you've got powerful computers, it helps you design the more powerful computers.
00:20:20 It helps you design the more high-density chips.
00:20:24 Part of it is the expectation of an exponential.
00:20:29 And one of the talks tomorrow will be how the fact that Dr. Moore said that there will be an exponential growth,
00:20:37 and said what that time constant alpha was,
00:20:40 made people sort of believe that that time constant alpha should be there,
00:20:44 and encouraged them to do the work to make that true.
00:20:48 So there was variation in what alpha was, and it got restated a little bit over time.
00:20:53 And then there's the cross-transfer from another exponential.
00:20:57 Sometimes you get an exponential growth in something because of something else.
00:21:01 So in the original paper, and please again, Dr. Moore, correct me if I'm wrong,
00:21:05 I don't believe in the original paper that it said besides more components, they'll also get faster.
00:21:16 They got faster because they got smaller, and a bunch of other factors coming together.
00:21:20 But the original paper was talking about the number of components rather than the speed,
00:21:24 which is often another characteristic that we think of when we think about computers,
00:21:28 that they have gotten faster and faster in an exponential sort of way.
00:21:31 And there's a transfer from one exponential domain to another,
00:21:35 with good reason for that transfer, but it's not a direct generator of the exponential.
00:21:43 So the existence of the law helped to drive the exponential at the predicted rate.
00:21:48 It's been going for quite a while now, 40 years.
00:21:51 Can we still make use of more of it?
00:21:54 We've got these fast computers.
00:21:58 You can send a lot of email, you can browse those web pages, you can play your movies,
00:22:03 you can do all that stuff.
00:22:05 Heck, who needs a faster computer?
00:22:07 Who needs a computer with more memory?
00:22:09 Who needs a computer with that stuff?
00:22:12 What if Moore's Law just stopped today?
00:22:14 Wouldn't we be happy forever after?
00:22:17 Haven't we come to the end of innovation?
00:22:19 I want to show you a few examples of things from the lab right now,
00:22:24 which are different than what we have in our everyday lives,
00:22:26 because they take 10 hours or 4 days to compute.
00:22:31 But if we had them, we might actually use them,
00:22:33 in the same way we carry our cell phones around.
00:22:35 They might be useful.
00:22:37 The first one I want to show you is more better digital photography.
00:22:43 Actually, I'm going to jump out of PowerPoint here.
00:22:47 Let me show you what's going to happen here.
00:22:52 We start with the image on the left, an input image, a single image.
00:22:56 And after lots of computation,
00:22:58 we get a three-dimensional model of that scene from a single image.
00:23:04 So imagine, walking around with a digital camera,
00:23:06 you take an image, and not only do you have that image,
00:23:09 but you have the full three-dimensional model
00:23:12 of the thing that you've just taken the photo of.
00:23:15 If we have 13 more doublings of computation speed,
00:23:18 we'll be able to do this in just seconds.
00:23:21 So there's some room for some more Moore's Law there.
00:23:25 And this is the work of Fredo Durand and his colleagues at CSAIL.
00:23:29 Let me jump out of here and go to a movie of this.
00:23:35 This is the original photo.
00:23:37 And after lots of computation and lots of inference,
00:23:40 lots of stuff happening, we get a full 3D model.
00:23:49 And you'll see parts of the building appear from behind
00:23:52 that weren't in the original.
00:23:54 So it has to infer what it must look like, generalize,
00:23:57 and that's where all the computation is going on,
00:24:00 figuring out what shapes must be behind the building.
00:24:06 So you see some other stuff that was occluded before
00:24:09 now becoming visible.
00:24:10 That's all synthesized in the background.
00:24:12 And as it moves around, there's going to be some change in lighting
00:24:15 from the original image also.
00:24:26 So, 2D sensor producing a 3D image.
00:24:29 Eh, we don't really need it.
00:24:30 We don't really need our cell phones.
00:24:32 We don't really need our iPods.
00:24:33 We don't need any of that stuff.
00:24:35 But it sure is fun.
00:24:45 Here's another technique.
00:24:46 This one takes about 10 hours of computing, motion magnification.
00:24:50 So here the idea is, I'm going to just tell you a little
00:24:54 and then I'm going to show you an example.
00:24:56 What I'm going to show you now will run in real time
00:24:59 with 13 doublings of computation.
00:25:01 And this is the work of Bill Freeman and his colleagues.
00:25:03 So I'm going to show you this little video of a movie
00:25:07 and then after 10 hours of processing on it,
00:25:09 and there's a little background by Antonio Torralba
00:25:13 in his very Spanish accent describing what's going on here.
00:25:17 This paper presents motion magnification,
00:25:19 a technical art like magnifying glass for visual motion.
00:25:22 It can act as a propulsion of your sequence
00:25:24 around the surface of the information
00:25:26 that we otherwise see in this world.
00:25:28 So this is the original image.
00:25:30 There are several objects moving.
00:25:32 Our goal is to magnify the motion of parts of the scene
00:25:35 that seem to be static.
00:25:37 Our approach relies on proportional estimation.
00:25:40 First, we find the trajectories of our reliable set of visual points
00:25:45 to represent the motions in the sequence.
00:25:47 By detecting feature points that have particular motion behavior,
00:25:50 we can look in at the regions
00:25:52 even when there are things that are still first
00:25:54 and when there is no special continuity.
00:25:56 The final representation of the motion of the sequence
00:25:59 is obtained by assigning every pixel
00:26:01 to one of the motion groups,
00:26:03 providing a layer representation of the region.
00:26:06 The user can now specify one of the layers
00:26:09 and amplify its motion.
00:26:12 Magnification of the motion
00:26:14 reveals seen regions not captured by the camera.
00:26:17 Therefore, a texture synthesis algorithm
00:26:20 is necessary to recreate the missing information.
00:26:23 Now we put the stuff in behind.
00:26:26 It's still not quite right over on the right there.
00:26:29 That will get fixed in a second.
00:26:32 So you're seeing all the actual motions in the world
00:26:35 being magnified there.
00:26:37 And now they're going to fix up
00:26:40 just those railings on the right.
00:26:44 So that's what that swing is really doing,
00:26:46 is she's swinging.
00:26:48 And here's another example.
00:26:52 So the guy on the left is trying to stand on his hand,
00:26:55 but we're magnifying how much his belly's moving on the right.
00:27:09 Now we're not used to seeing the world in that way,
00:27:11 but with just a whole lot of computation,
00:27:14 it's ready for us to be able to look at the world
00:27:17 in very different ways.
00:27:18 You can imagine not only the sorts of things
00:27:20 you can use this in industry for,
00:27:22 for measuring things,
00:27:24 but maybe even in your everyday life.
00:27:26 Monitoring your house,
00:27:28 monitoring what's going on,
00:27:29 seeing all sorts of things.
00:27:31 I think there'll be lots of applications of this
00:27:33 when we can do it in real time.
00:27:35 And it's only 13 doublings away.
00:27:37 So that's not too far.
00:27:39 I plan on being around.
00:27:42 And Pat, I hope Intel is going to continue
00:27:45 with all 13 doublings.
00:27:49 Here's one that doesn't...
00:27:52 We don't have to wait for 13 doublings.
00:27:54 We can do this in real time on a PC.
00:27:56 The trick now is to put it inside a digital camera,
00:27:59 in that little tiny processor that you've got
00:28:01 in your digital camera.
00:28:03 The human eye has a massive dynamic range
00:28:08 for light sensitivity.
00:28:10 We have a few different mechanisms.
00:28:12 We have a mechanical mechanism,
00:28:14 which changes the dilation,
00:28:16 which is the amount of light getting into our eyeball.
00:28:19 But then we have some dynamic changes in the retina
00:28:22 as we're...
00:28:24 for a certain amount of dilation
00:28:26 as we look at different parts of the scene.
00:28:27 So we see a very wide dynamic range.
00:28:31 Our rendering capabilities on a screen like this
00:28:34 are very, very small, relatively.
00:28:37 Only, really, in each color,
00:28:39 about six bits of depth
00:28:42 in red, green, and blue on a screen like this.
00:28:45 And on printed paper, it's really about the same order.
00:28:48 So you can't print or display,
00:28:51 or even on an LCD screen, it's even worse,
00:28:54 you can't print or display the range of brightness
00:28:57 that the human eye can see.
00:29:00 Digital cameras used to only collect a few bits per pixel,
00:29:04 but now they collect more bits per pixel than they can display.
00:29:07 So when you try and display this picture,
00:29:09 the sun is shining through the window in the end,
00:29:12 and it sort of looks like it's bloomed.
00:29:17 In the old CCD arrays, they actually did bloom.
00:29:20 The little buckets of electrons overflowed
00:29:22 into the neighboring ones.
00:29:24 This is just going to maximum brightness.
00:29:26 But if we go and look at what's happening inside the image
00:29:31 and the more bits that are actually collected,
00:29:34 and adjust how the brightness is at various points,
00:29:40 and take care of some spatial frequencies as you do that,
00:29:46 you can actually reconstruct something
00:29:48 which can be displayed in a normal sort of display.
00:29:51 You get the stuff over here in the dark,
00:29:54 the stuff that's really bright there,
00:29:56 in a thing that fits the human eye on this screen,
00:30:00 which only has a few bits of display.
00:30:02 This can be done in real time on a standard PC,
00:30:05 and now the challenge is to get it down
00:30:07 to run on a little tiny chip which will fit in a digital camera
00:30:10 so that even I can take good photographs.
00:30:13 So there's no problem.
00:30:15 Point into the sun with the person below the sun,
00:30:17 no problem, it will fix it.
00:30:20 So you can all see the benefit of that.
00:30:24 Also, audio-visual delivery.
00:30:28 I want to show you some work.
00:30:30 That last one was work of Fredo Durand.
00:30:32 I want to show you some work of Tommaso Poggio
00:30:35 and some of his students.
00:30:38 And here, this takes days, the stuff I'm going to show you.
00:30:43 This is not real time.
00:30:45 What we do, or what they do,
00:30:47 is they get a person and they take a video of the person
00:30:52 and then they analyse, of the person speaking,
00:30:55 then they analyse that video for days and days and days
00:30:59 and learn how the sound that the person makes
00:31:02 is related to their lip motion.
00:31:04 Then they can make a model of that person
00:31:08 say stuff they didn't really say originally.
00:31:11 LAUGHTER
00:31:13 So this is Mary. She's called Mary 101.
00:31:16 And here she's going to be saying things,
00:31:18 but this is not her saying them.
00:31:20 This is the synthesised version of her saying these things.
00:31:25 It's a real person's voice for this one, though.
00:31:28 So they're saying one-syllable words.
00:31:39 Now, as these things get longer,
00:31:41 you'll notice the synthesis around the lips, the neck is pretty good,
00:31:45 but you'll start to notice the blinking gets weird
00:31:48 because they didn't pay attention to that very well,
00:31:50 especially in the longer sentences.
00:32:07 Yeah, you see the blinking not quite working too well.
00:32:11 LAUGHTER
00:32:15 Now, this was trained on her speaking,
00:32:18 but now we'll give her something to sing.
00:32:30 That's not her singing at all.
00:32:41 And now we can change language, and it doesn't work quite as well.
00:32:52 Now, we might use this with fully synthesised voice and synthesised image.
00:32:57 So here's an information delivery service.
00:32:59 This is a synthesised voice.
00:33:01 Smoothed pretty well.
00:33:04 ..4695 from Greensboro is expected in Halifax at 10.08pm local time.
00:33:10 And it can be synthesised in other languages.
00:33:21 And if any of you have any doubt that what you see on TV is real or not,
00:33:26 once this gets quick enough,
00:33:31 here's some reality hacking.
00:33:36 This was only, with eight seconds of original video from Marilyn,
00:33:40 the system was trained.
00:33:50 So, my point with showing you a few things here was,
00:33:53 you know, our computers are fast,
00:33:55 they give us a lot of trouble already, they're so fast,
00:33:58 but there's more to come.
00:34:01 And we will find uses for faster and faster computers.
00:34:05 We'll find uses for more memory.
00:34:08 We'll find uses for lots of things.
00:34:10 Now I want to tell you some other exponential stories.
00:34:13 And the first one is sort of a case study,
00:34:15 and I thought I should try and relate things to chemistry somehow tonight.
00:34:18 So there is chemistry here, and it looks like...
00:34:20 Is anyone here from Kodak?
00:34:23 OK, good.
00:34:26 Here's a typical convoluted example of how exponentials
00:34:31 can change all sorts of systems in our society.
00:34:36 My own field, my field is robotics,
00:34:39 but I started out in computer vision, actually,
00:34:41 and I still held a thesis defense with one of my students
00:34:45 just this morning in computer vision.
00:34:49 Computer vision research used to require really expensive equipment,
00:34:54 and it was in the Stone Age.
00:34:58 There were just a few isolated... This is back in the early 60s.
00:35:01 Just a few isolated people here and there
00:35:03 trying to do computer vision research,
00:35:05 because it required a computer, and they were hard to come by,
00:35:08 and you had to process for hours and hours,
00:35:10 and it was really hard to get much time on a computer,
00:35:13 but it was even hard to get an image to process.
00:35:16 So Larry Roberts, who went on to be one of the people involved in the Internet,
00:35:20 did his PhD at MIT in 1963 using a TX0 computer,
00:35:26 and he had about three images of white objects against black backgrounds.
00:35:31 Takeo Kanade, who has been at Carnegie Mellon for many years,
00:35:35 was one of the founders of the Robotics Institute there.
00:35:38 In the late 1960s, he was at Kyoto University
00:35:41 doing a computer vision thesis.
00:35:45 He had one image that he processed,
00:35:47 and the bits that he processed,
00:35:49 his wife had put a grid down over the image
00:35:53 and had estimated the gray level to about four bits on each little pixel,
00:35:57 and that was where he got his image from.
00:35:59 So it was really expensive and really hard to get images.
00:36:02 By the time I did my PhD thesis in 1981,
00:36:06 I had three real images that I processed things on.
00:36:10 But today, I wouldn't consider granting a PhD to a student
00:36:15 unless their algorithm had run on billions of images.
00:36:18 Two things happened.
00:36:20 Certainly, the exponential increase in computer power
00:36:24 and the commoditization of computers was important.
00:36:27 But also, home VCRs exponentially, over time,
00:36:31 brought down the cost of a quarter megapixel CCD.
00:36:35 Because a standard TV image is a quarter of a million pixels.
00:36:39 Cameras were really, really expensive.
00:36:41 Once you started having the home video player,
00:36:44 people wanted to take their own pictures.
00:36:48 The cost got pushed down, down, down,
00:36:50 to now where you can buy a $2 complete camera
00:36:54 with a quarter megapixel from the thousands of dollars they cost
00:36:58 not too many years ago.
00:37:01 Two decades ago, they were still up in the tens of thousands of dollars.
00:37:05 So computer vision now flourishes,
00:37:08 and useful contributions can come from research labs all over the world.
00:37:12 It's completely democratized.
00:37:14 The barrier to entry is very low.
00:37:16 So that's the first part of the story,
00:37:18 how a consumer pressure,
00:37:23 wanting home video recordings,
00:37:25 or wanting to be able to record your own stuff,
00:37:28 to play on that home video,
00:37:30 pushed down the price exponentially over time
00:37:33 of a quarter megapixel CCD.
00:37:36 But after that commoditization, there was a new development.
00:37:40 There was an exponential growth in the number of pixels.
00:37:42 That started to go way up.
00:37:44 And it got to the point where it enabled digital cameras to replace film.
00:37:49 And this has had tremendous impacts on the industry,
00:37:52 which is still being felt today.
00:37:56 Yesterday, you may have read that Dan Karp,
00:37:58 the CEO of Kodak, is stepping down in June.
00:38:02 In 1980, when he was a young marketing executive,
00:38:06 he got his first glimpse of the battle ahead.
00:38:08 He was on a visit to the company's research labs,
00:38:10 and a scientist in Kodak demonstrated a 23-pound digital camera.
00:38:14 But he saw that that was going to be the future,
00:38:18 and that the digital technology would eventually bring a day
00:38:22 when the film business would go into steep decline.
00:38:24 That day is here.
00:38:27 In 1976, Kodak had 90% of the U.S. film market
00:38:31 and 85% of the camera market.
00:38:35 In 1983, the executives sort of knew about this,
00:38:39 but one of them, Leo Thomas,
00:38:41 said it's very hard to find anything with profit margins
00:38:44 like color photography that's legal.
00:38:46 That chemical process was a real profitable business.
00:38:50 And it was really hard for middle management
00:38:52 to wean themselves away from that
00:38:54 and worry about what was happening in the future.
00:38:56 But that exponential was on its way.
00:38:58 When an exponential is coming at you,
00:39:00 you better watch out.
00:39:04 When a new CEO came in in 1993 from Motorola,
00:39:08 he noticed that Kodak had spent $5 billion on digital R&D
00:39:13 but didn't have any products.
00:39:15 So the top management had been saying,
00:39:17 explore, explore, explore,
00:39:19 but it hadn't gotten down to how do we change the company.
00:39:23 And at one point in 1993,
00:39:25 they had 23 different digital scanner projects
00:39:27 scattered amongst the divisions.
00:39:29 So this is from January of this year,
00:39:31 Kodak cutting 12,000 to 15,000 jobs.
00:39:34 The goal is to cut fixed costs at a rate faster
00:39:37 than the decline in demand for film,
00:39:39 which could dip 30% in this year alone.
00:39:42 So we're in that decline.
00:39:44 So the lesson I wanted to show here was two things.
00:39:47 One, that things cross-couple,
00:39:50 but exponentials, you've got to watch out for
00:39:53 because they can really change your world
00:39:55 and change your business world.
00:39:57 So megapixel growth is now sort of saturated
00:39:59 for the general market.
00:40:00 There's not much point in getting more and more megapixels
00:40:03 in your handheld digital camera.
00:40:05 But as I pointed out before,
00:40:07 the bits per pixel is better than display technology.
00:40:10 So the value is now in processing the raw image
00:40:16 before the human eye gets to see it,
00:40:18 before it gets transferred to the human eye.
00:40:20 So that's going to change that business some more.
00:40:22 It's also going to change the display business.
00:40:24 And here's one example of a change in display business.
00:40:28 The pictures that we can take
00:40:31 are bigger than a projector or a display
00:40:35 that most people can afford right now.
00:40:37 So one way around that
00:40:39 is to have a whole bunch of projectors,
00:40:41 cheaper projectors, pointing at a screen.
00:40:43 And maybe they, you know,
00:40:45 if you try and line them up exactly,
00:40:47 that's really expensive.
00:40:48 And there's lots of companies out there trying to do that.
00:40:50 It's really hard to get them mechanically lined up.
00:40:52 So let's just let them overlap sloppily.
00:40:56 And let's assume that the displays
00:40:58 may be made by different manufacturers.
00:41:00 They're not even color-corrected for each other.
00:41:03 You know, they're just projected up there.
00:41:06 But then have a camera looking at what it sees,
00:41:09 maybe display very quickly some test patterns,
00:41:12 and then have that camera tell the computer
00:41:15 how to change what pixels are put on what projector
00:41:18 to account for changes in orientation,
00:41:20 overlap, changes in color.
00:41:22 So out of that, you get a very smooth-looking image
00:41:27 from multiple overlapping projectors.
00:41:29 And so there's a group up at MIT
00:41:31 who are trying to build a 32-megapixel display
00:41:34 out of cheap commodity projectors.
00:41:36 But here, what's happening here?
00:41:38 Here is replacing mechanical precision with computation.
00:41:43 In the digital camera system,
00:41:46 you might say, oh,
00:41:48 digital computation replaced the chemistry of film.
00:41:51 Actually, it didn't.
00:41:53 It moved the chemistry from film to silicon.
00:41:56 So chemists out there, this will still work for you.
00:41:59 It's just moving around.
00:42:01 You've got to follow it.
00:42:05 What else is happening in the exponential world?
00:42:07 Well, we're starting to get multiple cores on a chip.
00:42:10 We used to just have a chip with one processor on it.
00:42:13 Now, over the last 12, 18 months,
00:42:16 everyone's starting to put two cores on a chip.
00:42:21 Some for, not the notebook or laptop computer,
00:42:27 but for some routers and things,
00:42:29 are putting handfuls of cores on a chip
00:42:31 for embedded applications.
00:42:33 But this number is going to increase exponentially.
00:42:35 You're going to see more and more cores on a chip.
00:42:37 And that's going to change the whole structure of things.
00:42:40 This is one we built at MIT,
00:42:42 Anant Agarwal, a little while ago.
00:42:44 This is one that's building processors,
00:42:46 the raw processors on a chip.
00:42:48 This is another one that's being built now.
00:42:50 64 processors on a chip, each with 16 ALUs.
00:42:53 So that's 1,024 ALUs in there.
00:42:55 So the number of tiles is growing exponentially.
00:42:58 But the way in which software needs to be written
00:43:01 is going to be changed by that.
00:43:03 The value of innovation is shifting right now
00:43:07 from the hardware to the software,
00:43:10 and how you change that software.
00:43:12 Compilers that can compile stuff
00:43:14 that's written independently
00:43:17 of how many cores there are going to be,
00:43:19 and figure out how to splat it out
00:43:21 onto a chip with lots of cores,
00:43:23 are going to become more and more important.
00:43:26 And how that software works
00:43:28 is going to constrain the interconnects
00:43:32 that are used on the chips for these multiple cores.
00:43:34 And there will be an interplay between those.
00:43:36 There will be some winners and losers.
00:43:38 And things will go round and round for a while.
00:43:40 I'm sorry for the Intel people
00:43:42 where I say silicon just becomes a mere commodity.
00:43:45 But they're working on software, too.
00:43:49 Companies will have to change the game they're playing.
00:43:53 And this is an IBM chip, IBM, Toshiba, and Sony,
00:43:58 which is their cell chip,
00:44:01 which has a PowerPC and eight peripheral units.
00:44:04 There's going to be four of these
00:44:06 in the new PlayStation 3 from Sony.
00:44:08 And the typical caveat about quoted numbers
00:44:14 will be able to have a teraflop per second,
00:44:18 which is a lot of computation.
00:44:21 And it's going to be cheap.
00:44:23 Where else are exponentials happening?
00:44:25 Exponentials are happening in storage, personal storage.
00:44:28 And I like to think of the iPod as a unit.
00:44:31 Oh, there's one right there.
00:44:33 iPod is a unit of personal storage.
00:44:35 $400 iPod.
00:44:37 In 2003, the $400 iPod had 10 gigabytes.
00:44:44 In 2004, the $400 iPod had 20 gigabytes.
00:44:47 I went to Apple's website last night just to check,
00:44:50 because it's another year.
00:44:52 20 gigabytes is down to $249.
00:44:55 30 gigabytes is $349.
00:44:57 60 gigabytes is $449.
00:44:59 And it's so big, no one's got that much music,
00:45:01 so they sell it as the iPod Photo.
00:45:03 So roughly, $400, let's call it 40 gigabytes.
00:45:06 We've doubled again.
00:45:08 So every year, we're doubling the amount of storage
00:45:11 in this iPod for $400.
00:45:13 And then there are cheaper models with less storage
00:45:15 and more expensive models with more storage.
00:45:17 But let's just stick at the $400 level
00:45:20 and assume that doubling happens every year.
00:45:23 20 years from now, that means an iPod
00:45:26 will have 40 million gigabytes.
00:45:30 40 petabytes.
00:45:32 That's a lot of storage.
00:45:35 What could it be used for?
00:45:36 Raj Reddy at CMU has the Million Book Project,
00:45:38 where he's going around.
00:45:40 He's got digitization centers in India, China, Egypt,
00:45:43 digitizing a million books and making them free
00:45:45 across the world.
00:45:49 If you convert from the image, just extract the text,
00:45:53 that's only 500 gigabytes of text for a million books.
00:45:56 That's an iPod in 2009.
00:45:58 You can carry a text of a million books around on your hip.
00:46:01 That's a lot of books.
00:46:04 If you keep all the pages as image files,
00:46:07 that's 50 petabytes.
00:46:09 The Library of Congress has 20 million books.
00:46:13 That's a 2013 iPod as text.
00:46:17 Carry the Library of Congress around with you.
00:46:19 Now, we assume that the networks will be real good,
00:46:22 so you wouldn't need to carry that big, heavy thing around.
00:46:26 Current iPod, four full movies.
00:46:28 I went on to IMDb last night just to check how many movies they have.
00:46:32 They have, as of last night, 319,937 movies.
00:46:36 Now, I should point out, less than 60,000 of these had reviews,
00:46:40 so you don't need most of them.
00:46:42 They must be really bad,
00:46:45 because a lot of the ones that have reviews are really bad.
00:46:47 But if you include Bollywood, maybe 500,000 movies.
00:46:50 You project out a little bit how many movies there are.
00:46:53 An iPod in 2024, you'd better carry around every movie ever made,
00:46:58 including the 400,000 that you would never want to watch.
00:47:03 So the ones worth caring about, probably way less than 2020,
00:47:07 you'd better carry them all around.
00:47:09 Now, Recording Industry Association of America is getting worried.
00:47:12 There's been a whole fuss about downloading tunes.
00:47:16 Well, this guy's going to be carrying every movie ever made.
00:47:22 So it's going to change the politics of what our society worries about in gross ways.
00:47:29 That's just one other exponential.
00:47:32 What else are we seeing?
00:47:34 Low-level wireless.
00:47:35 Now we have barcodes that became pervasive.
00:47:38 Soon, RFID tags will be pervasive.
00:47:40 That's a way of wirelessly interrogating things.
00:47:43 Ad-hoc networks starting to form in the home.
00:47:46 802.15.4 and ZigBee.
00:47:50 I can never get that right.
00:47:51 I probably mistyped it.
00:47:53 Connect up devices to each other.
00:47:55 There's going to be a lot more sensors in the world
00:47:58 and a collection of sensor data and aggregation,
00:48:00 and that's going to increase radically over the next few years.
00:48:04 High-level wireless networks.
00:48:06 We're already seeing 802.11b is pervasive in Starbucks, airports, et cetera.
00:48:11 Here's my statistical experiment that I actually use it all the time.
00:48:18 In a cab in Manhattan on a cross street, when it stops,
00:48:22 at any time, my statistics say there are five 802.11b networks available or there,
00:48:31 and one of those five is not password protected.
00:48:34 So my wife calls me a Wi-Fi whore.
00:48:38 I just sit grabbing the Wi-Fi as I go.
00:48:43 And you can download a few e-mails at each stop.
00:48:48 We see 802.11g.
00:48:49 We see 802.16 coming.
00:48:52 There's going to be some convergence one would hope, GSM, 3G, et cetera.
00:48:56 So there's going to be much more pervasive wireless than we currently have,
00:48:59 and it's already changing the way we, certainly the early adopters,
00:49:04 use their computers and get their information as they move around the world.
00:49:08 There's other exponentials, too, and I'm just going to go through these without pictures.
00:49:13 Price of gene manipulation is going down.
00:49:18 Between, you know, the human genome is a few, what is it, 3.6 billion base pairs.
00:49:31 Is that right?
00:49:33 Someone tell me.
00:49:34 Roughly, okay.
00:49:35 But in terms of the number of places you and I differ,
00:49:38 the number of places we all differ is only at about 10 million places.
00:49:42 You and I only differ 3 million.
00:49:44 You and I only differ 3 million.
00:49:46 So a fairly small amount of stuff has to be extracted from that to uniquely identify people.
00:49:52 So that price is coming down, down, down.
00:49:57 As that technology comes down,
00:49:59 some of the things you've seen in some of the movies,
00:50:01 like taking a blood sample, identifying you genetically, uniquely,
00:50:06 will, I think, become commonplace.
00:50:10 You know, the saliva, you may have to carry your own napkins around
00:50:14 and keep them all so that no one can steal your identity.
00:50:18 It's not cost effective right now.
00:50:20 It may become cost effective.
00:50:22 You know, maybe you'll carry swabs of other people's saliva around.
00:50:25 I don't know.
00:50:27 It's going to change our society.
00:50:30 The volume of scientific literature,
00:50:32 I also already gave an example of that increasing exponentially,
00:50:34 but the machine, the last 15 to 20 years of scientific literature,
00:50:39 is all machine readable.
00:50:41 So going forward, most of the scientific literature ever made is machine readable.
00:50:46 And so that's amenable to Google and other sorts of search techniques.
00:50:50 That's going to change things also.
00:50:53 Wired bandwidth to the desktop.
00:50:56 I quoted exponentials because not all these things are exponential, exponential.
00:51:00 They're hard to estimate, but they're going up fairly rapidly.
00:51:03 Backbone bandwidth.
00:51:05 Number of cameras in the environment.
00:51:07 When you've got lots of cameras in the environment,
00:51:09 when they start getting networked together,
00:51:11 you can get full three-dimensional reconstruction,
00:51:13 full tracking of people as they move around.
00:51:16 There's already companies started,
00:51:18 I'm on the advisory board of one, to do this within stores
00:51:21 because right now they may have 100 cameras.
00:51:25 A gang comes in in broad daylight, steals stuff,
00:51:28 and they're trying to figure out who they were working with on the inside.
00:51:31 They've got 150 streams of video data,
00:51:34 and they have to manually go through and try and figure it out.
00:51:37 So people are working on algorithms which can say,
00:51:39 who did that person talk to?
00:51:41 Who did that person brush against in all these 150 video streams?
00:51:44 That's happening within a store,
00:51:46 as we get thousands and thousands of cameras outside,
00:51:49 and maybe even everyone's camera on their cell phones.
00:51:53 So there's all sorts of potentials for using all that data
00:51:55 for ways, good and bad.
00:51:58 The price and size of displays,
00:52:00 I think we're going to see change drastically over the next few years.
00:52:03 We need some chemistry to help with this.
00:52:06 The vocabulary size for spoken speech systems,
00:52:10 for interacting, was fairly small,
00:52:12 but it started to go up fairly drastically.
00:52:15 At C-Cell, over the last 12 months,
00:52:18 we've gone from 3,000 word vocabularies to 30,000 word vocabularies.
00:52:22 And we're going from one language to six languages.
00:52:25 So things are starting to change there.
00:52:27 Those sorts of interfaces will change.
00:52:30 The dimensionality of geometric algorithms,
00:52:33 this is pretty obscure.
00:52:35 Back in the old days, when you wrote FORTRAN programs,
00:52:37 you'd have a two-dimensional array or a three-dimensional array,
00:52:40 and you declared the array,
00:52:42 and pretty soon you filled up the computer memory.
00:52:44 Well, today, if you declare a 20-dimensional array,
00:52:47 it fills up the computer memory.
00:52:49 But if you can have algorithms which can represent sparse data
00:52:52 in a million-dimensional space,
00:52:54 you start to get all sorts of structure of the data
00:52:57 that you can't see otherwise.
00:52:59 So the theoreticians are developing
00:53:03 algorithmic representations
00:53:05 of bigger and bigger dimensional spaces,
00:53:08 which fit in a conventional computer.
00:53:10 And then you get new data analysis techniques,
00:53:12 which are at the core of all the machine learning work
00:53:15 that goes on in computer vision,
00:53:17 and speech understanding,
00:53:18 which leads to speech vocabulary sizes increasing drastically.
00:53:22 So a very obscure sort of thing,
00:53:24 measuring the dimensions of a representation
00:53:28 for geometry in the theory group.
00:53:32 Synthetic biology library size.
00:53:34 If you want to look at this, go to parts.mit.edu.
00:53:39 Some people at a number of universities
00:53:41 are working on the idea of taking computer programs,
00:53:47 compiling them to DNA sequences,
00:53:50 which then get inserted into the genome of a living E. coli,
00:53:54 and then you cook up 10 to the 12th of those E. coli.
00:53:57 And when the RNA transcription mechanism
00:54:00 reads that string of DNA,
00:54:02 it effectively does a digital computation.
00:54:04 A repressor protein stops another protein from being formed.
00:54:09 That's a NOT gate, and out of that you can build logic.
00:54:12 It's really slow.
00:54:14 It's 10-minute switching time.
00:54:16 But you get digital control over protein production inside cells.
00:54:21 And right now they have about 300 parts, standard parts.
00:54:25 They model it after the 7400 series TTL parts,
00:54:28 in concept at least,
00:54:29 that you can standardly put these parts together,
00:54:31 and make cells run programs,
00:54:34 and the cells can then communicate with each other,
00:54:36 have inputs and outputs.
00:54:38 So the long-term goal is to change
00:54:41 the basis of our industrial infrastructure.
00:54:43 Right now, we grow a tree, we cut it down,
00:54:46 and we build a piece of furniture.
00:54:48 The goal is, just genetically engineer the stuff
00:54:52 to grow the furniture.
00:54:54 And this is getting digital control over there.
00:54:59 And then the great one, brain-silicon interfaces.
00:55:03 We've seen a few of those.
00:55:04 We've seen them in rats.
00:55:06 We've seen them in monkeys.
00:55:08 Duke University, Brown,
00:55:10 had a whole series of experiments recently
00:55:13 where monkeys with silicon implants
00:55:15 and machine learning techniques
00:55:17 using high-dimensional representations
00:55:19 learned what the neurons were doing
00:55:23 as the monkey was controlling a joystick
00:55:26 to play a video game.
00:55:28 Then take the joystick away from the monkey.
00:55:30 The monkey just thinks, and it controls the video game.
00:55:33 So the monkeys are able to control
00:55:35 and play the video game by thinking.
00:55:37 They're able to drive a robot arm around
00:55:40 and get the robot arm to do stuff.
00:55:42 And on the human side,
00:55:45 maybe someone in this audience has a cochlear implant
00:55:49 where there's direct neural connections
00:55:51 from electronics
00:55:54 for people who have lost their hearing
00:55:57 in certain ways to get their hearing back.
00:55:59 And if you've been watching the news just over the last week,
00:56:01 there were a whole bunch of experiments
00:56:03 authorized for putting retinal implants for some patients.
00:56:07 And there are other experiments where there's a camera
00:56:10 going directly back into the V1 area of the brain.
00:56:13 So there's a medical push.
00:56:15 And then there's been a few clinical experiments
00:56:17 for quadriplegics,
00:56:19 giving them some of the same sorts of abilities as these monkeys.
00:56:22 They have a little bit of control over their life.
00:56:24 But as these technologies get developed
00:56:27 for good medical reasons,
00:56:29 and I don't think anyone would argue that someone's blind or deaf
00:56:32 shouldn't be given back some functionality,
00:56:35 or someone's quadriplegic shouldn't be given that functionality,
00:56:38 I think there'll be consumer push,
00:56:41 just like the home video,
00:56:43 to have neural interfaces.
00:56:45 And you might think, that could never happen.
00:56:47 Well, people inject Botox just to get rid of their frowns.
00:56:52 I think there's nothing to which the human race will not sink.
00:56:55 And we will see that over the next few years.
00:57:00 So to misquote Carl Feynman,
00:57:02 there's a lot of Zoom left at the top
00:57:05 with these exponentials going.
00:57:07 And we've all become, I think,
00:57:11 aware of the power of an exponential
00:57:13 through that marvelous paper,
00:57:15 which I just read from Dr. Moore.
00:57:17 And that has led the silicon revolution.
00:57:20 And we're going to see other things change in our lives over time.
00:57:23 Some will be good, some will be bad.
00:57:25 And we're all going to have to work at what we want and what we don't want
00:57:28 and see where it goes.
00:57:29 And I thank you for your attention.
00:57:31 Applause
00:57:45 Well, you can see we've had an absolutely astonishing
00:57:49 and wonderful glimpse into what may be coming ahead of us.
00:57:56 And that gives us the framework to tomorrow.
00:57:59 We shall be looking at some of the human stories,
00:58:03 some of the technical issues around the last 40 years
00:58:08 that have brought electronics and the chemical sciences to where they are
00:58:13 and have brought the first 40 years of Moore's Law.
00:58:17 Tomorrow morning, there's continental breakfast available from 8.30.
00:58:23 And we start at 9.00.
00:58:26 And we look forward to seeing you and continuing this discussion
00:58:32 and this wonderful glimpse into the past and future of chemistry and electronics.
00:58:38 And, Rodney, thank you again for a wonderful start.
00:58:42 Thank you.
00:58:43 Applause
00:59:13 Applause
00:59:42 Applause