Transcript of Professor Neil Gershenfeld’s Talk at AI World Society Summit

May 15th, 2019

 Transcript:

[00:00:02] Hi I’m Neil Gershenfeld director and like ts Center for Bits and Atoms and I chaired the fab foundation and I want to talk about.

[00:00:11] The future of A.I. and how the digital world relates to the physical world. So as background the Center for Bits and Atoms is created to look at the boundary between digital and physical. And we’ve done things like creating the first significant faster than classical quantum computations or creating synthetic life research rate at that boundary. And as background one of my students Jason Taylor built and runs all the computers at Facebook. One of my students Rafi recorded and built the computers for Twitter and then led the reboot of computing for the Democratic National Committee. One of my students spend wrecked 1 the test of time award from MIT. The big annual meeting and all of that. I’m not really a computer scientist and they’re not computer scientists but they could do that because they learned how to believe in physics not computing in a sense because computer science. The theory violates laws of physics. And once you understand how they relate you can do the kind of work I’m describing. Sowith that background it appears we’re now in a revolution. That’s what this meeting and a lot of attention is about. But it’s important to be aware that we’re really in about the fifth boom bust cycle that there have been these cycles of A.I. is going to solve all the problems that night is going to fail then it’s going to solve and fail. And we’ve been through about five of those. What that boom bust misses is the scaling. And so. Year by year processors have gotten faster. The memory you can process has increased the data you can store has increased the data you can collect over networks has increased. And so when you add all of that up together what it means is a brain does about 10 to the 17 operations per second. If you count the number of neurons and the rate they fire and the supercomputer now today does about 10 to the 17 operations per second. And so the computers have caught up to a brain in the number of operations they can perform and we would be fundamentally derelict in our duty if at that point the computers weren’t beginning to do things comparable to the brain. The real thing that happened wasn’t a breakthrough. It was the steady scaling increases so computers are matching the complexity of our brain. That’s what’s really leading to the result today. Today the steady scanning in processing speed storage networks but A.I. has a mind body problem because it doesn’t have a body essentially. And so What’s missed in those numbers is the supercomputer doing tend to the 17 operations a second is made out of about 10 to the 17 parts if you count to 10 followed by 17 zeros if you count the transistors. A billion dollar chip factory can make that many transistors in a month to produce all the parts you need a person you listening to this is doing that every second so you’re full of manufacturing machines molecular machines called Ariba zones and they’re placing tended the 17 parts every second. And so as you’re sitting there listening to this every second you’re making the complexity of the supercomputer and.

 [00:03:57] That’s done at the heart of that is through a process called Morpheus Genesis which is how genes give rise to physical form. Now that may sound a little remote from artificial intelligence but two of the fathers of computing one is Alan Turing. He gave us mark the modern models of computation. This is the last thing he studied. He studied how genes give rise to form. And John von Neumann gave us modern computer architecture. The last thing he studied was cell free producing machines how a machine can communicate a computation for its own construction. And so this is the literal mother of all A.I. problems. It’s the evolution of A.I. itself. How did intelligence create intelligence. And that didn’t come from the brain thinking it came from molecular intelligence. And so the way more for the genius this works is in one of the oldest parts of the genome. There’s a program and in the sense you think of as a computer program. But it’s a computer program that gives rise to form. And so your genome doesn’t store things like you have five fingers. It stores a program that produces five fingers and that may sound like a detail but it’s profound. One reason is a billion bases in your genome can specify a trillion cells. But the deeper reason is almost anything you did to the genome would either be fatal or inconsequential. But changes to these developmental programs are interesting. You can go from five to six fingers or fingers to webs or walking to flapping. And this is exactly the heart of A.I.. So what A.I. does is find representations. How you search for data hasn’t really changed what these A.I. algorithms do is represent where it’s an interesting place to search. And so in the same sentence evolution searches over programs that create life by finding a beautiful representation for this evolutionary search. And so this was the breakthrough of the last year last year in science according to Science Magazine. And it really is artificial intelligence but it’s natural intelligence. It’s embodied in your molecules and it’s what creates life. It’s what creates our intelligence. So now the connection to this meeting in the heart of what I want to talk about is to look now ahead at the scaling. And so we’re really living through a third digital revolution that unites the first two. So the first one was in communication. We used to send analog waves down the wire and they degraded with distance. Shannon Claude Shannon showed if you communicate digitally if the error is below a threshold the noise reduces exponentially. And what that means is unreliable devices can communicate reliably. And that observation gave us the Internet. John von Neumann built on that to show you can have an unreliable computing device but it can do a reliable computation again by detecting and correcting errors. So the first two digital revolutions were digital communication and computation which at heart means reliable operation with unreliable devices. That was kind of enshrined in Gordon Watts came to be called Moore’s Law. Gordon Moore one of the founders of Intel in 1965 plotted five data points for the transistors on a chip. And he showed if you take the logarithm they line up on a straight line and a straight line on a logarithm plot means something is doubling. So it’s going from one to two to four to eight. And so he saw it looked like this was doubling and he projected what if that happens for 10 years. And in fact his projection was wrong. It went for 50 years. And so that came to be known as Moore’s Law. And the scaling of Moore’s Law is what’s led to the digital revolution is transforming the world. And it led to the scaling I described that led to computers matching the complexity of reaching the complexity of a human brain that the largest computers.

 [00:08:34] So the heart of what I want to talk about is the same thing is now happening for going from digital to physical and this has come to be called Last as law. After Sherry Lasseter somebody I I worked with. And so we actually have more data than Gordon Moore had to see the same sort of exponential scaling but now not in digital computing or communication but in digital fabrication. Crossing the boundary from digital to physical. And so to look at the history of that M.I.T. invented computerized manufacturing in 1952. But that leap led up to the State of the art manufacturing today the most advanced things like 3D printing. But those are all analog. The computer is digital but there’s no information in materials life. Four billion years ago evolved what I described of this process of genes giving rise to form. That’s truly digital in the sense of digital computing and communication. And so you can just you can program that directly. My lab was part of a collaboration creating fully synthetic living organism organisms but we’re learning how to do that in. Inorganic systems in Engineered Materials. And so to trace what that history looks like around the same time that my team made that first computerized manufacturing it made a computer called the world and a few blocks. From where we are sitting right now. The whirlwind filled the building.

 [00:10:08] It was two floors of a building and it was the first significant real time computer a computer that could respond not to a batch of operations but in real time.

 [00:10:19] And you could trace really all modern computer operating systems grew out of that project. It’s very important historically and there is just one of those that fills the building. It’s the size of a planet in the same sense the first computerized manufacturing. There was one of those multimillion dollar initiative the whirlwind got transistor eyes did M.I.T. it was commercialized as the PDP. That was a mini computer and many computers historically went from filling a building to filling a room so he’d fill about a room of this size it would cost maybe one hundred thousand dollars way a few times. And so that was much too big for an individual but it was smaller than a whole organization. And on many computers that’s when the Internet e-mail video games word processing all the things you do with modern computing happened when computers reached the level of a workgroup. The analog for that for this digital fabrication revolution is something called Fab Labs. And in a moment I’ll tell you much more about that. So to continue tracing. The mini computer that cost one hundred thousand dollars there are about a thousand of them. Thousands of them and that’s roughly one per city. What came after that were hobbyist computers that weren’t useful but they were personal. And Microsoft and Apple and all of those companies grew out of these first hobbyist computers. And so there are millions of those roughly one per town and the analog to those isn’t just using a machine to make things but it’s actually machines making machines. Fab Labs making fab labs then the computers became truly personal. In the era of pieces and smartphones and there are now billions of those as many as people on earth and the research we’re doing for those is going from printing processes or cutting machines to assemblers that can make almost anything in one process can make things like integrated circuits are all the technology. And today we’ve reached what’s called the Internet of Things stage where a smart thought we might have trillions of smart devices like a thermostat and the thermostat has the computing power of the mini computer. But now one person might have thousands of those. And for those we’re waking up to things called self assemblers. And the reason I just skim through that is computing numerically went from one to a thousand to a million to a billion to a trillion over a 50 year period transforming every aspect of how we live in the same way I just described going from digital the physical is going through that same scaling each of those stages a thousand million billion trillion exists today in some form in the laboratory but it’s going to take between now and 50 years in the future for them to emerge. But the implication is that 50 years in the future. So in the same way that the Internet and all of that was created and many computers as a outreach project initially for the National Science Foundation we began setting up these fab labs and they fall in between the millions of dollars of tools they run in the lab at M.I.T. and the self replicating systems far in the future. So the fab lab today cost about one hundred thousand dollars it fills a room like this it weighs about two tons and it contains 10 different machines that together read computerized data and do manufacturing. It includes printers and lasers and precision milling and cutting and. Things like embedding and programming electronics. And once you have those tools you can make technology that grows food. You can make consumer electronics you can make furniture you can make houses boats icicles clothing just about everything you buy is a commercial product today. You can make when you have access to these digital fabrication tools. There are inputs you need but what like you can’t yet make the integrated circuits you need the research I described but with those you can really make just about anything. And so they’re not yet personal that’s coming down in costs but they’re a level of the community group and the dramatic thing that happened is we set up a few as an outreach project and then they’ve been doubling for the last decade every year and a half the number of these labs doubled. There’s about fifteen hundred now and they range from as far north to the top of Norway to the bottom of Africa. They’re in rural India. They’re at the bottom tip of South America. They’re in favelas. They’re in just about every sort of setting. And so in the same way that Africa got to skip landline phones and largely go to mobile phones. A significant part of the world is skipping the industrial revolution and go into this distributed manufacturing. And that in turn has a number of profound implications. One is for education. So behind me is that my tea’s campus which was added up a few years ago to businesses spun off from here in the world’s 10th economy it falls between the economic output of India and Russia from these just few square blocks. It’s not because the people here are usually smart it’s that this is a productive place for them to flourish. And what we’re finding is these fab labs all over the world are attracting exactly that profile of bright and then of outliers exactly the ones we see here. But now in rural African villages or in Arctic towns and so we started a program called the fab Academy where instead of traveling a distance to a central campus like this or instead of looking online at a screen students have peers in work groups with mentors and machines locally and we connect them with video and content sharing to make an educational network a distributed educational network that’s really growing. To tap the brainpower of the planet and so that was one unexpected thing and then another unexpected thing is the implication for academies and for cities and countries. So Barcelona for example has a great design sense but over 50 percent youth unemployment a whole generation can’t leave home and work. So my counterpart there fascinating guy art became the city architect the planner of Barcelona. And he started sending up fab labs in districts around the city. So in the same way you expect the city to provide clean water or electricity. The city is now providing the means to make the means to produce as part of urban infrastructure. And that launched a fab city initiative of many cities around the world of Detroit or Oakland or Mexico City or so signing up to this fab city initiative to turn their consumers into creators with the means to produce on the scale of a country of Bhutan for example is based not on gross domestic product but on gross national happiness which doesn’t mean they’re happy. But it means they measure well-being as the output of the economy. So it’s a profound initiative on the scale of a country. But they were limited physically by buying crap trucked in from India and China. And so we’re working closely with Bhutan to describe did play these labs throughout the country to take gross national happiness and make it physical. And so among the most sensitive issues right now in the world are diverging income income inequality. Tariffs economic races to the bottom. All of that package of news. And if you think about this connection of once you can go from digital to physical it’s fundamentally an end run around it. So digital fabrication is not separate. It completes the first two revolution digital computing is the means to think digital communication is the means to message digital fabrication lets bits become atoms and atoms become bits. And so if you go into a fab lab and you produce that sort of things you see me around in this office in the lab. It fundamentally changes a series of assumptions at the heart of these battles of income inequality and tariffs and taxes and all of that is the assumption that you need a job. You need a business to have a job to have work to get money should then be able to purchase something if you can go into the lab and make something. It fundamentally changes the equality of all of those things and really in a way it does an end run around it. Now it’s not utopia but if you think about the democratization that’s happened in computing and communication it does the same thing. So you could make some for yourself you could make it for your community. You could make it for your town a wonderful group of Fab Labs in Detroit. Run by Blair Evans called Insight focus has an explicit model of a third of the time in the lab. As for traditional economic activity for money a third is for a post salary economy that involves barter and exchange and community infrastructure and a third like boot time is economic activity. But not for money but for enrichment transformation. For improving yourself and your community really revisiting these very basic assumptions about what is an economy what is work what is money how do you meet people’s needs in a way it’s a very old idea to break global supply chains and consumption but it rests on the ability to think globally to be part of these global networks. But fabricate locally. So in the first two digital revolutions it took us decades. You could take Gordon Moore’s plot in 1965 but it took. US decades to catch up to spam. Fake News viruses. Differences in access to computing. We don’t need to wait a few decades now. We have a moment right now where we can shape how this revolution is going to unfold and there’s very interesting data points for that in the U.S. Congress in the House Representatives Foster and Massie and then the Senate. Senators Van Hollen and Murkowski are introducing a bill to do in the U.S. What Barcelona did which is universal access to digital fabrication a new notion of a national laboratory made out of connected local labs and so in a world where you do that a lot of what a government does today you don’t need if you just make a product in the lab the way it’s done today. It comes in say in a ship to a port you need a port you need somebody to build the port you need to figure out how you tariff the thing coming in. Then it needs to go down say to a train then it needs to go to a highway then it needs to get to a building and then you need a cash register and you need to account for the sale. If you just make it for yourself you eliminate much of that global supply chain and much of the function of what a government does. On the other hand if you have the ability to do it yourself you need a whole bunch of new functions government doesn’t do today about empowering and enabling this to make it efficient effective safe and all of that. But in the world where this is so distributed you can’t do it by command and control you can’t legislate it. You have to opt in and add value to the networks. And so merging fabrication with communication and computation fundamentally challenges how an economy works. It fundamentally challenges what are the functions of government performs. But it gives a real hopeful opportunity to not keep fighting the same battles we’ve been fighting for many years but step around them and empower anybody to make almost anything anywhere. So now to step back and conclude I started by talking about foundations of computing and how it’s led to what seems to be a breakthrough in a i i then connected with. The most profound part of a I for me is not how we think but how we evolved the ability to think the creation of thinking so not just A.I. but to create A.I. that creates a I that involves intelligence but it’s not simply software it’s a molecular intelligence that goes from digital the physical we’re not learning to build that technology ourselves it’s growing exponentially and it’s one of the most significant and disruptive things I know for the future of the planet. To do that we’ve been initially frustrated by working with schools for education and non-profits for aid and governments for governance and businesses for economy and we kept tripping over that because if anybody can make anything anywhere it breaks all of those boundaries. And so probably the hardest part of this whole story but the most interesting part is we had to build a whole new set of organizations around anybody being able to make anything anywhere. How do you live learn work and play. And so there’s a really interesting group of organizational innovators helping lead that story and I’ll share some references afterwards with that. One of them is I recently wrote a book with my called Designing reality with my younger brother Allen who ran the biggest videogame studio and my older brother Joel who led the National Labor Relations organization tracing out how this many technical roadmaps relates to the social roadmap for the impact.

 [00:25:05] And so that’s what it means for a guy to become body to become physical. And I invite you to help join us in building that new world.

 [00:25:14] Thank you.

 – End of Transcript –

Professor Neil Gershenfeld speaks at AI World Society – G7 Summit Conference at Loeb House, Harvard University, April 25, 2019.

A summary this speech can be found here.