Pioneering the Science of Information

good afternoon welcome to tonight’s IBM Centennial lecture on pioneering the science of knowledge to be delivered by the director of IBM Research and IBM senior vice president dr. John E Kelly my name is Glen Davis the vice-chancellor of the University of Melbourne and I’d like to begin proceedings in our customary way by acknowledging the traditional custodians of the land in which we meet this evening the one jury people of the cooler Nations but also on a more prosaic like nope invite you to turn off your phones and pages because this lecture will be filmed and broadcast and the audience probably wants to hear from John Kelly rather than that fabulous tune that you picked up and used as your tone from the University of Melbourne perspective I’d like to say that it’s an honor to welcome dr. Kelly and many of his colleagues from IBM on a much-anticipated two-day visit to Melbourne and Australia a partnership between IBM and the university has been developing in very exciting directions in recent years with a range of research initiatives touching on ICT the life sciences smarter cities sustainability the bionic eye and much much more and in that context I particularly like to acknowledge the presence this evening of Professor Peter Rathjen now the vice-chancellor of the University of Tasmania but a much-admired colleague of ours as we put this relationship together which began in 2007 it established a relationship of continuous exploration open to people from different parts of IBM and the University interested in working collaboratively on research on learning and on engagement and for both IBM and the University this was our first ever whole of University relationship it’s not about one faculty one school one program it’s about everything we do at this institution this week’s visit by dr. Kelly coincides with a series of important events in this developing partnership including tomorrow’s opening of the IBM R&D lab in Australia at 204 lagaan street another centrally important event is tonight Centennial lecture the lecture forms part of a series IBM is sponsoring campus locations around the world in this the hundredth year since the formation of the company in 1911 IBM has an extraordinary and diverse history as an information technology company yet I think one of the most interesting phases of this rich history is what’s happening in IBM today IBM CEO and president samp osmania has described the recent turn in the company from being a traditional multinational to becoming a globally integrated enterprise with the rise of supercomputing and the rapid expansion of knowledge in many fields one of IBM’s main aims now is to turn big data into big insight as the CEO put it with this fascinating history and exciting present in mind tonight’s lecture by dr. John Kelly will explore pioneering the science of information so to deliver the IBM Centennial lecture please join me in welcoming senior vice president and director of IBM Research dr. John Kelly well good afternoon everyone it’s a pleasure to be here at the University be here in Melbourne Australia as Glenn mentioned IBM is celebrating its centennial this year which is in industrial terms not a trivial task in the high-tech industry it’s a very rare occasion in fact the body count is one after a hundred years what I’d like to do this afternoon is talk about IT and where I T is going and many of the technologies that I think will change change the world there’s no doubt that information technology has played a critical role in the evolution of society in in the last century go back nearly a hundred years to tabulating machines and punch cards mechanical devices that were used to transform many aspects of our businesses go back 50 years to the IBM 360 which really revolutionized a number of different businesses whether it’s the airline business the banking industry the telecom industry etc or go back 30 years to the personal computer and how that brought IT out of the glass house into the hands of all of us and began to advanced computing in society in a much different way go back just 10 or 15 years to the internet which really in a sense interconnected and flattened the world and began to transform not only

businesses but many aspects of our life to today when technology is ubiquitous it’s in all of our pockets it is a constant communication device and just recently machines such as the Watson the Insignia know right that we’ll talk about a few minutes are a glimpse at where information technology is going in the future so this technology revolution that’s occurred over the last century has played a vital role in transforming businesses and society and as I said we’re a hundred years old and we like to think that we played a major all and much of that transformation over the last century going back to the punch card the system/360 that I mentioned some of these insignias you may or may not recognize the hard disk drive in the middle IBM invent that the hard disk drive invented many of the technologies that created the storage that we carry around in our personal devices today and we even embrace some very revolutionary different technologies the the penguin Linux open source we in we embraced an open source technology that was in a sense very threatening to our core business but we saw it as the future of the development of software and operating systems I point to the upper right one because it’s really the origin of IBM research and I think it’s relevant to where we stand today IBM Research was founded at Columbia University in New York City the first research center that mr. Watson our founder created was at a university because he realized that this is where the intellectual horsepower was and this is where you need to do research you need to be associated with great universities and that model today is the model we most often follow when we open new research labs around the world and this is our current CEO and chairman Sam Palmisano in a speech that he gave during our Centennial at the computer history museum and I think that this is a very interesting statement for one of the top fortune five CEOs in the world to make and here he recognizes that r & d play a very very significant role in our corporation and then I think in a sense that he would recognize that this is why IBM is still here after a hundred years in the technology industry and perhaps many others have failed and it’s this nature of both R&D and its importance that we think is incredibly relevant now I’m often asked well John why is IBM Research still alive and vital place when many of your competitors research organizations hour-long sense go on the bell labs a hitachi central research the xerox part and i think there’s a few reasons but i think the most important reason is that the research and innovation culture is fundamental to our business model we have decided as a company that we will compete based on innovation as opposed to other business models which may be supplied supply chain low cost etc we have decided as a company for 100 years that we will compete based on innovation the minute you make that decision you have to excel it research and you have to put your money where your mouth is and this gentleman does that this gentleman put six billion dollars a year into our D which is I think the largest in certainly in our industry but fundamentally believes in both let me translate that into sort of now the beginnings of the the technical presentation I’ve in our Centennial sort of SAT back and looked at technology and the growth of Technology in our industry over the decades and it’s very clear to me that the exponential growth in the capabilities of our largest computers or our handheld devices is the result over that long period of time of both incremental improvement development and radical innovation disruptive innovation that all of these technologies tend to get on these exponential curves but as you know no exponential is going to last forever and so what happens is we get on these technology curves but we reach a point where plateaus and we need a radical disruptive innovation to take the next step forward and if you look over decades or much less essential you’ll see this pattern of continual and if you back to sams comment that is

nothing more than R and D as you probably have gathered our agenda has moved from one of back office I t21 that we refer to as matter martyr planet but this is basically applying information technology to some of the world’s most complex systems and going from basically a reactive mode to a modeled predictive mode so the problems that were most interested in studying are not just single systems but real-world systems of systems just as in biology we’re made up of systems of systems the real world complex nature whether it’s physical systems man-made systems or others are all complex systems of systems no models exist in in that right hand area and we’re very interested in modeling and predicting and if you look at the left hand side it was primarily an analog world on the right very much a digital world and very much a statistical world that we need to deal with our research centers span the globe and I think this gives us a tremendous advantage as we do research into these areas of systems of systems as you can see we’re global we’re very proud that we’ve been expanding dramatically over the last several years around the globe doing real world research to give you a sense for the kinds of things that we’re doing in these centers that we’ve recently established in Dublin Ireland a focus on smarter cities we are co-resident at the city control center with the city management there is not a separate location we’re doing traffic management we’re doing a disaster recovery a number of different things there and Saudi Arabia we’re doing high performance computing around oil discovery an energy generation I’ll talk about the different agendas that we have here in Melbourne and Taiwan we’re very interested in health care to some very interesting emerging models that are developing on the island of Taiwan so this gives us a global reach it gives us a sense of what’s going on it gives us a sense of the art of the possible it creates a network around the globe for these research labs to to collaborate and in a sense it takes our researchers out of their brick-and-mortar labs and puts them into the real world because as you’ll see many of the problems we’re now working on the the world is our lab we cannot create in our laboratories the kinds of environments that we want to study in the real world so as Glenn mentioned just a year ago today we announced that we were going to open a research and development center here in Melbourne in Australia and we were going to do it associated with the with the university interestingly we also decided that we were going to run an experiment of doing both research and development together co-located with a university something we’ve never done before and what we’re tempt attempting to do with this model is do much more rapid transfer of Technology and innovation from the research mode into the development of products that we can rapidly deploy around the world so it’s a new model we’re experimenting in our d in the paradigm itself here in Melbourne the three areas that we’ve chosen to do research on here in Melbourne are shown on this slide first in their innocence their obvious but resource management from the discovery through the supply chain through the operations natural disaster management we just ran a whole seminar today on resilience societies and of course I don’t need to tell you about the impacts of natural disasters on society how do we sense how do we predict how do we model those how do we go from from predicting to reacting to recovery all part of what we want to study in that bill column and then because there’s such a tremendous capability around the Health Sciences and biological sciences here in Australia and here in the melbourne area this is an area that is ripe for working closely between our tremendous capability and data data management high-performance computing with those that are studying health care here at the University of Melbourne and across the bottom these are the types of areas that we’re interested in applying to these kinds of problems high performance computing stream computing that’s coming in out

very rapidly as I’ll discuss in a minute cloud offerings analytics etc now I want a bridge to a little bit of the future and talk about what i think are four technologies that are going to be transformative and are going to be the backbone of many of the types of systems that we want to work on and model let me begin at the bottom of the chart and talk about nano devices the computer chips that we use in our most advanced systems today pack roughly 11 billion transistors on a chip the size of your thumbnail tremendous compute power I remember 30 years ago when I joined IBM we were struggling to pack 1,000 transistors on a chip and we’re now doing 1 billion we are going to experience over the next decade or so another factor of a thousand we will have over one trillion devices on our semiconductor chips powering our large computing systems a thousand X that’s hard for the human mind to comprehend we’re used to ten percent twenty percent or doubling we’re not used to a thousand the implications of a thousand x are beyond what we can comprehend in terms of the capability of those systems going up the stack computer systems we now are in the pet what we call a petascale generation we are moving up by a factor of a thousand i’ll show in subsequent charts to the exascale generation exascale being a billion times a billion in data we’re moving from large terabyte petabyte data up by a factor of a thousand to a million to exit and zettabyte databases and again as we heard earlier supposing that not structured unstructured data which presents a tremendous problem for traditional computing but it has huge implications and then at the top probably most far-reaching but perhaps most interesting is we are going to go from computer systems that are programmed by humans the computer systems that learn and I’ll talk a little bit about the work we’re doing in that area so these factors of a thousand for a million or complete changes in paradigms and compute systems are going to have profound impacts in terms of what we’re going to be able to do with information technology in the future so I’d like to talk about each of these sequentially and let me start at the bottom behind computing for the last hundred years has been some type of switch or device over the last century that switching device has changed from mechanical through electromechanical through vacuum tubes transistors integrated circuits that I described if you look at this semi-log plot it’s it’s it’s incredible what we have done over the last century right we have gone up by what 15 orders of magnitude we’re entering a regime now in nanotechnology where we’re basically assembling atomically and through molecular assembly bottoms up in devices that are measured at the nano scale that’s going to allow us to again achieve trillions of devices in these computer chips it’s interesting again to look at this and I think we’re the only one again left standing we built in 1911 mechanical systems we survived every one of these transitions not only did we survive we let as a computer company and when we stand back and look at this this curve as a company if i put the company’s on this chart that competed with us in every one of those technologies you’d see very few of them make it over one transition in technology because they were not doing the research to understand what the technology was going to do next and so the field would be littered with companies that you would recognize or have heard about in the past and so when people ask me well why are you investing in nanotechnology why did you build a hundred-million-dollar nanotech center in Zurich Switzerland it’s because we want to not only lead in that generation of technology but we need to survive the leap over that barrier between those two technologies very interestingly when we work in the nano scale regime we also find other applications of the technology beyond computing devices and I show here just two examples on the top is something we call a DNA transistor and this sort of speaks to having an interdisciplinary lab I have biologists computer scientists nanomaterials folks working in the same lap what our folks

came up with is a device which it begins to address the challenge of how do we read every individual’s human DNA for less than a thousand dollars now why do you want to do that you want to do that so that you can get to personalized medicine at a very low cost well they invented a device which is a multi-layered device of electrodes with a thin nanopore or hole drilled through it with a special laser device constructed in such a way that we can pull a single strand of DNA through that hole and electrically read the DNA tremendous advances and we’re working with Roche and partnership to commercialize this device the bottom is a very interesting again side advantage of working in some of these materials we work in nano scale polymers for obvious reasons of insulators other types of materials that we use in these nano electronic devices our team discovered a set of polymer-based nanoparticle materials which have as you can see on the left is a bacteria on the right is a bacteria and these by the way our staff infection bacterias which are very very resistant to current antibiotics we discovered a material that we could make with great uniformity in size we could tailor the electrical charge of the surface of the particle such that it would attach itself that these bacteria penetrate the wall and kill the bacteria so the bacteria can try to mutate as fast as it wants it makes no difference to this particle and very interesting this is a whole class of materials that we think we can develop to address other types of bacteria so as you might imagine a number of pharmaceutical companies are interested in that kind of technology so a little bit off the main may an IT line but I think it’s very interesting that when you work in these fields you never know what you’re going to find in adjacent spaces moving up to large scale computing so as I mentioned we’re going to go up by a factor of a thousand again how we going to do that the state of the art today is shown on the left a pedophile pedal aired system or a blue jean p to achieve a petaflop of compute power requires 72 racks iraq being about the size of a refrigerator we’re shooting for that factor of a thousand before the end of a decade at exascale to do that as you can see we’re shooting for a factor of a thousand and compute power but we have tremendous tremendous energy barriers to overcome we need to hold it to a factor of about a hundred x not a thousand x these systems today consume megawatts of power if we don’t do something different these systems on the right will consume gigawatts and you’ll literally will need a power plant next to a large installation of one of these super computer centers we’re working on the technologies both at the processor level the memory level some of the three-dimensional packing of these devices working on using silicon photonics light as opposed to electrons for transport within these systems for reduced power but we know that we can achieve these numbers to put that in perspective what this means is these 72 racks at one petaflop will fit into a third of one rack when we achieve this 72 into a third of one rack now I don’t think anybody’s going to buy a third of Iraq Romus they’re going to want to buy 72 racks of that and have very large scale systems but again it’s it’s it’s hard for us to comprehend that kind of advance in technology but it will happen it will happen in the step from here to here we have systems that are in the tens and hundreds of pedal flops or the so-called blue jean cue system which will be an interim step to these larger exascale generation systems and that’s what the partnership here at the University of Melbourne is about is using that high performance computing in this collaboratory to generate new types of insights I’d like to say that these computer systems are analogous to the microscope they basically give us insights into the behavior of these molecules that we never had before with traditional microscopes they also allow us by the way to compress time because we can accelerate interactions and run them at real time or faster than real

time to run the experiment so we’re really thrilled to be able to apply our most advanced computing technologies to these biological and healthcare challenges here in Melbourne another application of these large supercomputers is in the area that we talked about before which is natural disaster prediction and prevention we have written computer models that give us dramatically improved ability to forecast weather on a much much finer grain resolution kilometer resolution than has ever been achieved before with other computing systems this we call it deep thunder this capability is at the core of what we’re doing in Brazil in Rio de Janeiro where you recall they had massive rain mud slides loss of life some time ago this system now is being deployed in Rio and several other cities in Brazil to do very very localized prediction of whether water runoff flood mudslides so again building these systems into a very resilient society in a place like real moving up up to data this is the one you’ll notice that not only is going up by a thousand but it’s going up by a million X we are generating data at rates we’ve never seen before in the history of mankind the question is how we going to extract knowledge from something that’s that fat large it’s not only going up in size it’s going up in the speed at which is coming at us this date is coming at us and requiring processing in milla and sometimes microseconds fractions of a second so we literally no longer have the time to take the data store it on a disk drive pull it out doing analysis and come out with an answer but a time you do that you’ve already been swamped by a tsunami of data that’s come over your bow during just that time and we have some very fast systems getting on and off of disk so we need real-time processing so here’s the problem that we’re facing we have enormous volumes I talked about going from terabytes to zettabytes of data we have that data coming at us from what used to be batch mode to literally streaming data and we have what used to be structured data in the olds are a banking world ok I’ve got ones and zeros and columns I can use relational databases to operate on that to this unstructured data multimedia data and the fourth dimension I didn’t put on this chart is that the data that we’re getting now in these newer real world applications is very noisy it is full of erroneous signals erroneous data if you look at the applications of some of this kind of information Homeland Security is an obvious one telcos and I’ll get to this one in a second but look at me we’re talking about hundreds of thousands of records per second whether it’s Homeland Security or what’s going on in toco networks and some of the largest telephones in the world when you look at the 50 billion per day 6 billion when you look at the response times you need millisecond decisions in a homeland security application particularly if it’s a cyber threat in some cases even milliseconds is too late your systems are corrupted in in a few milliseconds in the case of telcos and here’s a case where we’re working with Bharti the largest telco in India massive massive wireless traffic 10 milliseconds per decision on looking at what their customers are doing and how they want to tune and refine their infrastructure for their wireless network and then the system I hope by now everyone’s heard of this Watson Watson deep Q&A system that we that we research developed and demonstrated on this this game show Jeopardy but again this was a system that competed head-to-head with humans in an open domain question and answering field in that case we had three seconds to beat humans at at this open domain question within three seconds we had roughly under a second to understand the question under a second to find the information and we had about 500 milliseconds to decide whether we were going to bet or not in this in this game show that is very fast it is so little time once again that we could not go to disk drives all of the equivalent of 1 million books of information were in local memory in that system it had to have fundamentally instant recall of all

information at its disposal so whether it’s only a security massive telcos was these new analytics operations it’s it’s huge amounts of data and very little time to decide what to do with it another example is smarter traffic where we’re talking enormous amounts of data flowing in real-time 250,000 GPS probes per second in an average city going back to my comment on a noisy data I was over in our Dublin lab just a few weeks ago looking in in the real control center they have instrumented gps’s on all of their buses around the whole city and I’m looking at the screen of traffic flow for the whole city of Dublin and I noticed that two or three of the buses were not only off the road they were in the rivers and I said are you concerned about that oh no no that’s just noise it’s a bad GPS system so you get into these real-world applications and often the data is full of extraneous signals but you don’t know whether that’s true or not and so what the system has to do is look at the data around it look at temporal data before and try to predict is that a possible location of that bus could it be in the river or not do we need to send out a rescue team to to that bus at the top is this very interesting emerging area of computer science research where as I said we’re going to fundamentally learning systems this is important because we are no longer smart enough to program computer systems to deal with these real-world situations and as I showed you before we don’t have the time to be able to get in there and program or reprogram the system has to learn based on what it is seeing so on this journey the first system that we wanted to create in this regard was this this Watson system we demonstrated it again in the in this competition against two human beings I should first tell you that these two gentlemen Jennings and rudder are not normal human beings well they may be in this for him but they’re not in the world I’m from these these two fellows are incredible in fact I had a chance to meet with them individually before the actual competition and I asked them why are you this good how come you’re so fast it have what how do you study how do you recall something when you’re when you’re asked a question like that and independently they both gave me the same answer they said we don’t study anything I never sit down and try to learn something I just remember everything I’ve seen read and heard throughout my life I said well then how do you when faced with the question how does your brain work you know I’ll start to look for associated information and they say we don’t know it’s just instantly there it’s just instantly there they have no filing system the answer is just there and they trust that they’ve done this for so long they trust the information the answer will be there when presented with a question so they go into this match they were extremely confident let me just they were extremely confident that no machine was going to beat them I told them it was just a matter I knew the the factors of a thousand I told him it was not a matter of if it was it might be a matter of when but a system is going to be human beings so we went up against these two gentlemen and of course we’re very successful the system that did this the system that beat those two very very bright men was our most powerful commercial we call power 7 system at the time that system was roughly the size of this entire stage and it consumed 85 thousand watts of electricity now those two human beings that your your brain consumes about 20 watts so incredible thing to me is that it took eighty five thousand watts to beat a cue for tee watt machine but that tells us something about we need to do bio-inspired computer science we need to understand how this does what it does and if we can do that and cut power consumption and and do the tort sorts of reasoning that we can do with with our brain we can do some very very interesting things this system also had the ability to learn it was primitive but it had machine learning if you go back and watch the game show you’ll see

that as the as the machine went down these question columns which were in categories the machine would become more accurate and that’s because it was learning the relationships within that category as it went down the column so it learned the context of the question it learned how its competitors were playing with in that column and it adopted within minutes as it went down that column what we’re doing now is taking that system and applying it to what we think is is probably one of the most challenging but perhaps the largest opportunity and that’s healthcare we all know that in healthcare system the number of errors is is just unacceptable we know that the cost is unacceptable and we think that this kind of technology will be a breakthrough technology in healthcare well point is one of the largest healthcare providers and insurers in in the United States we have a very unique partnership where we’re applying Watson technology to their massive massive historical data we’re basically using their historical data of treatments and outcomes to Train Watson to not compete with doctors but assist doctors in diagnosis and the pilots that we run with real-world cases with real-world doctors is astonishing what this system can do it’s astonishing from there we’re moving on in our research agenda to other forms of inputs not only voice voice activated or natural language but we need to deal with images the first thing the docs asked is for is let me feed in the radiological data that we have the MRIs or lab tests into the database that Watson was going to search on so that that’s a whole area of research but we think that this kind of capability is going to literally transform the healthcare system moving beyond just a primitive machine learning in Watson we’re doing research and again sort of bio-inspired a computation if one looks at neuroscience and one looks at the density of neurons and synapses big numbers out here 10 to the 12 10 to the 9th if we look at the number of nodes we have in our super computers we’re sitting this a blue jean p we can roughly roughly fit a cat’s brain into one of our super computers just matching number of neurons and synapses the exascale system that i showed you we will be able to mimic in our largest supercomputers the number of neurons and synapses in a human brain that’s the easy part we do not understand how this is water we do not understand fundamentally how the neurons and synapses are behaving all we know is that they’re not not ones and zeros switches they’re multi-state devices and they learn over time and their behavior changes based on what they have experienced and so using this again bio-inspired computer science we have a program underway in our nano electronics area to fundamentally start to build a raise of electronic synapse like devices so that we can physically mimic the human brain we’re starting with very very simple devices the first device we’ve built is actually with a traditional ones and zeros CMOS 45 nanometer device I can go back we packed all of 256 neurons not 10 to 12 but 256 neurons 256k equivalent synapses on a device we’ve built this chip and we’ve already begun to use this device to recognize images and train images the next step that we’re underway with is to replace the ones and zeros synapses with very very special materials that actually behave like neurons and synapses in that they’re multi-state devices and depending on what the device has seen or how it’s been triggered in the past it will have a different configuration than ones that have never seen or experienced that stimulus or that electric charge so between what we’re doing with computer modeling and what we’re doing with physical devices were fundamentally on a journey of these so-called learning systems so in summary as I stand back we’ve gone from the early tabulating machines through what

we have all experienced which is a computing error a programmable computing her some of this experienced it early with punch card systems we’ve all experienced it with pcs we’re now experiencing it with with mobile devices but really we’re still within the same regime of computing things or compete or or communicate the world that we’re entering in now is it is and we saw a glimpse of it in Watson in that competition with humans because we’re entering a regime now of what I would call smarter systems and these are systems that can deal with massive amounts of data data that’s moving at incredible rates data that’s ambiguous and machines that have the ability to learn and recognize patterns and literally become smarter as they have experiences verses are reprogrammed by human beings as crazy as this may sound I’m sure that the people who worked in the tabulating error never could have predicted that we would have systems that had XO flop capability that could model molecules and predict diseases so the research that we’re doing again is the completion of this era moving towards exascale and very very focused on this area of smarter systems and so as we establish our global labs we’re looking for the kinds of problems whether they’re natural disaster problems healthcare problems etc problems that these kinds of systems will be particularly good at solving where it’s no longer just calculating the answer to some arithmetic problem more quickly systems we’re learning and sensing in real time have tremendous impact on society thank you very much and I’d be happy to take any questions Dean economy from Nick to national ICT Australia on the on the chips that were simulating the neurons how are you going to start the patterns of connections I’m fascinated by that is it you’re just going to subject it and hope they connect some way or is it as they’re a starting point yeah we start and we start by just subjecting it through various detector arrays to patterns and we start to store those patterns on the device and then the device through a support computer begins to overlay new images that it sees and starts to do comparative imaging and so literally we’re and it’s weird a very simple stage I should have brought some of the images where we’re the chip is recognizing simple letters so you can show it portions of a letter and it will recognize that this is an a or this is a be based on the fact that is seen those shapes before so right now it’s a very simple comparative technology that we’re using but again it’s at very very low densities what we’re very interested in doing quickly is building chips out of these multi-state devices that we we know we can build indiscreetly but we’ve not yet built arrays of them because now we can get finer grain now we find our pixel grain but we can get multi-state within a within a pixel hi I’m Liz day from I teenies you mentioned that you’d be looking into the health care sector with the Melbourne Research Center have you had any problems internationally with getting access to healthcare information and how do you plan on getting secondary use access to healthcare data over here yeah it depends on the country were working in obviously in the US or Australia that we we need the data to be sanitized and anonymous if we’re going to deal with in other places around the world that is not the case so in China they’re very open to sharing health care data so we can do things they’re much more quickly and easily than we could do in the Western world so we have to we have to deal with the privacy issues in in each country Ian Blair the question I have is how are you going to do what the brain does when you combine smell and color and sound and images and and all of those together

that strikes me as being fairly difficult actually those are very well they’re they’re complex devices but they’re just analog inputs they’re just signal inputs and they’re actually fairly slow inputs from a computer speed standpoint so we would view is the analogy is in the natural system we want all sorts of sensors out there we want temperature pressure vibration etc that’s no no different than sort of human sensing and these systems will have no difficulty dealing with multiple inputs because they’re just analog signals what we what we don’t know yet is how do those things combined to produce a higher level of knowledge in the brain and so you know when we get these multi-state devices that’s one of the first things we want to start to study is the interaction and overlay of these different senses as discrete inputs absolutely no problem but we believe that there are interactions that we don’t understand some of that I didn’t get into it but one of the things that’s most interesting is is how is this wire and one of the schematics you saw up there we had we collected all of the discrete work that’s been published on monkey brains and how they’re how they’re wired and there’s some very interesting wiring patterns that certain things have long-term connectivity versus short-term connectivity that somehow relate to these different areas of the brain that specialize in different senses so there might be interconnect implications to how we’re going to have to interconnect these devices to deal with the multi sensory inputs whole new area of research here and my name is osofsky my area is philosophy it’s very well to say that they’re all analog input devices yet touching a soft toy is different to touching ones lovers body how would you hand this to the computer there is a fundamental difference here in surface texture or firm lesser I’ve one responds differently to one end to the other so you see it’s not just a straight analog input well again the sensors and the input is fairly straightforward what we again what we don’t understand and where the real interesting research is is how does the brain learn the difference between those two things and how does it begin to associate soft from firm with certain kinds of objects how do we know that I know this table is going to be firm right based on experience that is a completely new area there’s a lot of brain mapping going on has going on with with in terms of where do certain things reside in the brain but there’s no deep understanding of how those things are stored within synapses and neurons thank you John right you mentioned about sitting in the Dublin control center seeing the bus in the river the operator said that’s noise it’s a faulty GPS signal right I how does the operator know that it’s a folder GPS signal how does the system then start to learn that it really is faulty and not a real bus yeah it was it was a scary thought that the operator just sort of threw that data data out right to today they’re operating with very primitive optimization programs for traffic which do no snow scrubbing of the data whatsoever so they’re just statistically saying that that’s an outlier we think that through different analytics techniques though again looking at last location of that bus looking at other types of relevant information will be able to tell whether that’s an outlier or not in other areas where it’s say even more important know if it’s an outlier let’s say you’re talking about cyber security and you have one signal that says something is going on over here in the corner this database this this piece information move from here to here that normally doesn’t occur and so in this case we’re using analytics to say okay you know what’s the probability that someone interacted with that data have we ever seen that movement before who moved it did that person have the right now wherewithal and the authorization to move that piece of

information so in the area of cyber security we’re looking at other types of information to determine whether that’s an outlier whether it’s noise or whether it’s signal and there are cases through you know basically every industry of that kind of thing I use the bus because it’s a very visual thing but we’re going to have two more and more distinguish between signal and noise in these world situations and we’re exploring and doing research and all the analytics techniques to try to sort through through that noisy data in these real-world situations is I think the biggest problem we’re going to face we’ll build a computer systems that will be big enough and fast enough to deal with it but the noise is going to be something that we’re going to really be challenged to handle particularly if we want to tease low-level signal out of all that noise there’s a lot of analogies in the electrical engineering world for the electrical engineers in the world of techniques to do that and we’re going to apply those I think to this new domain of noisy data so the research needed to move towards a human brain is in one direction but to solve the Dublin traffic problem is quite different and humans are not very good at that is it a completely different direction of research or are they all part of the same ultimate system different but I think different different in time like the cognitive work is is very long-term research that we believe that that is where these smart systems are going but these systems are decades away true cognitive systems are decades away the real world problems were dealing with now the bus in the river we need to deal with now so we need new algorithms new techniques to sort through filter this data looking for signal versus noise today and you can go through every level of system problem that we’re dealing with and we’re seeing this same pattern in the earlier symposium i think it was coming up routinely that we’re getting you know very very noisy information we see it in the large telecommunications the carriers when they’re dealing with billions of records per day there’s tremendous noise in that that data set and as they’re trying to decide what to do with their network and how to build and how to change their marketing programs there’s huge amounts of noise in that data and so we’re trying to invent new algorithms to deal with that in real time with today’s systems but they’re all I think dealing with directionally large analytics problems and really the kinds of problems we’re talking about those smart systems are analytics type problems where you have massive amounts of information and you’re trying to sort through it so directionally the same with different totally different timescales bitter engine I’m very interested also in the concept of the biology of computing and mimicking the ryman and the chips that you have in which presume we will have considerable flexibility in them like the brand does what you’re doing i think is assuming you can expose those chips to experience they learn from that experience and if we become more powerful that’s true of the brain a a more experienced brain is a more powerful brain to a point at which stage it starts to H now I don’t know if there’s a relationship between experience and number of signals that come in and the fact that it ages itself but it’s a very simple question do you think there’s a chance that issue design in these machines they may infect themselves age that’s a that is a very good question I hadn’t thought that through I suppose is a there’s a chance that they will just as the human brain gets in will get into recognizing patterns in that pattern will become the dominant theory and will start to bias the system there are techniques to deal with that the the Watson machine is one of the fundamental decisions we made with the Watson machine which made it so powerful relative to other people who have attempted to do this is it uses very very few rules and very very few predetermined pieces of information it takes every question and statistically wipes the slate clean and goes out and searches and comes up with a statistical answer to whatever it’s asked and it will come up with three or four and say i’m ninety-five percent confidence this sixty percent this this and this starting with a completely new

calculation we did that because it was too easy for the system to become biased interestingly if you if you look at health care that is one of the phenomenon in healthcare that becomes a problem that the docs as they start to see the same the same disease chances are they’ll diagnose you just based on what they last saw right we think that Watson is going to have a huge advantage in healthcare because it will start you as a completely new statistical analysis it will not it will have a record of previous but it will wait that equivalent to every other new piece of information available so some of these statistical techniques we should be able to use in these new types of chips to make sure that it doesn’t over recognize and become trapped in its own learnings but that’s a very good question the same thing to hear from dr. Kelly to get I think a sense of how analytics and high-end computing now interact with a set of substantial policy and human problems and offer the prospect of very different ways of framing responses and changing our world so on behalf of the University of Melbourne can I say thank you very much you you