 Open the pod bay doors hell I'm sorry Dave. I'm afraid I can't do that. What's the problem? I Think you know what the problem is just as well as I do. What are you talking about? This mission is too important for me to allow you to jeopardize it. I Don't know what you're talking about hell. I know that you and Frank were planning to disconnect me And I'm afraid that's something I cannot allow to happen Where the hell did you get that idea hell? Hey Although you took very thorough precautions in the part against my hearing you I Could see your lips move We are not there yet This is science fiction We have in front of us to make challenge To resolve before to arrive to this kind of a solutions The first is thinking about how we are able to find The right mathematical functions to mimic the intelligent behavior Now we are thinking the things based on errors and trials We don't know what is the real way to Suffocate the way that we are executing that kind of artificial intelligence But also we have an a second challenge The second challenge is rated that in the during the last 20 years Owing the increase of the computational power and the availability of the amount of data a vast amount of data Machine learning algorithms had quite remarkable success in tax from in several tax from computer vision to Play games However to reach the point When the current computation on tools We are arriving to the point that the current computational tools are not longer be sufficient also To tailoring that we also we have tailoring architectures that could be the GPUs or could be the TPUs They the graphical processing units of the tanks tensor professors units They don't are Standard enough to be able to escalate so the conference today is about How we are working to move forward the future I An elisa martin I am the city of for IVN Spain. I am member of the Academy of Technology a Lot of my work is working with technology to introduce Disruption with the technology and thank you very much to join me because even in our young working in artificial intelligence I knew the human war to feel better and also to transmit and to have an adorable direction of communication of course that during the last year during the last decades We advance a lot We are in the middle of the journey and of course the majority of the people and for sure that during this conference you are seeing several things and Thousand of conference that the people is is really looking for the future and The people is looking for the artificial intelligence like opportunity Also, we are bouncing a lot To understand that when we talk about artificial intelligence, we need to talk also about information architecture Very similar words isn't it very similar capital letters in different positions But you can do you cannot do something You cannot do anything in artificial intelligence if you don't understand how you are going to manage the information Even though you are not going to do anything if you are not able to understand what data is required In your artificial intelligence goals Also, we know very well what are the steps to to secure We are able to collect data organize data analyze data trust And transparency in the in the things that we are doing and also how to infuse all of these information to automatically Automate some process Also, we understand that We need the data is in different places and we have different types of data So we need to think in also in a multi data architecture We cannot relate just in one shape of the information And at the end as companies as IBM and another companies was also realized That we need to use open technologies in this roadmap that we have in front We need to use Open technologies for the data science to build and train at the scale But also we need to thinking about for matching learning and for the people that build the algorithms How to be able to embed that algorithms in our process how we are able to operate and operate and Operationalize these algorithms and also how we are able to deploy and manage and monitoring that kind of algorithms but still There are problems that we are not able to resolve with the current with the current approach We have problems that you have in the bottom of this diagram We have problems that is in the end complexity That means that we are talking about linear problems That kind of problems is like when we are in the top of a building just to try to use another Referencing of the complexity The second the second level of complexity in the problems is when we are talking about the potential Functions some problems that could be represented by potential function That's could be when you try to organize and you try to organize you try to find the primate The primate numbers, okay the discomposition of the primate numbers so And the third level of complexity is the problem that you can express them in an exponential through exponential Cops or exponential functions this kind of We need to advance in order to resolve the two the two top lines Because if it's the first line we were in the in the top of the building When we are in the second line, we are in the level of the atmosphere the planet atmosphere But we like to go to the third level the complexity of the problem We are talking about problem in the level of the universe To this to respond this kind of challenge Historically, we are moving through you know to resolve Challenge itself. We are no moving in a in a let me say a systematic Abundance of the things we are moving Challenging people to do different kind of things Starting from the beginning the first demonstration to be able to train a machine learning network in order to To play in an automatic way a game it was in the 1989 Where this a colleague from IBM was able to was able to execute in an automatic in a ton of us way a Play to the check checkers Okay from that point Through the big blue Geopardyne and arriving to the alpha go game All of this is helping us to advance in the artificial intelligence, but sometimes things happen before for instance In this in this in this diagram in this picture, you have Gerald tesoro that is also coming from IBM and He is the father of the reinforcement learning Working and he's the first time that he was able to make it working automatically He did it With a game that is the badminton and he used also the pictures in order to training the neural network to make it happen Alpha go is an Sophisticated way to use the same kind of techniques several years after But also we are able now to do something else As IBM we have an a new challenge in front Geopardyne was in the in 2011 In 2012 as a company we decided to have an a new challenge and that was to design a system to be able to debate debate is a More complex problem is not only answer questions. It's not only to understand the context debate is Thinking about the ambiguities how you are going to manage the ambiguities and also How you are going to manage not only the understanding of the questions But also how you are going to build the answers Sorry, how you are going to build the arguments Okay, again the different themes the different themes of the different topics that you have in the debate You need to understand but also generate the Information I think the argument is going to be an a clear and a very good Entertainment for all the people here in Spain because we love argument So I suppose that we like to have something like this enough in with us in a house But taking taking care of the media Subsidizing space exploration is like investing in really good tires It may not be fun to spend the extra money But ultimately you know both you and everyone else on the road will be better off So you can see how the system generate the speech the system is able to Organize the information in order to support the first the first argument about the topic Afterwards the system is able to listen the person that also put on the table the arguments and with all of these Elements the system was able to generate a new speech Giving the reasons why he is supporting that position in the in the debate and after that he is able to create and Generate again a summary of the different content for the argument So we are talking about a different dimension about what are the capabilities of the systems when we talk about the evolution of the artificial intelligence And I suppose that everyone is very familiar with that really today. We are talking about Neural networks and inside of the Neural networks. We are talking about deep learning But the reason why we are we are now in this explosion of the deep learning Has a lot of relation with the capacity of the systems If we take a look of what happened in 2012 Why if we take if we see this this graph where we have the Accuracy of our Neural network in order to recognizing image You can see the real Optimization happen when we introduce in the in the execution of this Neural networks we introduced accelerators the GPUs hardware and The after the years after we are able to arrive to the point that the Optimization of the seclusion of the Neural networks are close and better of the human But still we are in the narrow what we call narrow artificial intelligence so Really we are talking about systems that are able to execute single tasks in a single domain That maybe we they have an a superhuman Accuracy a superhuman speed, but they still are not a very standardized Every time you need to define a specific system You need to design and execute a bill and execute a specific system in this specific domain It's true that we advance a lot we advance alone in the legman's translation for sure We advance a lot in the space Transcription in the language processing in the face recognition in the object detection but this is simple tasks simple element to be analyzed by a Neural network But also we need to afford to to affront another Challenge and we are doing well also to do that. So this the next challenge is about how we are able to resolve the with another Neural network to different tasks That means to recognize the image and at the same time to be able to write To the description about what the image is about Okay, I think if you take a look of the different image and if you take a look of the description It looks quite well Okay, so the description described very well quite well the image itself And if you go to the last one, maybe you know, I don't know if a person is able to do a better description that this one Okay, so it looks quite well that we are able to do these kind of things in a very good way So now it was if we have binds in the artificial intelligence We need to afford Multitask That means to be able to have these kind of algorithms able to Execute or able to manage different kind of parameters in a multi-domain in a very distributed way and information And with an a semi-autonomous in a semi-autonomous way and this is the The the time that we are starting now So the sample that you have and you are seeing in the video is exactly the best moment in the open In the open Golf Master, thank you. Thank you. This is the reason that I like to have people in front of me to help me so in the master This moment was selected by an artificial intelligence system And was selected not because someone telling them No, because you have an specific annotations on the videos because this is in real time The system select that because recognize different kinds of elements on the image They recognize sentimental when the people really You know love or loud and make a note of noise when the things happen Okay to recognize also The behavior of the different players and to recognize also the statistics everything together and With all of these parameters together The name of our network was able to define that this is the best Moment for the master and this is the moment that they are going to reproduce in the TV news So if we like to to move forward in the artificial intelligence to move to this new step We need to think about the following topics We need to think about as explain ability We need to thinking about not only to secure the models To thinking about how we are able to explain. What is the behavior of the models? We need to create the sense of trust in the models and the transparency About what we are executing Also, we need to thinking about about security. We need to avoid to to have another Networks Neural networks attacking the behavior of our algorithm. Okay, and Introduce errors in the behavior of the algorithm Also, we need to thinking about how we are able to train the system with a small data Because for the different use cases, we are not plenty of data some cases Maybe yes, but in another cases and if we are moving in the in the business in the Enterprise area, not all the all institutions has a Bass amount of data for all the use cases to be able to train the systems and the only way to Resolve this is to to be sure that you are going to be able to capture the knowledge that you have to store your knowledge and to use this knowledge as as analogous in order to be able to resolve the algorithms But also we need to thinking about the ethics If we need to trust in the artificial intelligence We need to be sure that the algorithms has an accurate behavior has an ethic behavior So everything that we are doing related with vias That we are doing in order to identify if if a data set has vias So it doesn't has vias and how we are able to correct a data set in order to correct The behavior of the algorithms is something that we need to put in place Joely with with the Specific or not specific with Joely with the platform that is an another Another Topic to move forward Platforms to manage the artificial intelligent life cycle all of them together We need to move forward in order to be able to manage the algorithms because previously We learn a lot about how we are able to manage the applications now We need to understand how we are going to manage the algorithms, but not from the point only from the point of view of the How we are designing that how we are executing that how we are training that but also how we monitoring that and also How we are taking care of these concepts of the ethics In order to do that in IBM we create one one new services in IBM cloud that is calling OpenScale but also we put in the Putting the open open Open source community we put fair freshness 360 in order to help the people with seven with seven with 70 Datasets and with ten algorithms in order to be able to use it to avoid This kind of vias and remember we like to build things like this Now the next steps is the general artificial intelligence and put attention about what IBM research put on the bottom of the of the of the of the slide They said that this is going to happen in 2050 That means that really we don't know what is going to happen because When we talk about General artificial intelligence, maybe we are talking something about very similar to what we saw What we see in the video. Okay, we are talking about cross domains learn some reasons synthesize synthesize new approach and broad autonomy What happened is when we when we see a videos like how We thinking that the artificial intelligence what we are able to do in artificial intelligence We are talking about the general artificial intelligence and really we are not there as I mentioned before We are crossing the boundary between narrow artificial intelligence to broad artificial intelligence The majority of the questions that I don't know you experience, but the majority of the questions that I receive They put the questions in the general artificial intelligence Maybe that could be a very philosophical Discussions, okay, but really we are not there And we are not there because we need to increase our understanding in different areas We need to increase our understanding To in in in the science and the computational areas We need to accelerate the research to understand the behavior of the brain to understand the behavior of the of the Intelligent in the people, but also we need to understand and to and to accelerate The computing models how we are going to support that kind of advance also, we need to advance in the To see and to recognize the environment around us. We need to have computational agents Continuously and progressively learning using feedback to in order to behave Taking in account the context where they are moving but also we need to advance to in the persistent knowledge based on the reasoning So if we are not able to advance in these different areas We are not going to arrive to the artificial intelligence and to advance on this. We need time So we focus one challenge and we are going to focus the second challenge We said that there are three type of problems Where with different kind of complexity what we need to resolve and In order to resolve the second and the third a kind of problems a We need to go to a different kind of computational models. We need to go to quantum computing To go to quantum computing. It's not we are not talking about only to have more power on the hardware We are talking about to really to have more power and more efficiency in the algorithms Why do we need quantum computers? Classical computers gave us the internet smartphones and even sent humans to the moon But try asking when to simulate a single molecule like caffeine to understand how it impacts our brains. It's impossible There's just too much information quantum computers may be ideally suited to it by encoding information into quantum states quantum computers May make calculations that we only dream of today So we could one day use a quantum computer to finally understand how caffeine's wake-up magic works IBM is already making quantum computing more accessible than before with open-source developer tools called kiskit and the IBM Q Experience a free way to experiment. It's accessible from virtually anywhere Millions of experiments have already been run on every continent even Antarctica Here's how it works You enter instructions on a classical computer which travel to a quantum computer hosted on the IBM cloud The instructions translate into microwave pulses with frequencies and shapes that control qubits and change their quantum states As the pulses travel over cables to reach the qubits They go from room temperature to negative 459 degrees Fahrenheit which makes outer space look warm this movement to the area of such cold temperature only takes about 0.0001 of a second after the microwave pulses interact with the qubits any results are returned back along the cables where they are Converted into data that classical computers can interpret Finally those results are sent back to you Each of these IBM Q experiments gets us a little closer to realizing quantum computing's awesome potential to solve a new class of science and business problems Learn about quantum computing or try it yourself share this video if you're excited about a future with quantum computing Okay, do you cite it? Yeah, go and say Okay, so we back to these curves how do the artificial intelligence behave There are a very few exponential problems in artificial intelligence the majority of them are non-linear Algebra related problems it means that I have an apollin nomenical shape Like could be the matrix in in version So What we like to do and this is the experimental area that we are working on is to thinking about how we can reproduce for a For instance using the the problem of classification How we how we can reproduce the neighborhood networks Behavior that we already knows how to do classification in the traditional of the classics computers Okay, how we are able to reproduce that with an a quantum circuit And in order to do that we are going to use one of the characteristics on the entanglement of the Sorry of the quantum physics that is entanglement Okay, these characteristics really is going to help in us in order to resolve this problem I am not quantum physics at all Okay, so but it looks like and if we go to the handset back and such some principle It said that in the quantum mechanic we cannot know both positions of Sorry, we cannot know at the same time the two characteristics of a particular That is the position and the speed if we know the position We are we don't know the speed and we we know the speed we don't know the position Okay so if we have two particulars and We are light and we entanglement these particulars And we put the two particulars in two different sides of the of the universe Okay We what we can observe is If we focus in one of the particulars and we decided to measure one of the aspects that could be the position of the speed What we realize that? Instantly we know the position of the speed of the other particular When we talk about that and I suppose that I don't know if you are familiar with that But the first time that someone told me about that. Okay. I really surprised Because this could happen only for two reasons First is because the communication across the two points in the universe is faster that the speed The speed light, but this is unlikely and the second one of the or maybe the other reason is because a Quantum description of the universe or the world is much more complex that everyday experience we have So really what is happening? We really we are in the second in the second option Because but it's all of you for sure you have the feel that this is not real Because we are be late one of the main concepts that we have that is the localization and The localization concept telling us if I am here and I am make I am take some actions over one object The reaction of the orchard are going to be immediate and close to me And what we are seeing with this characteristic Is that the same happen even if you have this particulate at the end of the universe? And this is something that we can know image, but it's real so Thinking about the two characteristics that we are using for our circuit in the in the In our in our quantum circuit, we are going to use the superposition That means that we are going to put at a particular in Superposition that means they are going to have different kind of the same But they are going to have one and two value one and zero value at the same time Okay, but also we are going to see to use the entanglement characteristics that means that when one of the particulars when we are able when we collapse one of the Particulates we are going to measure what is the state in the second particular so If we like to mimic in order to create our quantum circuit to classify to classify something, okay, we need with the current Systems we need to mimic the entanglement And to do that we need an expose. We need an exponential classic resources Exponential classic resources is very expensive Thinking about that if we have if we decided to to do the entanglement between two particulars Okay to represent that in a in a traditional computing We need 512 bits But if we like to represent 100 Cubits we need more that the atoms on the planet earth This is what it means is very expensive Okay, we are no able even if we need a circuit able to represent the if we need to represent at circuit we need 100 Cubits and we need to mimic it with the current systems. We are not able to do that So thinking about that every time that we like to represent One state, okay, we are talking about to the number of Bits the number of bits that we need to use it is to Exponential and Let me explain a little on this is noisy for me But let me explain a little how how the quantum algorithms work Okay thinking about the first that we are going to do in a quantum circuit is To Put the circuit to put the the qubit in superposition Superpositioning, okay That means that if we have two bits We are going to have four different states We are going to have zero zero zero one One zero one one at the same time Okay Thinking about that. This is a vector Where you have the face and you have the the the amplitude of the vector Okay, so the first that we are doing is to put the particulars the cupids in superpositioning now what we are going to do is to Load the data How we are looting the data? We are looting the that we are loading the data Changing the face of the vector. So we are changing The vectors in the sphere, okay to represent The different information that we are putting in place in the circuit Okay, and afterwards with the interference of the different phases Okay, we are going to have to solve the output So this is how the algorithms works in quantum This is what happened enough quantum System, this is IBM Q Okay Maybe it's not the traditional. It doesn't look like the traditional servers Okay, it looks beautiful at least has different kind of colors And as the video said you have this piece that is in the middle is where the cubits are the particulars are Okay, and it's the part of the system that is a close to the zero absolute the other one is a is Is the the different Claves that is I'm sorry. I forget the name what that Introduce or is able to to move the different Microwaves, okay, that is the the microwaves And make the introduce the chains in the particulars Well, come back to that problem So currently with quantum computing we are able to resolve two kind of problems. We are able to resolve Modeling nature or we are working in modeling nature quantum level, but also we are working in mathematical problems And in the mathematical problems, we are we are focusing in the matching learning Okay, so How we are going to exploit the entanglement for artificial intelligence I'm in the classic If we like to to do a classification Thinking about if we have different points or different that could be the linear could represent Different clusters. Okay, we have two clusters. We have the The the the blue the the blue clusters and the dark blue clusters Okay so if we like to separate if we like to To separate this car these two clusters with just one line is impossible Yes, but if we introduce the concept of dimension Different if we include you introduce different dimensions Okay, and we position in the information in different dimensions. We are able to separate that because we we introduce an Hyper hyper layer that is able hyper plan that is able to put in one side one Yeah Okay, sorry because I didn't see that. Okay, so so if we are able to Okay, so if we introduce different dimensions, we are able to to take a look of the we are able to Put in different places. We are able to find the way to separate the different clusters of information. Okay Okay, so The idea here is how we are going to create our function quantum Using that kind of approach. Okay So here we have the training set So we have black dots and we have yellow dots. Okay We like to train in a function that is the the quantum function that this quantum function is going to classify these these dots Okay To do these Implications what we are going to do in the function is exactly that we are going we are going to represent the the function is going to represent the different dimensions Okay, in order to be able to classify the the dots in the different areas So in order to do that and you can reproduce this this is in in the in quantum experience in the quantum experience tool The way to do that is to coding to coding the problem using kiss kit Okay, we can prepare the quantum That means that we are going to put the the the qubits in superposition Later on we are going to establish the using the internal demand between the different the different qubits We are going to map the information that to load the system and later on We are going to classify we are going to use an a quantum classifier algorithm in order to update the quantum network So now if we execute this what we discover is that as much as as much as we Establish a bigger number of entanglement between the different particulars As much as a curing is the classification Maybe at this point of time. It doesn't means that what we are doing We can say that it's better that what we are doing in the classic computing with the neural networks But currently we really discover something that when we we label to escalate With an enough number of cubits We can discover the way how to do the classification in a better with an a better Results in a better way with a better performance We are in the position of what we call quantum ready That means that we have tools in order to to create that kind of circuit This is what this is quantum experience and again the the quantum circuit that you see before you are able to have to Assept to the to the to it from here. Okay, I Hope some of you already tried to train Sorry try to use the system for your research So maybe just to to enjoy and to to play with it I don't know maybe you are one of the users or one of the simulations or one of the secutions That we already captured during the last four years Okay, at the at the end at the end of these Of these diagram, you are able to see that we have more than four millions of executions of exercise of trainings in the in the quantum currently In order to work with quantum the the the foundation of the quantum is kiskit Currently, but it's not only kiskit kiskit is there is well the the foundation. We are calling Terra Terra is the representation of the minimal elements that you know that you need To to be able to compute something in the quantum in quantum computing. Okay, it's open source It's completely able Completely open to anyone and what they are and they are managing the resources of the of the of the processor But also we have an another friends Joely with Terra with kiskit We have Aqua Aqua is the is the new framework Where we have different a set of algorithms. Okay To helping you to run all their algorithms and to run applications on top of kiskit And we are going to have and another two elements one of them is going to be her that is the The part of the framework that is going to helping with the simulators and also to do the different To do the the baggings of the system and so on and we are going to have Ignis in order to measure The errors. Okay, I'm monitoring the errors of the different of the circuits but also we have what we call the call the the Quantum network and is the way How you are able to access the resources currently? To access to five cubits and 60 cubits is completely open for everyone Okay Now if you like to go to 20 cubits So really you like to do something very serious and with another objective research or to do an experimental Experimental exercise, okay or for for business you it's better that you go to the 20 in order to do that we already Establish this network because the processors are in the cloud Okay, and to and to execute your algorithms. You need to go through the cloud And this is what it is IBM Q. This is what we have in the cloud This is in your town in in New York Okay, and currently in these machines We are running the kind of the exercise that you see before and that's it go Because I think it's important that if you are thinking too classical Maybe you are not going to be in the future. Thank you very much No question, I don't know it's too late because I Because I sat I move there It's a moment for the question any question that you have Yes, there is one guy. No, he's just walking just walking but no question. No question. No question No Victoria have one question. We are going Wait one second. We give you the microphone so we can hear you Can you explain if we are using real quantum computers physically or sometimes people think it's just simulators Okay, we are using real quantum computing processors always Always this is your choice What happened is we have in the Qs perils, okay? You have two kind of resources you see in the diagram You are able to use the real quantum's and you are able to use the simulators What is the people is doing? They are using the simulators in order to approach the first testing of the algorithms Okay, it's easier and also because there are what we see is that there are a lot of people that are doing the same Can they are running the same kind of algorithms? So sometimes you don't need to understand to run you you algorithm in order to understand the results because someone else are doing that Okay, so this is the reason that we are using the simulators However, it's very relevant to go to the real quantum processors Because if really you like to understand How you are going to manage in your algorithms the errors Okay, you need to go to the real quantum processors Okay, and we are running real quantum processors We have in the quantum network currently we have more more than 25 institutions Working with the 20 qubit servers Okay However, I think there are as you see in the in the in the scale in the calendar in the schedule that we have We are not in productions But we are not in production only because there are something we need to to work hard to manage the errors Okay, but also because we need to create this the application is take Okay, we need to reproduce all the software. This is a completely new way to compute algorithms So we need to create all the stake and we are doing with the community the community is doing that Okay, so we cannot go to production until to be sure that the community and the institutions are ready to afford to have that kind of resources and to make and to do something and to do something valuable for business With that Okay, but real quantum processors It's true that there are a lot of people that asking the same question Okay Any other question? three two One go big applause for Elisa