is the AI the oil of 21st century this is the topic for professor of technical Czech Technical University in Prague where he found it and has the artificial intelligence center and department of computer science a man who was named one of new Europe handed significant innovators in 2015 ladies and gentlemen make up a whole check thank you for coming 20 years ago give her a student of artificial intelligence at the University of Edinburgh and you were taught that there are two things that artificial intelligence will never be able to do what two things yeah that's a known story I was a student in Edinburgh like one of the best high schools in the world in time and they told me that you know one day computers will be better than chess players right the day the day came on a couple of years after then they told me but you know it will never happen that computer will be better in go then then human players and Korean players and Chinese players it will have to happen because God is too complex and it happened two years ago and they told me that you're both difficult driving a car is like real a complex problem and now it's it's soft it's done it's done did you believe if your teach respect then yeah you always tend to believe but also you tend to question you know in this part of the world students question students and they deliberate but the body a big shot is really true do you feel that your students question you enough these days they do because I encouraged them unlike others the floor is yours ladies and gentlemen I I believe that you and the audience usually follow what tech celebrities like Elan musk Mark Zuckerberg or Stephen Hawkings are telling about AI about this big general AI that it made it's very coming one day and nobody knows how dangerous what would be its potential whether it will conquer human mankind but it did it will eat us alive I don't know I disappoint you I don't know what is going to happen my next speaker maybe give you may give you more insight in this and this topic at the truth is I don't care that much and the reason why I don't care is that I care about what's now I care about current narrow vertical specialized industrial AI because I think that it's not only as interesting as the general AI but it is full of potential it can change the world and it's dangerous it's similar dangerous as the feature AI so I am glad that I was invited to come to you and speak at singularity because I will be sharing my opinions with you on what is going on currently and not what is going to come in the future I'm actually basing my hypothesis and my observation out of my long term academic career 25 years in at the university but also through building AI startups I am strong opinion that people need to really understand this is the only way how we can take the benefit out of AI and how how can we feel safe and it's important that general public and politicians sit together as technologists and they they understand I was impressed that Barack Obama in the last weeks of his presidency made his famous interview with with John I toe for the for the viral magazine and he was able really to explain to general public what a is and what is its potential he was drawing his conclusion for from his famous AI report that he procured his government to to to produce and to and to provide but it's only Barack Obama Justin Trudeau is also making Canada an AR country he has made a deliberate decision that for Canada oil not any longer source of their future wealth about its AI and they've built an economy where in Toronto and Montreal they Rai startups ai businesses big tech companies starting there there are dissenters it's it's not only u. s. and Canada but it's also Britain right so in 2013 one of the most advanced AI company deepmind was acquired by by Google for half a billion and it was pre revenue right if there is anybody BC pre-revenue pre-revenue half a billion right and I don't know this is a coincidence but last last year when British government were approving the budget they lost up half a billion to the budget oleh on AI research so Britain is taking guys seriously China no debate China in the future we will see that your China will be a super super super power in in artificial intelligence I wanted to give you a definition I just came from my university I was giving my lecture like couple of hours ago it was proof mathematics today you will see no mathematics no diagrams this is the only definition right so in my in my worlds in the way how understand AI it is a family of tech and its scientific field that allows computers to alternate accelerate and scale three things that people can do and people can perceive people can perceive data images videos sound people can perceive the environment people can also reason make a deductive reasoning underlies this causes and effects predict and also people make decisions people decide what to do and by these people are changing the environment what is interesting is that by perceiving reasoning and and and and reasoning people are making money people actually making lots of money and this is why and their oer is there to take a fairer share hey I want to take a fairer share why would be the only people would be taking the benefits of being able to perceive reason and take decisions and be paid for this I wants to have the same opportunity and a is actually thinking like really big steps to compete and you know if you look around you see that there whenever people are imprecise slow inefficient the eye comes and takes the job right so I think that there is this is what narrow a is narrow AI is sweeping human inefficiencies by alternating accelerating and scaling what people used to do there are couple of a is right so as I have explained in my introduction I'm passionate about narrow narrow AI this vertical specialized specialized AI this is the capability of human machines to take one aspect of human reasoning human human intelligent capability and to be able to perfect it replicate it automated scale it right for example you can speak about things like understanding images video sequences playing chess making financial predictions all those are specializing in capabilities and even every computer is capable of doing the same thing we call him it's it's an narrow AI system but the truth is I think there is nothing like a narrow AI these days because everything is extended ai ai ai which is better than human in this particular aspect of human reasoning that's why we are alternating accelerating and scaling there is an opportunity for taking all those narrow or extend a is aggregate them across the entire space of human capability and to create an artificial big artificial reasoning machine and if we can do this aggregation we are arriving at what we call a general AI and Maricruz develop speaker will be speaking about this AI is getting better and better every day and there is no reason why I should stop getting better so whenever we have general AI replicas of humans we just you know next second we'll have a super AIII which is better than people a hundred years ago we saw this interesting event in our history right we thought that you we will continue to have horses and slaves and neurosis and slaves will be the the engine of a human civilization but there was a fossil fuels we were able to find out that there is much cheaper and much more efficient and much more could be in the way how to create an energy storage and transport it and once we put the energy problem out of the question we as a human mankind were able to focus on new stuff we were able to innovate we were able to bring new businesses create new businesses not only for getting oil out of the out of the ground but also how to transport it how to get oil from a place to a place and how to create a new businesses that depended on oil we were able those hundred years ago to take the advantage of the external source of energy so currently we are the situations where we are taking the advantage of the external source of intelligence it's similar a is the similar way as I was helping as oil was holding a hundred years ago is making Alive's is a more efficient more convenient you know we can we can do transport but then we did before you know we can do shopping more you know in a more convenient way that we did before we actually cut friendships in zero way that we did before your AI is making our life easier and you know not no longer now you not long time ago in front of us will be driven by by cars without the driver there is lots of businesses that is being created by AI there are a I companies that create this core AI engines there are there is the AI enabled economy businesses that rely on AI there is also of wealth being created I I similarly to to how the world was created by oil hundred years ago what's interesting is that you in the past oil divided people all divided countries there are countries rich on oil countries that were able to make benefits out of being a trans transporters of oil there were countries that were able to make benefits out of being a consumers of oil I actually think that you're a I will divide nations in a similar way I'm not I'm not the only one who thinks so even like a highly ranked politician thing that you know capability of understanding and using and developing AI will be dividing nations there'll be nations with good capabilities and there'll be the nations who will be dependent as you know in every chunk of this kind I shall give you some numbers so what can I looked hard to give you some market data so in this data I came up with are the data from IDC market research company and they are here to support my argument that there is no AI when they're coming you know I'm old enough to be able to live through cup of a I've interest therefore the the times where AI research wasn't as successful boson as funded all companies were not that interested so like 80s late 90s early 2000s there are those pipes and even today when I speak in you know in the corner with people you are asking me you know this this AI is it's kind of everywhere isn't that fake isn't this trend that will go away and I'm telling you it actually will not go away when I started couldn't've security the cybersecurity business 10 years ago people are asking you know this cyber isn't this weaith its isn't that piece of fashion so you know cyber is a problem that will stay for a long time and similar with ai ai is eating people's jobs and currently there is 88 trillion worth of wealth created by people each year on the planet it looks like your people predict that within two decades we'll be replacing half of people jobs so divided and this is you know this is where you see what is the ultimate value where AI can depend I'm not I'm not trying to do AI will be dipping in all those for trillion US dollars but this estimate of 46 billion by 2020 I think this is reasonable and doable so that's why I think that there is no AI winter coming so just to give you an idea about how much money there is in AI these days so it's like thirty billion dollars in 2016 spent on AI out of which 90% was in the internal R&D we have you know we are seeing 12 billion dollars in 2016 out of the VCS and PE money 60% goes from machine learning distribution that 66% is us 17% china and 17% rest of the world this is dated data now it's more in favor of China so I just wanted to kind of give you like early heads at about what what actually AI is right so this this narrow AI is very successful if industries like automotive driverless cars your day will replace these engines sooner than than we think you I think that you may be in the next term of Angela Merkel we will see will we see this this Swope cybersecurity is driven driven by AI know you're the most part of the cyber defense these days is run run by AI we see AI in in banks and I just wanted to tell you that one of the most successful AI applications that was able to saw two samples of money were a film industry in the past because people didn't need to use so many actors for massive scenes and they were saving lots of money in film through through AI simulations so AI is three things it's an hour machine learning data analytics this high performance super great AI these days which is all around us data videos sound images then there is automated reasoning intelligent planning optimization which is this important o field of AI which is there from modeling and building a complex intelligent reasoning that leads to running robots that leads to for example a human workforce automation that is also very important for a playing algorithms in chess and go and poker and the last piece is decentralized ai ai which runs interaction between intelligent agents software and human AI which runs combinatorial auctions and distributed distributed systems this is the powerhouse of the internet the way how we are selling and buying advertiser manages the biggest source of income on the internet comes from decentralized AI very very important so those three fields are the important disciplines of AI research and also ai ai applications I wanted to give you something concrete something specific I wanted to show you what are the success stories of 2017 so that you can see what actually happened and what was successful last year in artificial intelligence and this is my present opinion so the first one is deep stack who has heard of this deep stack have you guys heard of deep stack yes okay a couple of hands as I was asked about chess and gone alphago this is his famous Google invented algorithm that concurs best players in golf last year human mankind was defeated by computer in poker right so there is a computer poker algorithm called deep deep stack that is better than the best human players and it has actually checked trace because my team Wardak Sheik Martin Schmidt and Belinelli see our three three check researchers that were the first three authors of this famous paper and of this famous invention following the deep stack success story deep mine came over to Alberta and I have built a deep mind lab out of out of this out of this group so the second important source of value created by AI last year was reported by Amazon Amazon acquired Kiva robots Kiva robots is a robotic company that became part of Amazon in 2016 those orange guys right at they deploy those in the river house they were able to reduce the click to ship time from 75 minutes to 15 minutes only huge huge savings as a result of which they were able to cost the amount of required inventory space by 50% they were able to reduce the running costs for inventory by 20% as a result of which they saved 310 million u.
s. and 2017 this is 40 percent of what they have paid for cavora boats it was almost 800 million u. s.
so it's Amazon Newton on a new trauma is interesting start up it was built by an ally of Rizzoli who is an MIT professor currently I think is he's now ETH but he used to be MIT professor and they've built this autonomous car startup like really successful startup that was the first start of it and that in 2016 built a autonomous taxi operation no in the world it was in Singapore it was a pilot it was like a piloting service like small part of like suburban environment in in Singapore and in the second quarter 2018 they plan to go commercial so they'll be running a commercial fleet of taxis in Singapore in between between 2016 and 18 and 2017 they became part of Delphi Delphi acquired this group of researchers for almost half a billion u. s. another success story that yo I saw an I liked was improbable improbable worlds it's like the second biggest a I started out of London out of UK and these are the guys that are building large-scale high-fidelity simulation of urban behavior right you know in London one is complex you in you a nobody guys travelled by a London Underground recently this is just mess right so you cannot get in the underground you cannot get on the you're on the sidewalk you cannot get texting you know the transportation in London is just mess and nobody knows how to run it how to manage it how to chain because there are no data and into the world's our ability Inc large-scale AI simulations that are capable of building of of building systems that give you the answers about situations that have never occurred those guys know what happens if a Waterloo Bridge goes bust for example right they tell you what happens if there is a bomb stumble on tráfico square and they can plan the evacuation scenarios and softbank's came last year and they invested half a billion us to support R&D are in the arm of improbable so major major major investment into into this company and the last result that I like very much is result reported by Netflix and Netflix reported that they were able to earn may create the wealth of 1 billion u.
s. only by improving their machine learning algorithms for for recommendation your recommendation algorithms it's like it has been here for years there is no rocket science it's commodity right and this commodity by getting better at at Netflix they were able to reduce the search time for for the right movie for the subscribers and this study says that you know if the subscribers are unable to find a movie with the 90s 90 seconds they can solve the subscribe the subscription and the reduction of the search time saved 10% of their subscribers which equals 1 billion of the revenue so these were the four interesting fantastic results out of the last year just gave you this this chart of those different companies in AI that have been here last year like vision all tokenization balls but all sorts of care Corre I companies but you know I I don't wanna spend more time there I'll speak about deals there has been couple of deals that happen last year in the AI space and I tell you what nothing really major I think that the markets are are waiting Google acquired Lego which is machine learning crowdsourcing but platform nothing major they did a couple of aqua hires in a dimension learning space Facebook acquired Oslo which is like a cereal like an LP bought company Microsoft acquired Malou Bob which is a Canadian and LLP and a reinforcement learning startup at all they did interesting acquisitions they both will face which is image tradition company but also that is which is a data company those are the guys that are creating like good quality data for training and algorithms out of unstructured data we're interesting Amazon did couple of good deals I liked body body labs which is image recognition company that is helping Amazon to give the right sizes of people based on videos and images and then a couple of small admirers Spotify acquired an island for capability to 33 music sprang dead signal sense which is a machine Kappa higher similar to Cisco who also did an aqua higher in this in the space of machine learning people are asking what are the next big things for 2018 and again disclaimer this is my personal view right this is it's my opinion I think that you know the world of AI will be asking for learning not only models hypothesis results but learning actionable intelligence people needs what Bradford Cross out of the CDC is calling full stack AI so that you get data out of data you get a program out of the data you get a trajectory out of the data you get a logistic schedule right so people need to get like fully-fledged or resolve that they can plug into their businesses right so this call for putting together machine learning together with optimization and planning and automated reasoning this will be next trend next year similarly explained ability the fact that a machine learning deep networks are not explainable that we don't know how they come up conclusion it's it's big obstacle oh it's my personal surprise that you know our cars will be driven by non explainable ai ai ai that doesn't explain itself however for using the whole potential of AI for example in healthcare which is extremely inefficient lots of money lots of financial waste so I'm sure that AI will come and sweep this all space but it needs to resolve in explained ability problem first so I assume that like lots of results will happen in explained ability in 2018 advance it all machine learning ai is under attack people are trying to attack AI from outside of the world people are trying to deceive manipulate evade AI algorithms and not only in cyber but also in e-commerce and will be in cars will be in healthcare and raise lots of requirement to both such AI that is robust against attackers so there will be more and more research and more results applied from this field of adversarial national learning I've already put it on the improbable scenario that machine learning is dependent on data and even now we are seeing lots of scenarios where machine learning doesn't have the data that it needs quality and scale and size for example you would like to mention learn the proper infrastructure for autonomous traffic in like in major city how would you do this because you know you don't have data you don't know how people are using autonomous cars you need to do AI simulations so and people will be simulating quite a bit through AI to get the good quality data that can be used for training machine learning algorithms and the last one is data markets people still and our businesses and the way how we live we still don't understand exact financial value of data that we have that we provide that we collect that we exchange as a reason why there are no markets people are not buying selling data so if you give away some data you actually don't know how much of financial value are giving to to somebody right so I assume that there will be lots of back-end technology created this year to be able to measure understand and an invalid data as a result of which data markets will be will be created I would conclude my my my talk with four points where I'm worried about narrow political specialized AI point number one is monopolies right so we all know that you know a is actually manipulating us we're a subject of manipulation and you know irrespective whether it's the news feed that we read or whether it's the stuff that we buy online or whether it's where we go go for holiday your choices are heavily influenced by by AI this is perfect I'm calling for lots of open research that the AI research is as open as possible a big tech companies pushing AI open source I'm so glad for Google that they were able to outsource tensorflow open source to the research community that lots of startups and businesses are using the value that Google created for 1 billion US and I think that monopoly and manipulation in AI is something that we need to be aware of and we would be debating in the future next security and privacy definitely our dependence on AI makes us vulnerable so whatever somebody is capable of doing the right set of attacks the manipulation will bite us in our back so security is a big problem it continues to be a big problem and similarly with privacy I assume that countries will have a different level of AI capability based on what kind data they have so and look around which citizens of which country are giving out data about wherever they go and those countries will have the best possible data for training the best possible algorithms unlike GDP our countries they're worth less data and training class sophisticated algorithms I'm not Clinic any opinion I need to tell you that privacy is a function of the quality of AI the last but one is regulation and oversight so AI no matter what but AI will be regulated we need to be able to understand what kind of AI is driving our cars would a I is getting the prescriptions when I go and buy medicine what kind of AI is investing my my money and I believe that the human society will be there trying to understand and regulate the use of AI and the last last piece which is dear to my heart is jobs right note that there is no jobs these days there is still plenty of jobs but in 20 years from now people period that half of the jobs will will go and you know we all know that human mankind needs two things safety security and jobs and if you're unable to restructure society so that we will be able to operate in the world where Jews will be different will be less jobs people break glass there'll be much less jobs that will be creating direct economical value I don't know how the society will function I'll end up with the results of the Barack Obama procured ai ai report and this report says interesting observation that 47 percent of jobs will go in the u. s.