 Terima kasih, Justin. Ini panel Artificial Intelligence. Kita semua adalah bahasa natural di sini. Jadi saya tak tahu di mana panel Artificial Intelligence. Adakah anda dah jumpa? Saya tak tahu di mana mereka ada. Baiklah. Jadi, apa yang saya suka buat... adalah untuk menghubungi ini ke dalam hubungan. Bukan kita kata-kata banyak-banyak. Ia akan bercakap dengan anda. Dan saya suka mendengar dari anda juga. Jadi anda mendengar cukup idea dan cadangan dan masalah. Dan projek dan sebagainya. Jadi, pada awal saya... menunjukkan beberapa perkara. Satu perkara yang kita ada... adalah cuba faham apa yang bermakna... untuk memperkenalkan. Ada sebuah definisi atau apa yang bermakna. Berapa banyak... ada kita... atau ada kita nampak Google Brain? Itu pilihan anda. Berapa banyak... yang anda fikir ada di Google Brain? Jadi, apa yang saya suka buat... adalah untuk menghubungi itu dan cuba melihat... beberapa jawapan... dari perspektif apa yang ialah intelligence? Apa yang anda fikir? Dan bagaimana ia berkaitan dengan apa yang kita nampak? Dan kita ambil hubungan daripada itu. Bagaimana lagi? Ada sebuah mic yang lain di sini? Jadi, apabila anda bersedia, anda akan berjalan-jalan. Bagaimana anda akan berjalan-jalan dengan Oprah Winfrey... dan bertanya untuk jawapan dan pertanyaan? Jadi, saya akan mulakan dengan Andrew di sini. Jadi, saya rasa... teknologi telah berjalan-jalan... dan terus berjalan-jalan. Dan perkara yang penting yang kita dapat... adalah kita dapat... lebih baik pengalaman untuk hidup kita. Tapi jika anda melihat bagaimana ini berlaku... dalam perjalanan, anda melihat peralatan. Dan peralatan ini adalah... lebih-lebih-lebih mempunyai. Dan banyak peralatan yang kita selalu menggunakan... dan kita terus menggunakan... sangat magical sebelumnya. Apabila saya belajar komputer sains, sebelum saya mengambil peralatan... saya fikir peralatan itu... mempunyai aplikasi yang sangat hebat. Bagaimana kita mengambil peralatan ini... dan memasukkan... kode asam. Jadi, saya tidak perlu mempunyai... peralatan atau apa-apa yang saya menggunakan. Jadi, saya rasa... dalam peralatan itu... peralatan yang sangat hebat... tetapi dalam peralatan itu... mereka mengawal. Saya rasa lebih anda melihat dan lebih anda bekerja... dalam peralatan ini... lebih magical ia menjadi dan lebih... ia mempunyai... peralatan yang lebih hebat... dan lebih mempunyai. Terima kasih kerana mempunyai kita. Saya mahu mengambil peralatan... quite some interesting talks today. To take a little overview... what we're seeing is that Google is... really active and doing a great job... in doing AI also open source. We see that Microsoft is moving... into that direction. And we are working with a lot of... corporate customers worldwide. And we see that all these companies are also... starting to technologize their value chain... and base them more and more on data. And now become more and more active... and seeing that they will not be capable... in doing that by themselves. So they move towards open source environments... also in the eye space. And so the tech companies are already way ahead of course. And now the old industry is trying to... to pick up or do something about that. And that is also because of... the algorithms and all that... what the tech companies are providing... making it so easy for them to use. So I was wondering if you perceive that... as an opportunity that is coming up... if you perceive that as a threat... or as a large opportunity... for the open source movement also in the AI... that all these corporate money... is moving into the space. And I think that will be very strongly happening... in the next years. I think that's interesting. The question of corporate money... going into AI stuff... Is that a good thing or a bad thing? Bunny. Corporate money getting into AI. Do you think that's a good thing or a bad thing? I don't think... I think that... there's plenty of... I mean it's a more general question. Who should be funding research... or whatever, these sorts of things. I think that it's perfectly fine... for corporate money to be getting into AI... or whatever they want to get into or government money. It's more about... the applications and motivations... that are behind... those particular... why are they funding it... and what they're trying to do with it... at the end of the day. So from that perspective... would there be... Michael, your Suzy tool... do you think there are aspects... about the tool that you have... that could exhibit... the notion of... like the question I asked... intelligence? So again, it goes back to... what would the definition be? Would it be... what... if I see something... do I know it is intelligent? I don't know there's a Turing test. But is that sufficient? It's like someone was asking... can you define what is porn? It's like when you see it, you know. But how do you define it? Artificial intelligence... comes in different flavours... and it's difficult to find the right definition. A good definition is maybe... one that's very simple. For example, artificial intelligence is... if a machine can do something... that a human can do. So humans can do a lot of... intelligent things... but also a lot of stupid things... and useful and not useful things. In the past, for example... artificial intelligence was considered... in the field of gameplay... like being a very good chess player... was considered in the 80s, I think. People said machines... they have that much inspiration... like a very good chess player... and it will take a long time... until they keep up and play a very good chess. But now this game is over. Chess playing is an easy thing. This is kind of a solved problem. But now people say, oh this was never a problem before... because it's just an easy program... you have to write and then you are... as intelligent as a human in this field. So human intelligence comes at places... where you don't expect that you need... so much intelligence like... if it's about driving a car. This was for a long time an unsolved problem... and now in the next years... we will see that self-driving cars... come there and they get a license... if the number of accidents... is one tenth of what a human would do. So artificial intelligence... comes in different places... where it's not only about... doing a conversation like... chatting with each other... and expressing emotions and so on... but you would expect... if you think about a touring test. So we have a lot of... commercial applications like... for example do financial consultancy. In the past it's considered that... you need an expert to tell you... where you should invest your money. And today you can just put in some numbers... on the web page and if you go to a bank... they do nothing else. They don't do actual consultancy... they just read what's on their page. So you would say this is artificial intelligence... but still it takes a job... of a human before. So I think the number of places... where technology replaces what humans did... will increase... very fast, I think. And you wouldn't say... this is artificial intelligence... but in the sum of it... I think we see that... there is machine intelligence coming... and replacing humans... but mostly for good reason... because they are more safe... or a couple more information. Yeah, I think you can debate this... for a long time... but one of the most important things... about machine learning is it's been useful... and we judge things... whether they're useful or not. So you look at the chess example... is it useful for computers to be really good at chess? Well, only so far as you can use it... as a trainer to become good at chess... to enhance your enjoyment. I mean, in Google we have... Gmail and it has a smart reply. And so you can say... and I can say... hmm, here are some replies... that the machine learning has given me. And if I provide them... and if they're good, I can use them. If not, I don't have to use them... but my life has been shortened... the amount of typing that I have to do... because I don't have to think as hard... about how I'm going to respond to my email. And I think that's what I find exciting... about all the open source... is that people can take technology... and all these free toolkits... and plug them into... whatever their software... and make it better. And I think... we've only hit the tip of the iceberg on that. And I totally agree... and I think you have two important points... and that matches with yours. I think... the AI or machine learning capabilities... pop up all over the place... and sometimes we expect it to pop up... like for the cars sometimes... we don't expect it to pop up in that space... but what is happening... and we see that very strongly... is that all of this will... as good as they are, it will come... it will eliminate a lot of jobs... a lot of white collar jobs... a lot of traditional jobs... and we see in most corporates... and also governments, we see not enough... actually... time and also power invested... in thinking about what is going to happen... to the people that are probably... going to lose their jobs in the next few years. And there's a lot of transformation projects... going on... and I think there's a lot of good in AI... and it's very important that it comes... and it will come and open source will actually speed... or it's speeding that up a lot... but we had it a couple of times today... the norms and values around it... and what do we do with society... is still something that is like... widely not really discussed in depth, I think. So if I were to take that statement slightly further... if you say... to be able to create some stuff, right? So... is creativity... curiosity... a qualifier... for display of intelligence? So, can... potentially... an AI, whatever... create the Novena Project? I mean, I think there's... I don't know... I'm not... I'm one of the kids that don't belong here... but... Presumably, there's a difference between expertise... and... like creation and innovation, right? So what I see... people, a lot of what's coming on now... are essentially experts... AI experts in driving, AI experts in... reading your email, AI experts in... advising on things. So they're replacing select human... positions because of their expertise... that gets rid of a lot of training... and otherwise things that you would need to do. But... what I expect a self-driving car... to be able to come up with... a new sous vide recipe... for my salmon belly, no. It's not something that... I would ever expect to come out of... like an intelligent car, right? And so... at what point do the systems... become... sufficiently... I guess self-aware... to write their own rules and say, hey, like... I can reason about... where cars are. Therefore, I should be able to reason about... logistics and help plan... the routing of packages and UPS. And a car says, you know, actually... you're solving the wrong problem... with the traffic flow. Actually, the problem you should be solving is... don't send your packages from the depot... to here, just do it directly from there... because the car realized that, right? Somehow, because the intelligence... at this point, you know, systems... can scale up to become intelligent... in that fashion. But I think there's definitely... I think today, what we're talking about... replacing jobs is a lot of expertise... being captured in computers. But I don't know if I would qualify... as intelligence in a human sense. I would agree with that. And going back to designing a board... if you're placing... decoupling capacitors, that's something... that a computer could automate. Or if you're... if you had not a router that actually worked... you wouldn't have to place all the traces by hand. And so... if you had that, would you be... out of a job? I think you would just make more boards. And this is kind of what's happened. Whenever you have an assistive technology... that makes your life easier, you just do... you have more productivity. And musicians had this when the MIDI came out... in the 80s. They could start doing their own arrangements. Playing with instruments themselves... and doing their own orchestrations... and seeing what it was going to be like. And they had things like... auto arpeggiation... and auto rhythms. And these didn't make their creativity any less. Now when we have... at Google there's an open source project... about Magenta, which is kind of exploring... this whole issue. I encourage you to check it out. But what's cool about it is that... you're still directing where it's going to go. I think it's because of something that's Bach-like. And it's that, oh, I know how to play something that's like that... because it's been trained on that. That doesn't mean that it's taking over creativity... necessarily. So I think these kind of assistive technologies... are what we're really seeing right now. And we're going to continue to see a lot more of them. And we're driving them very well. So the issue is... the question of assisting you... rather than... and assisting you... because you have expertise in that space. So I know... I have these four bars of Bach... and I play that... and the system says, oh, I know what this is... and it continues and then innovates on that. Is that a display of... intelligence? Or is it a display of expertise? So in other words, the way I'm seeing it... is the intelligence in... the artificial intelligence perspective... is very localized. So if I... if you say that... the computer beat... the human in chess... can that same computer... after the chess game is over... walk away... or go drink a beer... or do anything like that... which the chess player can do. But the computer is not going to do it. So... is that an element... that we need to look at? Do you have multifaceted levels of intelligence... that you need to... in order to make it... more holistic? I actually... You can respond there first... because that's going to take us off topic. There was the idea around... that if you want to make a machine human like... and to give it intelligence... that it must be... it must learn like a human... and grow up... and be a baby first... and learn language and so on... to have all these capabilities... to be like a human. But I don't think this holds... and I don't think... the comparison... is valid that you say... if you are intelligent... after a chess game... you can do also something else. If the point comes... that a machine is... more intelligent than human... then they will... do the same... and do a comparison... after human capabilities... if they are able to do what they are doing... because they are able to... compute so many numbers... and access so many databases... and it's also not valid. So I don't think this is the right... way to... view... about artificial intelligence... maybe. I was actually just... going back to the original thing... talking about assistive... sort of... until... the AI is becoming assistive. One thing you said that... I wasn't aware of that was really alarming... was when you said that Google Gmail... gives you responses... that you can give. I don't use Gmail... but when you said that I was like... oh my god, that's like... crazy. I use SwiftKey and it has... the little Markov learning thing... and sometimes I do the thing... where I start typing a couple of words... and just hit the complete button... to see what it's learned about me... and what it would say. And sometimes responses are pretty good. I'll just send that. Oh, that's kind of fun. It's not what I intended... but I'll send it just to troll the other person. But so... the interesting part though... is that... the thing that's scary to me... is when you can do. If someone's giving me an assistive technology... that will help me place decoupling capacitors... on a PCB... but I didn't actually understand... why I placed them where I do. They just seem to look right. People will start designing boards... where actually capacitors might not go where they do... and they'll never have to learn... why the capacitors go where they go... because they've had the assistive technology... and eventually that computer... exceeds the... they're in the right spot, right? And so at the end of the day... there needs to be like a human... sort of intervening... and filtering the stuff... that's coming out of assistive technology. And what is a little scary... is that... inbox responses are very suggestive. Like one thing that I... I don't like about SwiftKey... is that I might want to respond one way to a person... because that's what I feel. But then I'll see the suggestion... of what I would have said in another context... and that actually changes... how I feel about that topic. Like SwiftKey is now driving me in a way. I really don't like it... when I feel like I've lost my... sort of... my internal self-compass... to that little suggestive thing... in the keyboard, right? And having an inbox reply... that is suggesting to what a set of responses are... makes you a little lazier... you're talking about how to vote. And someone says, hey, like, you know, are you going to vote this way? And it has like a set of responses... that you could go for. And whoever wrote the algorithm decides... oh, we're just going to bias it one way or the other... and so that you're going to always respond... in favor of one view or another. That response... because you're coming up with in your head... how you should respond to that. That AI is actually just whispering in your ear... ideas almost. It's like the little devil sitting in your shoulder... telling you these things. It seems like... that's really scary. Like when you start having suggested responses... and tweets and stuff like that. So... so basically... by reading the email, you're also getting that suggestion. So it's not really unique to the... so if you look at your inbox, you've read these email from people... and they're creating suggestions and you... like this is the idea of persuasion... of an email. So this is happening from spam. But... I mean... I wasn't responsible for creating this technology... but this is one of the things that Google... looks at when they implement it is... how to balance these trade-offs... and how to make sure that... privacy is maintained as well. So the key difference is that... for someone to persuade me... another human must make the effort... as little as it is to forward it to me. Or they have to be a script... that's mail-botting or whatever it is. The difference is that... when there's a single centralized AI... whoever controls the parameters on the AI... can actually... they don't even have to know the people... or the conversation or whatever it is. They just load the parameter set... to sway the responses in a particular way. They're completely non-contextual... influence and biasing... of the conversations... of every single conversation that's on Gmail. So the amount of power that's embedded... into that kind of AI system... and how it's programmed... and how the weights of the network are set... can have systemic effects. And when you're looking at elections... they're decided by 1%, 2% of a population vote. You all of a sudden realize... this may be the most powerful lever... to turn in the world. So the algorithms that we use... are based on... research that we publish... and open-source algorithms that we have. And these are constantly evaluated... and bias in the algorithms is a research topic... and I'm not an expert on it, but it's definitely something that... is aware of. Let me take the opportunity to... open it to the floor. Yes, someone at the back there. The mic, where's the mic? Oh, you have it. Go ahead. Yeah, I have two questions. One is, how is AI going to affect... skilled human jobs? The other one is... what is your opinion on... how real is the risk of... skynet-like event? Alright, skynet. I was waiting to hear skynet. First time I'm hearing it today. So, question. Is AI going to replace jobs? Is that the first one? And the second one is... when is skynet happening? I hope you know what skynet is. I would like to answer with a quote... of the former Chief Technology Officer of IBM... Gunter Dück in Germany. And he said that... all jobs will be taken over by machines... instead of one, information technology jobs. Because of all the abilities... which are there. Even the job of a doctor... and everything else that you can think of... can be replaced by artificial intelligence... big data and moving machines. But only one is left... that of information technology. Yeah, and it's also not about being scared about. I think this major shift, like you say, will happen. And what I was stating before is that... corporate and the government... also think about how to keep up with the speed of a shift. I mean, if we look at cars... if we look at also decision-making processes in companies... they will be highly taken over by... more data-driven systems, I would say. And also by AI systems. And this is nothing to be scared of... because it will happen and it's good it happened... due to the efficiency reasons... and also to more productivity and all that stuff. But I think it will happen much faster... than the examples you mentioned before. It happens much quicker. And I think that's just important... that we also think about... how can we actually support these people... and moving quickly forward into new... productive jobs then. So does it mean that... geeks will finally inherit the world? That's what it sounds to me like, right? I would like it here, yes. So we have job security, guys. So we are okay. So it's... the others who have a problem. Did that kind of answer your question for you? Alright, another question. Is there somebody else? Yeah. Oh, there you are. We were talking about touring tests... and I think that it's not very valid anymore. It should be more like physical tests... that, for example, can we think a very simple example... that I could ask a robot to tie my shoes? How long... simple question, how long it will take that I can... let's say so for a robot... that this is how you tie my shoes... and it would tie them. It seems for us very simple problem... but it's actually very complex... because it's like in three-dimensional movement... you curve the shoelaces... so it seems very complex. Which one is more difficult? Driving a car or tying my shoes for the robot? Tying a shoelace, right? Which one is more difficult? Tying a shoelace or driving a car? Which one is more difficult for a robot? Is that your question? Well, yes. There's a difference between robotics and AI, though. So, there is a... Tying a shoelace doesn't have the ethical decision... of do I run into a group of people... or do I ram into the other car in front of me... which is something that a self-driving AI has to make. So there's a different sort of parameter set there. So, building robots that can... so the one thing that you find... if you look on factory floors and stuff... and talk about robots replacing human factory workers... and one of the hardest problems is that programming a robot... to just pick something up and move it around... requires an engineer that gets paid six figures. So a lot of factories don't replace human jobs... because there's a limiting factor... if people can program the robots to replace the humans... because it's an exact problem. But there's a whole field of people who are building robots... that you can teach. It's a general 3D robot and it has eyes... and you like show it a job... and then it tries to replicate it in a very natural sort of copy... see what you do and reproduce that motion. And a lot of the limitation there is... it's a lot of it's around the mechanics... and a lot of it's around sort of the vision problems and so forth. I think eventually they'll get good enough... that they'll build a robot that can... you can just show out a tie of shoe and they can do it. But there's a lot of sort of just technical dexterity issues... that you can't conflate with the idea of artificial intelligence. I mean the idea of solving a knot problem... I think again is more like expertise. You can build an algorithm that can do that... but the problem is building the mechanics... that can grasp the shoelace... is a separate problem from the intelligence... to learn how to tie a shoe. Ya, but can it like understand what I do... for example if I would show with my hand... that this is how you write, then could it... like is it possible that... when can we build this kind of device... that can see with the camera what I saw... and then it mimics it and learns like... how to do physical interaction. I think there's AI involved in that. They are building those systems... but it's a bit different than the stuff... a lot of the big data sort of AI... that you guys have been talking about. I think one of the things you want to keep... remind ourselves is... I think it was Neil deGrasse Tyson... in one of his shows he mentioned... the difference between us humans... and our next nearest mammalian intelligent being... is 0.5% in our genes. And you know how different chimpanzees are from us. Can chimpanzees for example... have this conversation as we are having today? Maybe a chimpanzee can tie a shoelace, I don't know. Maybe a chimpanzee can. But the chimpanzee does exhibit... what we would call intelligence... to a certain extent. So the difference between us and 5%... 0.5% difference to the next level... could that be what machine intelligence would be? So they are so different from us... that we look like chimpanzees... to machine learning algorithms... as what chimpanzees looks to us... from an intelligence perspective. Think about that. Any comments? I would like to comment on your statement... where you said is the Turing Test valid anymore... as a criteria. And I think you are right. It's pointing to the wrong direction... if you want to see what machine learning can do... and artificial intelligence in general. But it's still an inspiration... because the Turing Test raises philosophical questions. But to detect artificial intelligence... or to show what it can do... it's maybe pointing in the wrong direction. Okay, you got a question here? Yes, so what I understand from the discussion... is that AI and robots... will take over our jobs basically. So my question is rather... what does it mean for the society as such? So what's the impact for all of us? We heard just recently... we heard a couple of people talking... about let's say a basic income... like Elon Musk or Joe Cazer, Siemens CEO... or Bill Gates talking about tags on robots... stuff like that. I think one of the things that you may want to look at... is I mentioned this in my opening remarks... that there is this project... a group of people came together in January in California... Future of Life. So they have a set of 23 principles of AI... that they have crafted. And I think one of the comments... one of the stuff that you're mentioning... is mentioned in that 23 items... which means essentially... the humans are meant to be... default to be higher ranking... than AI algorithms are supposed to be. So we need to try to create... some kind of a mechanism... where we manage the algorithms in the AI space. So you don't want to have two AI systems... trying to compete with each other... to the detriment of all of us. So there are some ideas. So I think this is something... that we need to have an ongoing conversation. We need to figure out what it means to all of us. Is there a line there? Oh, there is a line there. Okay. I hope I gave you an answer. So you want someone to reply? Yeah, he wanted to. Well, I think my comment is that... again, if you want to understand machine learning... and what it really can do... you should experiment with it... and understand it. And there's playgrounds... to understand what neural networks do... and how the algorithms are working. And as open source engineers... you are all capable of doing that. And the more we can make these tools... and machine learning algorithms accessible... to people that don't program... would be a great way to get this understanding... out so that people realize in a concrete way... what things are actually happening right now. Okay, thank you. Okay, thank you. Just have... asked for comments... from the panel... about the use of AI... in the views of literature and science. So the use of AI in literature and science. Yes. Is that the question? Literature. Literature, yeah. Literature and science. No, literature and arts. And arts. Okay. So you're talking about the use of AI... in those two fields? Yes. Any comments, gentlemen? I already pointed out the Project Magenta... which is a Google Project... to help with music composition. There's also many papers on stylization... where you take in different styles... and it allows you to apply them to new images... including kind of a real-time technique... where you can blend different styles. I have kind of an animated gif... in my talk tomorrow. But I think it's really an interesting thing. And my background is in computer graphics. And a lot of that is trying to figure out... how artists should interact with computers. And there's been a huge step in computer animation... towards more technical ways... of doing computer graphics. And I think the same is going to happen... with machine learning and imagery and music. Okay, I ask another question. So why is the possibility that AI... will actually achieve emotion? If we achieve emotion? Yes. What is the possibility of AI achieving emotion? I mean the possibility that AI will have emotions. I don't know. You can ask yourself... why is it useful for humans... to detect something like beauty? Why is it useful to think something is beautiful for humans? What is your use in life? Maybe you are able to perform in some way better. And that's the reason we have this kind of emotion. And there could be an analog way... that machines detect... that they find some beauty in their environment. Because they say, I'm able to function better in this environment... therefore I'm seeking for this kind of beauty. So maybe it can be important... but for a practical purpose. But maybe we humans have the same practical purpose. So it's also a philosophical question. So next person, this mic. Hello. Emotions. I have a question about... Closer to the mic, please. I have a question about the multiple AI implementations... where one company does NLP, the other one does a little peak data. Maybe somebody does something else. There is some... Google is in some stuff, Microsoft, IBM... or any other third-party implementations. They have different characters. And the question just came about the Skynet. There is no parallel in the Skynet concept. There is no parallel second AI. Do you think that those implementations... they somehow get merged into one single AI? Or do you think there will be multiple AIs existing in the world... and maybe coordinating and talking to each other... and then we just take part of this interesting multi-AI society... like multiple gods or single god? So multiple AI coming together to become... And become one single AI? One single AI is called Skynet. Or not only Skynet, but maybe the thing is about computation power. And if the one who has the most computation power... will most probably win this race... and do you think that an alternative without that much power... can somehow run in this race or has a chance... there will be only one winner. That's obvious. Can I? Yes, please. Services in the internet should be decentralized... because it's better for everyone. We have a lot of centralized services now. And I think it's a good idea... to formulate the same for artificial intelligence. Centralized artificial intelligence may be bad for humanity... and decentralized may be better. So the answer is decentralized is good, centralized is no good. Thank you. Alright, okay. This mic. So the topic of the conference is... the open source part of it and the power of individual, right? And throughout the talk, we came to this topic of, hey, there are no publications out there that everybody can access. There were days that, for example, Steve Jobs... and was could sit in their garage and then write code... or hadoop in the Google can become very big. And everybody could do it on their own. But isn't it a little bit, you know, kidding ourselves to think today... that we have this power of actually working on open source level... on AI, artificial intelligence and neural networks... knowing that they actually need super-super... big kind of resources to be able to work on that. And this kind of towards this democratization of AI... that at the moment most of the power is in the hand of Google... because they have super resources behind them... while a guy in a garage actually cannot make something like that happen. So basically if you look at the kind of computation power... you need to do research in AI, it's typically... you can do many papers on a single GPU. So I think in the way that a computation is really... quite commoditized and affordable, you can do a lot on your own. At the same time, there are cloud services... where if you get a community together, you can get significant resources in order to do AI. In addition, a lot of the open source toolkits... are providing pre-trained models... and these types of things like the image model... that Google provides, give you a checkpoint... that you can use immediately for inference... and do a lot of applications. The power is not in computation, it's in the data itself. Who can store all this data that's out there? Ask Watson guys. They will tell you the IPs in the data, not in the tools. Well no, there's power in the algorithms too. Like there's powers in the architecture. What most is done with deep learning right now... is what is the exact architecture... where you connect how many layers should I do? How many skip connections, for example? So these things, you can then take your data... and you can apply it to a pre-trained model... and get a specialized result. So for example, there's a tutorial on our website... which allows you to specialize the ImageNet... so you can detect flowers. And people have done lots of interesting things with that. Including use it as... I think there was a train detector for a Raspberry Pi. I think we can only take one more question. So... Inny minny minny mo. Because you're up. Sorry. Yeah, thank you for... your enlightenment on a lot of things here. So on the topic of deep learning... a few days, a few weeks ago... I posted this question on my website... is that can deep learning... actually count the number of fishes in the ocean? I'm not talking about species. I'm talking about counting. And my answer and my challenge is it can't. So I'd like to hear the panelist view. I have a second question which is rather short. It's about democratizing IP addresses. In what direction does AI do... on democratizing IP addresses? Thank you. Alright. First, second question is... to democratize IP addresses? I have no idea. Maybe we're at the wrong panel for that. We need to ask the AI to answer that. Maybe somebody in the audience. Somebody else in the audience can answer that. Second question. Nobody? Alright, but you have an answer for that? Mike, Mike, Mike, Mike. Can't hear you. Do you mean when you say... I don't know where she went? She's right here. Oh. Democratizing IP addresses. What does that mean? Typically the word democratizing. Because currently is controlled. Unless you say it isn't. Okay. I mean, that's why I ask for clarification. I think... pretty sure what she meant by controlled is... the ICA are the people who decide, you know... A level BC and the D level IP address. I think that's what she meant by... democratizing. While you're thinking about it... let me go to the first question. Can AI count the number of fish in the sea? That's your question, right? First question. Difficult to imagine. For me. I mean, if you actually want to count the fish... you have to have something that you can see the fish first. And so, I mean, I guess... I don't even... I think you could build... it wouldn't necessarily even be an AI problem. It's more of just... can I build a sensor network big enough? Or maybe you can build something where you take all the sonar pings... from all the networks of all the subs... and do some deep learning on it and figure out... how many fish based upon the interference pattern or something. But it seems different, right? You could argue that you can only count in a statistical sense... like all counting methods are based on... if I sample this in a certain region... then I'm going to see this many fish... and therefore it extrapolates to some estimate. I really like to challenge this... because there are many reasons that force me... to say that it's not possible... at least in the near future. Because all these things that you just mentioned... that can't be done because fish... do not stay static in a place. They move all the time. And there's fish migration. The pattern can change depend on climate... and all sorts of patterns are in the ocean. So this is a near impossible question. So actually I'm here... to ask these luminaries in the panel... to enlighten me because it's a real question. It's not I posted this today. It's a few weeks ago. It's on my web pages. I think there's an easy answer. Following the definition of artificial intelligence... you can say that... a machine is doing what a human would do... and that means to a reasonable guess. I think also you turn it the other way... can the humans count the number of humans... that we have in the world? The answer is no. Senses are extremely hard. We don't actually know how many people are... in a given country or area, in a given time for the exact problems you point out. People are being made, people are dying, people are moving. It's a hard question. I wouldn't want to count the number of fish... that love the fish on the grill. That's what I like. We're out of time actually. Can I have a few more minutes? I just wanted to make a comment on that. So the lady mentioned that... the complication of counting fish... is because of the fact that the fish move. There's another aspect to it... which is the lifetime of a fish. So you need to be able to count all of the fish in the sea... in the time that the shortest lived fish lives. So it's not just a question of... counting the number of fish in the sea. It's also being able to count fast enough... which is probably a far more complicated... mechanical problem, not exactly an AI problem. There's also a biological question of what is a fish. I'm sure if you get everything in the ocean... and you showed an expert everything in the ocean... species that we don't even know if they're fish or not. I think it's a dangerous question... because artificial intelligence with super power... would try to dry out the sea to count the fish. To dry out the sea to count the fish. As opposed to boiling the ocean. Well, on that fishy note... we've come to an end of what I enjoyed as a panel discussion. I hope gentlemen you had fun time as well. Thank you for the questions from the audience. Andrew, Michael, Stefan and Andrew. Thank you very much for your time... and your comments and your conversation. Thank you everybody for your patience... as well as your interesting questions. And I think with this we are done, right? So thank you very much guys. Let's have a round of applause for our panelist... for artificial intelligence.