 Mae'n meddwl i'r ddweud wedi'u cyfrifio'r Gruffydd Nôr, ac yma'n mynd i'ch gweithio'r gwaith o'r ddatatau. Dwi'n dechrau, rydw i'n ddweud i chi, i gael eu gwirio i'r ddechrau eu phoenu ohono. Can I say our speakers main talk will be on the record. After that the question and answer will be off the record. So, we're I think privileged today to have Dr Stefan Heumann with us. He's a member of the management board of the foundation for new responsibility in Basiton, Berlin. His topic is China, the United States and Europe. Germany's AI strategy in a global context. Felly, os yw'n mynd i ni'n gyllidio y Dyn entitledau ichi ddweud wedi'i ddechrau ar gyfer y dysgu'i ddechrau. Ieithio ddweud amgylcheddurau ar gyfer y dyfodol, ac yn debyg ar ôl o'r strategio ar gyfer y dyfodol. Yn ddweud, mae'n ddweud o'r ddweud o'r ddweud o hynny, yw ddweud o'r ddweud o'r ddweud o'r cyffredig. Mae'n ddweud o'r ddweud o'r ddweud o dr Hoymans Dyftung, yn y amser i'r gwahodd y gallwn ei gwahoddiad yn ei wneud ohol ar gymrydeithasol. Roeddwn i'n g uhhhol o gyda'r bysgel. Roeddwn i'n g Sisters Okaith. Roeddwn i'n gynnig i ymddangoson mwy o'r sefyllfa IEZO? Roeddwn i'n gynnig i'r newid yddiad ar hwnnw. Roeddwn i'n gynnig i'n gynnig i'r newid yn gyffredinol i ddiwethaf, Catherine, Yn ystod y gallu'n cyffredinol i Dublin. Mae'n fyddion i chi i gyddiolol i gael a'i gyda'u am ymweld i chi, ac yma'n edrych i'r gyda'u cyffredinol i'r target i'r cyfrifolio i Irish. Mae unrhyw gyffredinol yn ddiwethaf, yn ffwrdd ymlaen, yn y ffordd a'r cyfnodol i gael cyfleol cyffredinol o'r cyflwyno cyflwyno. Mae'n cyfrifolio o 20 cyfrifolio ac yn Ymbylwn. Mae'n ganddaraf, ac mae'n ganddaraf. We work with both the tech sector and with government officials, but also with NGOs and academics, and we really try to bring them together to develop expertise. We don't think that you can find the answers on how to react and assess the rapid technological changes at your desktop. You really have to go out and build networks and find the experts that engage in conversations with them. Over the next two years, in 2017, we had a conversation with the German Foreign Ministry, which is a close partner. And we run a lot of projects with them. They were interested in a project that was a new technology that was kind of a horizon that they should start thinking about. They were self-critical, kind of missed the internet trend. wrth dweud y tynnu gennymol, oeddeithasio mewn cyfleol, a opyniaeth newydd sy'n gallu cyfrifonau cyfioedd hynny, mae'r gwyloedd yng nghymru a'r desgol. Rydyn ni'n arferwad i'r rhaglen, dyweddwn ni'n cyfaciliadau, ma'r rhaglen newydd. Rydyn ni'n gweinydd am hyn arill yn nhw, a rithog rydym yn sefydlu yng nghymru ychydig eisiau ymddangos ates. erioedd yn ei ddweud o'r economi digital a'r bobl yn ei ddweud o'r economi data i'r economi i'r plathu US i'r cyffredin iawn. Mae'r bobl yn ei ddweud o'r 21 yma eiffan grei fynd o'r economi data. Ac mae'r ddweud o'r cyffredin a'r ddweud o'r cyffredin a'r cyffredin iawn, ac yn ddim yn ei ddweud o'r economi. Mae ddechrau'r data am yw, mae'r un o'r cwyrdd yn intech arnynol cwyrdd, fyddai'r graffa i bwysig, mae dafwyrd o'r gyfrifol, o ffwrdd gwirig ymgyrch, ymgyrch, a'r internet of things, mae'r ddweud o ddim yn higwrdd ar y dyma'r amgwil, a mae'r ddau cyfrifol mewn data ac dasgwyr. Mae hynny i'n ei fwyaf o gymaint i ddim yn ddweud i'r ddweud i'n ddweud i'r ddweud. It's that. We will need intelligent automated systems that make sense of all that data. That's basically what under this catchphrase of AI we are talking about. We are talking about tons of data that is out there that no human being can process in any way, meaningfully. We need systems, computer systems that help us to make sense of that data. Aie is going to be a key technology as we moving into that connected data driven world. We started out, like we usually start out, we do a research sprint, we look at what we can find. Could we find any papers and publications on AI informed policy? We didn't find very much but we did find AI experts and we were talking to diplomats and we structured an initial input paper around that and started a conversation with them. And part of that research and looking internationally we found of course what everybody of you has heard about, this global race for AI that is largely led by the big tech companies from Silicon Valley that also China has now produced a very ambitious strategy for AI and China wants to become a global leader in this technology by 2030. We also identified that as a key technology moving forward and we also seen that inside of the EU other countries were really working on AI. We saw that Great Britain was having discussions on an AI strategy that France had started a process and what we were frustrated about at that point is that Germany hadn't started yet working on an AI strategy given how important that technology is. And Germany was a little bit reluctant to join that and I will explain to you why and I also will show you the moment when that changed. So that is Chancellor Merkel in early May 2018 on her visit to China and when she visited China she also went to Shenzhen and some of you may have heard that Shenzhen is one of the AI hubs and especially hardware hubs in China and she was visiting start-ups there, robotic start-ups and she was personally very very impressed what the Chinese were already doing with that technology and she came back from that trip and she was really questioning her cabinet and her ministers what are we doing on this technology and is it really good enough what we have been doing and the question is really good enough what we have been doing in Germany is a longer going conversation that we are having about digital and tech because the German track record is not super great if you look at this statistic from that the EU has the digital economy and society index and I marked for you where Germany is right at the middle it sorts of assesses how well developed the member states are of their digital economy and digital society and they look at connectivity, they look at skills, human capital they look at how widespread is the use of internet services they look at the integration of digital technology into companies in the business sector at large and they also look at how digital public services are and you will see that Germany is right at the EU average and you will probably pleasantly have noticed that Ireland is ahead among the top groups but there has been a long going conversation in Germany that we have had for years that we are struggling to adopt our public sector to digital that we need to be more proactive in pushing digital technologies into our society into our education system but also into our business sector and so the conversation about AI started at a point where there was already a lot of criticism on the German government that we are not strong enough and not doing enough at the moment and so Germany joined that race late they decided we really need to turn this around and we really need to move Germany forward and you can see that we followed a trend that has been going on globally that a lot of countries have been adopting AI strategies and this is from Tim Dutton who wrote a block entry that is artificial intelligence strategies that I highly recommend for anyone who wants to dig deeper into it he constantly updates it and he gives you an overview in terms of where the discussion is in all these different countries that have started or already adopted AI national strategies and you see we also have a process going on now on the European level but we have quite some member states being also from the EU already being very active in the field particularly myself like the Finnish AI strategy so if somebody wants to look at a particularly innovative one from my perspective I recommend them to take a look at the Finnish one and it's been translated into English so you don't need to learn Finnish to read it so what complicated the conversation in Germany is that AI is not a new research field in Germany last year the German Research Centre for Artificial Intelligence celebrated its 30th anniversary and this was one of the reasons why the German government has been reluctant to adopt an AI strategy because our research ministry has always been saying we have been working on AI for 30 years this is a very large research centre that has several locations in Germany and a network of over 800 researchers that work in it but they have been mostly working and focused on what's called symbolic AI AI is really not a new research topic the traditional AI systems go a long way back over 50 years people have been trying to make computers intelligent and there's really two camps on how you approach it and the traditional camp is what you see in the slide is the symbolic AI that is you start with knowledge you understand something and then you try with logic to tell the machine to follow that logic so a very good example for a symbolic based AI system would be for example systems you use for your tax returns so there's lots of rules around if you're in this income category these kinds of rules apply and you can deduct this and that and a lot of that you can translate into logic for a machine to learn and then the machine can help you calculate your taxes that's what's called traditionally an expert system and that's the symbolic AI the problem with the symbolic AI is when it gets really complicated it's really complicated to translate that into logical systems that work consistently for example how do you explain to that sort of system to recognize a horse or a human being you have to say it has to have a nose and ears you describe it but monkeys have that too and other animals how do you differentiate that so that kind of approach also ran into lots of roadblocks and what we're seeing now really taking off in AI is the statistical based approach the data based approach which is we don't start from logic we don't start from knowledge that we understand what a human is and then we try to translate that into rules and if when rules to a system that can apply it we rather let the machine learn it based from data so we give a machine lots of pictures of a human being and the machine will figure itself out the patterns behind a human being and get so good at the recognition of that pattern that it in the end can recognize human beings by itself and that's the statistical machine learning approach and this machine learning approach is really what took off and it's really what's driving around all the AI hype right now I think the AI term we can have a conversation how useful that is because I don't think we're talking about artificial intelligence we're talking about very specialized systems that can do very specific things that is very different from human intelligence and really the big hype is on machine learning is on training computers at certain tasks or at the ability to recognize faces to recognize speech based on massive data inputs and these are really the drivers of machine learning that have come together in the past years so neural networks is this learning based approach and it's derived from neurons this is how our brain functions it was inspired by neurosciences and how the brain works and translated into computer software that simulates the learning process then we have data and during the past years the availability of data has just exploded the problem by neural networks didn't really take off earlier even though we had the concepts already developed in the 1980s was that you need a lot of computing power to build neural networks especially deep neural networks the better you want to get at the pattern recognition the more layers you need in your neural networks and the more computing power you need to do that and we just didn't have that kind of computing power in the 1980s and 1990s to do that and so when we started to have the computing power and also the large availability of data we had key ingredients in place there are two more that you need that's talent that's people who are able to work with neural networks to work with data to write the algorithms to put it together and that's currently the biggest bottleneck because these breakthroughs are very new we don't have so many people trained in this specialized area that there's a global competition for experts in this field and that's why you read a lot about this in the newspapers that AI experts get these very high salaries why so many people get moving into the tech sectors from the universities because you just can earn so much money with this knowledge at the moment and then I wouldn't underestimate this this is why Silicon Valley was really leading in the application of machine learning and practical use cases why would you invest so much money in building computing power in analyzing data and hiring big talent if you don't have a specific use case and Facebook and Google and other companies Amazon they have the use cases Amazon wants to make predictions on what you want to buy for that they need to analyze all the data that you leave on the platform companies can't do that you want to automate that process Google wants to show you relevant search results also for that they analyze a lot of data and want to give you automated predictions Facebook wants to understand your interest based on your activity on the platforms they even want to understand what kind of pictures you upload for that they need image recognition to recognize pictures also to do their content policing to hire millions and millions of people to do that so they have really the use cases to work with all their data so that's the reason why they were pioneers in this but the use cases are much larger now and they go beyond the internet platforms they go towards diagnostics for medicine they go towards smart manufacturing or mobility and that's why we have this broader conversation how AI is a key technology all across industry and society so everyone talks about China and the USA leading and there's a great resource again if you want to see what are the most important trends in the field I highly recommend to you the state of AI report that two leading experts publish every year I don't even want to go into the details of the slides but it builds on what I said before they look at hardware, data, research and algorithms and they also look at the commercialization what worries me about their state of AI report is there's something missing there in my perspective where's Europe? they're talking about as if Europe is not even in that race at the moment and that is what I'm very concerned about if we follow the assumption that AI is such a key technology to make sense of data we're moving into a data economy we really need that capability and we're looking at the state of AI report and the leading report that's cited by experts and they're not even talking about Europe in their comparison so what do we need to do? we have woken up in Europe there's a lot of awareness about the importance of AI I think that's well understood and the European Union has started a strategy process they have already published a first paper they had discussions with member states and a further agreement how to move the EU forward but you see it's a long-term process and the EU, that's a high-level approach they like to call their expert working groups the high-level expert workgroup on AI and things take time in Brussels and Brussels is also in my opinion quite removed from the ecosystems that are driving the development of AI so there's a very important role for member states to play and the EU is actually also encouraging in its own AI strategy the member states to adopt AI strategies so let's talk about the German AI strategy and what it can be the contribution to Europe I apologize for this German slide but what it shows you in Germany so the lighter blue is the number of AI startups and companies and the darker blue is the number of AI research institutes and what you will see is Germany as a federal state has pretty much equally distributed around Germany AI research institutes but you will see that in terms of commercialization it's really focused on a few cities it's Berlin, Munich and then to lesser degree there's Hamburg but it's really the big cities in Germany Berlin and Munich that are the startup ecosystems for commercialization of AI and this is one of the biggest challenges for Germany at the moment because we usually want to promote research and innovation across the country we want to have it across all the lander we call them across all the states but we see that the ecosystems really take off in the big cities and I think we have a similar challenge in the EU everybody wants to have AI research but we really see that the strong ecosystems are based in cities the strongest AI ecosystems are in London and in Paris in Ireland I would imagine it's around here in Dublin in Germany it's Berlin and Munich and so we either need to figure out how we strengthen the commercialization in the smaller cities or we need to just accept the reality and say well we have these strong clusters what's happening there and we just need to focus on our policy there what you see is that across Germany now the different states are starting to look at their ecosystems and trying to promote them this is the governor from the Green Party the only Green Party governor in Germany Mr Kretschmar who is signing a program for what they call the cyber valley that's the state of Baden-Württemberg in the south west a very strong manufacturing machine producing area also the home tool Mercedes-Benz and they have developed a program connecting their research their leading research institutes on AI with industry and they're bringing in Bosch, they're bringing in Daimler they're bringing in SMEs also Amazon is now developing an AI development center these are kind of the local ecosystems that we need to build and that is a contribution that we can make thinking about what are our strengths in certain areas and how can we develop these ecosystems the second challenge is data the big US platforms have lots of data and we've talked about that for machine learning you need large amounts of data and we don't have these huge platforms in Europe at the moment so we need to encourage companies and industries to collaborate and to share data and this is an initiative in Germany called the industrial data space that's trying to develop a platform where companies can pool data and work together and share data and for example if you have lots of small companies that if they would come together and pool their data resources they could develop much better machine learning based algorithms than they could do on their own you even see that the big car companies are talking to each other to collaborate on developing autonomous driving systems because even a huge car company doesn't have enough resources but this is a challenge how do we build platforms for them to collaborate we can go more into detail in the Q&A if there's interest one interesting way to do this is for example what's called federated AI it's where you have the machine learning happening on site and then only what you learn you share with others and so you develop a master algorithm basically that benefits from what many different places learn you could also use that in the health sector that you have patients, algorithms that do diagnoses and what they learn and how they improve you combine what's learned in many different hospitals the advantage when that happens on premise is that you don't have to transfer any data you don't have to have a centralized data repository but rather you learn decentralized and you just bring a centralized the results of the learning together these are sort of technical approaches that I think for European AI we need to look much more into because they are much more compatible with our values and with our ideas about data protection and of course skills that are very, very critical we were just talking about this over lunch this is a traditional strength in Germany we have a very strong apprenticeship model we have strong trade unions that work together with the companies to develop training and qualification courses we have a very educated workforce in Germany that also is used to have a say in the workplace that critical interacts with machinery with management, with engineers and we need to get them ready to understand machine learning AI as they are going to be working with it and they need to be able to understand the basics of it they don't need to be able to code the machine but they need to be able to understand what are the limitations of the technology what could go wrong how can I help my engineers and my management to improve processes and use the technology and I think here we really could have a cutting edge in Europe with our trained and highly qualified workforces not only over China but also in the US as we are thinking about bringing AI into industry and then here is what the EU could really do this is an initiative also that came bottom up it's so-called the ELIS letter it's a letter that was signed by leading machine learning scientists in Europe that are demanding for more European support to build centres of excellence on machine learning they see three challenges first they argue that machine learning, as I told you before is really at the centre of this artificial intelligence revolution they think that Europe is not keeping up that the leading research institutes are overseas at the moment and that China is building them more to develop academic research centres of excellence that are combined with industrial labs and testing and to do that to really have visibility we should have a few of these flagship centres across the EU that collaborate amongst each other and really show European excellence in AI research but Germany is also the country of environmentalists about skeptics on technology and I showed you earlier the slide on the cyber valley there are protests against the cyber valley in Tübing which is a small university city where one of the leading research centres sit people are protesting against Amazon developing a huge research centre there and they are very concerned about AI taking away their employment they are worried about unethical AI they are worried about this technology being used to fight wars to develop weapons systems so I think that ethical discussion is really important there are legitimate concerns and if you don't engage with the population I think we will see a backlash against that technology and we see some backlash already I'm concerned about it I think much of the backlash is also based on misinformation about the technology because of the term artificial intelligence I think people widely overestimate what's possible at the moment because it's not human based intelligence and if you look at the media reports they like to use images of human looking robots to visualise artificial intelligence it's very far from how these systems work there are a lot of concerns in the broader population I think the main concern is about people's wealth and economic opportunities so what does that mean to their employment, to their future and the other one is am I going to be subjected to some sort of artificial system that decides over me whether I get a loan or whether I get a car insurance and do I have any way to verify that the decision was being made fair and that's why I do think that the European focus also on talking about ethics and the framework for developing ethic guidelines for trustworthy AI is a very important component of the European approach I like about it that they don't only have ethical guidelines that really centre around putting that technology in the service of human beings but that they also include recommendations for technical verification and governance mechanisms so for accountability in AI is a technical problem how do I understand if I use a machine learning algorithm how the machine learning algorithm learnt and how it makes decisions that's a technical problem and it's also a governance problem what kind of rules should a company adhere to when it thinks about training data when it thinks about quality assurance when it thinks about when can I put this into the marketplace so the framework is a very important starting point for broader discussion about how we want to bring AI into Europe and that's it I'm looking forward for a lively discussion on this and thank you very much for your attention