 Hey Thomas, it's great to see you today and you're doing some amazing work with this focus group AI for health So let's get into a dialogue about it and so you can share some of your experiences and especially from the past week I see. Yeah. Nice to meet you. Yeah, let's start. Oh, you know, Thomas Can you outline some of the top milestones that you're like professionally and then also, you know, some of your top leadership roles as well Well, let me start with the thing that I'm probably most known for which is my role in video compression. In 2000, I have been appointed the associate reporter of ITU, question six of study group 15. We were able to actually produce some things that actually are mattering today. H264 or impact AVC is now 80% of all the bits on the internet. Just a there was in 2003 and in 2013 we produced HEBC or HN65 and now all 4k video is in that format and I was very fortunate to be awarded the ITU 150 awards in 2015 at the 150th anniversary of the ITU and I could stand among giants like Bob Khan and others who have received the same award. So that's the thing I'm most known for. I'm professor at the Technical University of Berlin and I have been actually publishing actively about my work and it turns out that Thomson Reuters named me in the list of most influential scientific minds as one of the most cited researchers in my field and I've gotten some awards and best paper awards and I was a co-founder and advisor to several startups. Some of them made it with 100 million dollar exits and some didn't. So that's that and then I'm currently the executive director of the Powell behind the Hertz Institute. It's a research institute with about 500 researchers and it shows the steady growths and increased output and I'm very proud of everybody I'm working with and last but not least I just recently became this chair of the focus group artificial intelligence for health and we are very excited about that and I guess we're going to be talking about that more today. Yeah I mean again amazing work you have just a quite a remarkable background and you're quite young too so I'm always amazed. You know what are your top research interests and the predictive outcomes and why? So the research field of the Henry Hertz Institute are machine learning and artificial intelligence as one of obviously computer vision, video coding and then communication topics like 5G optical fiber and visible light communication and the nice thing about my university position and the institute position is that the scope of our research can range from basic research to application of our research to practical use cases and the application areas that we can be active in are health industry mobility entertainment and we have lots of engagement with all of these people have many many thousands of projects a year and the aim for the institute is to actually produce technology for the digital transformation as this you know the digital model of the world in some way will be one of the strongest driving forces for progress. Well again just find it so fascinating now these next questions are about this program the focus group AI for health and people abbreviating this as the FGAI4H that's kind of like a meme you know how did the focus group come about and why are the major organizations driving this global initiative? So the the abbreviation comes from the ITU, the ITU was in likes abbreviations but it's also good to have a term that you can search easily. So at the AI for Good Summit in Geneva May 15 to 17 in 2018 that you are also co-organizing and I'm helping sometimes to find speakers we had AI for Health session and there were 15 16 17 projects been presented we also have our AI for Health projects we do proteomics, gate analysis, ECG analysis, EEG etc and with all of these projects it was became clear that they are limited to the data set that was available to them and then it wasn't clear what happened with the results and whatever came out as an algorithm afterwards. So we saw that there would be a strong demand for standards in particular and evaluation how well these AI methods are performing and we would basically have to apply this to health data and then see how well these algorithms are. So we then discussed this further and we would basically say in order to bring AI to the global scale we would first of all you know need to identify use cases that are relevant to public and clinical health. We need to understand the data sources like how are these data being sourced and we have created something to discuss that. We would then need to evaluate the performance of these AI for Health solutions and we will then at the end having documented everything from the data sources to the use case the ethical aspects everything and how well these are performing these can be used as documents for recommendations like by WHO or national authorities. So why is it special? Well the ITU is a UN specialized agency and it provides the engineers in this and WHO is responsible for worldwide health. So if we combine the forces of engineers and health experts I think we have the chance to accomplish what we want to accomplish. I guess it's a really great combination of forces given the World Health Organization definitely is number one and many aspects of public health worldwide is recognized as such for many many years and of course the ITU is over 150 years old and has set many of the global standards that we just take for granted today. So really a perfect combination of the two and then you as chair with you especially your tremendous background and the contributions you've made you know you have those two Emmys in the background I don't know if people could see them but you know you've done some amazing work so and then of course your research background and a lot of it sort of practical basis to it. I think it's just a perfect storm as they say in sort of a common term. Now what problems are being solved? Well let me give an example so worldwide the thousands or millions of health workers are missing. There's actually worse 1.7 billion people on this planet to not have bank accounts but 1.3 billion of them have the phone. So we could actually bring in this in order to come up with new solutions to the health problem once it becomes a problem and so for instance using telecommunications to provide health advice for example is something that we are looking at. If we take a step back and look at the problem on a more generic scale and a high algorithm in health to some degree produces a mapping from data health data or environmental data or other data to something an output value that is relevant to health to health like an indication for a health situation etc etc and it may suggest treatment decisions etc so what the activities for this focus group is basically find use cases for AI algorithms in the space and then see how well they are performing so how could we do this this is what we are currently working on. Well you know it's a it's a grand mission and definitely will have a practical impact on the entire world going forward so what are the activities of the focus group and specific outcomes from these activities? Well first of all we we have a structured activity structure is important to achieve something. The first meeting we just had from 26 to 27 of September 25th of September we had a workshop and we were pleased to be able to have this meeting at WHO headquarters in Geneva and we had a thrilling 130 participants and many have been listening in through remote participation and there was not only I have to say enthusiasm at the beginning but there was also some skepticism whether this is actually possible what we are discussing here and so what we took is we took some real steps towards what we want to achieve by having a very constructive discussion at our first meeting we would basically find a way to solicit inputs to our next meeting which will be from 14th to 16th November in New York City and we would then basically issue documents that would start to describe the process that we are developing the process is to request submissions on use cases for AI and health together with associated associated data we can go there in more detail we would then also have to tell only which data we are accepting and which kind of algorithms there are certain important rules and also what are we going to do if something gets submitted to us all of that has to be laid out and has to put in documents so that people can understand what they are contributing to and so we have laid out our meeting plan until September next year and our third meeting our third meeting at DPFL in Lausanne in January next year I hope we'll have some really important progress because these standardized assessment methods then that we want to develop for AI for health solutions they would help us to assure the quality of these methods then once your AI method is evaluated in terms of quality people start using it on a bigger scale that would foster your adoption and practice it would basically help you to get whatever you have to develop out in the world on a massive scale you would then be able to create more data and more turnaround and improve even your solution and in this way we hope that some of the things that we are evaluating have a strong impact on global health you know it's an ambitious plan and one that will be carried out because just due to the parties involved and so how will these solutions be achieved so we had some first exchanges and this is obviously going to change over time but in principle what we want to do is we want to create standardized input data sets so we ask people to submit data to us some of those data should be open so that people can understand the structure the nature we need a full description of how these data have been gathered and we also need undisclosed data because those undisclosed data would then be used for testing so the dataset contains not only input data to the AI algorithm but it would also need to have confirmed output data so if there's a indication on a diagnostic value for example we would need to have this to be confirmed afterwards and so once we have split the data that we have into the public the open ones and into the undisclosed ones we will then create metrics for comparisons in health a true false and a false true can actually mean very very different things so we have to be very careful about that because if you are missing out on something maybe worse than if you overdo something but both are not good so we have to weigh them in and basically have metrics that also include other aspects like how much data is needed to achieve a certain output or other costs involved in using a certain AI algorithm and then once we have it all together on a particular use case we will then go ahead and evaluate AI algorithms so to give you an example we had a proposal on breast cancer data where we would on on the pictures be able to identify the cancer cells using the AI algorithm we have it seems that there's some high quality data sets which hasn't been public so some of it will be made public so people understand it and some of it will be undisclosed and we will be able to for instance test the identification of cancer cells in these pictures through the assessment framework you know you got this program you got this framework you got this great team in terms of moving forward you know the two best organizations to really get their bite behind it as well you know but we have all of these people out there in the in the world and so how can they the audience participate the title of the of the group that we have created is the focus and the focus group is an instrument that the ITU has created to tackle new topics and one of the aspects of a focus group is that it is open to anyone anyone can participate and we have all means of participation so you can attend our meetings you can join our mailing lists you can actually get lots of information through a website and social media and the whole process is open and transparent so that everybody can understand how we arrived at the end of our at the end and our documentation for a specific use case for example and how we arrived at the assessment framework and then what are the timelines that milestones desire so at at the last meeting we issued a draft call for use cases and data so then i hope that we will get some indications of use cases coming in and also understand better how the data resources are that will be made available to us based on that and based on some other discussions on what are we going to do and and how are we going to technically also do the evaluation we would at the next meeting in November issue the call for use cases and data and then if we see input by January at the EPFL meeting we would then be evaluating this data and by the March meeting and we could actually if things go well have first results by May and at May we are planning to meet with the AI for the summit in Niva where in May this year the whole thing was born so it will take until May 2019 full circle to actually hopefully provide a first result of our evaluation yeah it's interesting the AI for good summit is really has made history it's grown each year founded in 2017 it was even bigger in 2018 in 2019 we expect it to even double beyond that so you have all of these UN agencies 33 contributing this year over 50 global media and participants from every domain actively involved and in fact even the audience was pretty notable as well just there to watch so with the 148 speakers it's a great venue to target at the main meeting of the AI for good summit it's pretty interesting so it's not only the talks that are interesting at AI for good but it's also the crowd that's there so the breaks are as interesting as the talks yeah exactly yeah the networking that these side and hallway stories are compelling now where can people go to find out the whole story of this initiative well we are in the process of creating lots of documentation so we have written a white paper which might be a good start to understand what's going on the website is available maybe you can when you post this put at the website link so that people find it there but you can also just search for the term AI for each and you'll find already quite some amount of resources to read what we are doing and again if you want to participate we are open for those who want to constructively contribute to participate you know you got this open program where anybody can contribute so what resources does the focus group need to achieve success well to start with we need data health data in particular we need some of them to be open and some of them to be nondescript that is not easy but there are quite a few research projects and other projects that have produced such data so I hope we're going to get strong submissions then once we have the data we also need AI algorithms that will help us to actually do what we are set out to do then we need people that are devoted to the cause of bringing AI health to the global scale and was that you know they have AI doing good and the people need to have an expertise that you know we need to have health professionals that have medical biological knowledge we want to need we need engineering experts computer science AI etc we need people that are for good data signal and image processing and then last but at least we also need financial support because lots if this grows we will have the need for staff to help we might need some we do need some hardware for the rules operation the data handling the evaluation and also the promotion so lots of support is needed to make this work this program really resonates among so many constituents out there so what are the outcomes and why should each of these constituencies care well let's let's go one one by one button to healthcare so for healthcare I think it's important to make AI a useful tool in public and clinical health and what we are providing here is maybe solutions that provide support for health workers and we also could this way address question issues such as the shortest of health workers that may lead to no health care at all in some countries or long waiting times in other countries and this way where when the health workers can concentrate on other aspects that can be covered with the solution for example they can actually do a better job on those aspects so we could actually increase the quality of healthcare the next constituency you might want to look at is government the government know that we they need to do something about the digital story of health but what and how and what's the method so what we could maybe help them with is to provide data for tangible action for digital health and we may also at the end help them to also not only increase the quality of healthcare but also lower the cost of healthcare at the same time business industry well if you think about it you are building your AI algorithm in your start-up company in your big company you want it to be used to create a business how do you go from your series A type of system even series B to worldwide adoption well AI for health could be that possible roadmap to worldwide adoption for your AI for health solution and then once it's adopted you can make it even better this constituency would be academia and researchers so you would basically have a worldwide recognition of your contributions to use cases to the data and the algorithm and also again whatever you have created in your research could end up being adopted worldwide and so this will be interesting we have the United Nations the the one that stopped the people that support the sustainable development goals I think this activity addresses many pressing problems that are expressed by the SDGs and I hope that it will help there for citizens obviously better healthcare in terms of quality cost accessibility usability speed sometimes you want quick help and you don't get it so speed for AI for health is a big argument as well and then for entrepreneurs this the whole AI for health topic received lots of investment now I think this strengthens the investment situation and it's providing tremendous opportunity for you to basically also take into account this activity and then you could also with your existing venture look into how you move it further in terms of deploying your solution and if you're investing in this space well you could leverage on this focus group to bring your current venture to the next level and have it evaluated and have some approval through legislature and then have potentially large adoption and and this may also spur additional founding of AI for health companies because maybe this contributes to their exit story and last but not least if you are foundation well you are supporting an activity that is aimed for goods that is aimed to help people that aim to improve a very important part of our life it also helps you will also help to bring the topic in the health area from a promise that is being made quite often to reality because we will be telling what these algorithms are good for and how good they are performing and then you can also you know push us maybe also insist on topics that are helping the poor children and those that need the most as well so because they should be in the focus of our group here as well together as with many many other of those applications so yeah I think there's something in for almost everybody here well you actually name every group so definitely a lot of value act to be shared throughout the world and for all the different stakeholders in communities now there's a lot of work to be done it's going to be completed what happens after the focus group completes its work in 2020 well I agree with you we first have to show credibility we first have to show that what what we want to achieve is achievable and then many things can happen I think that the sky's the limit if we do things right well you know Thomas you're you're sharing this amazing initiative you have always laid out a vision that can resonate through so many different communities but also support the UN and government and industry and academia and I thank you for coming in today and just sharing your deep insights and also your commitment your dedication your passion to make a difference thank you very much