 Good afternoon. I'm Jennifer Shanker, Editor-in-Chief of the Innovator, a global magazine about digital transformation and very happy to be here today with a super panel to discuss the rise of the dataocracy. History is replete with examples of over-reliance on metrics which have led to unfortunate outcomes. Today, we are relying more and more on big data and algorithms for our decision-making to solve big problems, to evaluate people, and predict outcomes. You could almost say that big data and algorithms have become a kind of religion. Today, we're going to ask the question, is our faith in its accuracy well-founded? And to discuss this issue, we have with us here today, Igor Tullcheninsky, who is the head of a five billion dollar hedge fund called WorldQuant, that uses algorithms to predict outcomes. We have Yu Yang, who is the founder of a company called YiSites, which was a world-leading cross-language big data analysis platform. We have Frida Pauly, who is a award-winning Harvard and MIT trained neuro-scientist turned CEO, and last but certainly not least, we have Dr. Hillary Kottum, who is a social entrepreneur and the author of a book called Radical Help. So I now like to get right into our discussion and start by asking our panelists, what are the new opportunities that advanced technology allows? And I'll start off with you, Igor. Tell us a little bit about how you were applying algorithms and how this can help society. Yes, let me give you a little bit of background. WorldQuant asset management is an asset management firm, we're quantitative. We're global in the 27 offices in 16 different countries. But what makes us unique is the fact that we have about 19 million algorithms that we use to predict security prices. So I wouldn't call this world a data accuracy, I would call it an algorithm accuracy, because data is growing exponentially, and you know anything that grows exponentially becomes commoditized. So it's really the algorithms that derive the meaning from this data. So let me give you some examples of something that we've been doing outside of our main business. We have a joint venture with Wild Cornell Medical Center. It's called the WorldQuant Center for Predictive Medicine. And one of the things we did, I think everybody in the audience knows somebody who was unable to have kids easily, so they go to in vitro fertilization. And that's a very heart-breaking process sometimes, because you cannot tell easily which embryos are viable. So we have developed an algorithm together with Wild Cornell, which can pinpoint with 90 percent accuracy the viability of embryos by looking at the image of the embryo. So that's one example. Another example, I'm sure everybody in the room has heard of astronaut twin studies where the idea is everybody wants to space travel, but it's important to see what happens to humans in space, in particular what happens to human genome in space. So one twin was left on Earth, one twin was sent into space for a year, a ton of data was collected, and we've developed an algorithm that'll predict what happens to a human genome when you send it to space, basically. So these are three examples. Okay, so, again, let's talk about how you've been applying algorithms. You've also worked in the medical field, and you're also applying them to fake news. So let's hear about that. Ashley, GTCOM is a company getting involved in, like, vitro everything, the big data, some machine translations, and also spatial recognition and image recognition. But when we talk about the big data, what I mean by the big data is not just simply, like, because GTCOM is a Chinese company, but we are not mentioning just Chinese data. What we mean is multi-lingual data. When we put in the search engine, we put a keyword, Apple, in the Google, you're about to find out all the news or data and information in English. But it's supposed to have the data in English, Chinese, Russian, German, but we are everything. That's what we mean, the multi-lingual text, and also the content of the audio, and also the content of the images. And we try to leverage in the machine learning and our piece to calculate those three different types of the, like, formats of the data. And also, what is the most important, we just would like to propose a new concept, like when we talk about data, when we talk about news, for example, we think it's only three status. And the first one is the data. The second one is alternative data. The third one is the mirror data. And that could be no fake news. What I mean is that if there is no truth standing out, and the fake news could be there for real, for one month, for one year, or even for ten years. So we try to calculate the three status of the different, the three status of the data and to try to find out the balance and try to tell the other people to make the judgment. And also we have a company who focused on the medical imaging areas called the Paraglop. And as we all know, China enjoys the largest population in the world. And the Chinese people, one year, need to have like 1.3 billions of CT scans. Clearly, in China, we still haven't got sufficient qualified doctors to review those CT scans. So we try to let the machine to learn the medical images. Right now the Paraglop are able to diagnosis lung cancer, liver cancer, brain tumor, and pathological analysis. Take the lung cancer as an example. The machine already learned like 100,000 confirmed patients' cases. Each patient actually have 200 to 400 CT images. That means the machine already learned that many images for doctors when they graduate from a medical school for his whole career, life. A doctor only can see like 2,000 patients. But the machine already learned like 100,000 confirmed lung cancers. So clearly, algorithms are helping us solve some huge problems. But there can be some downsides to this. And one of them is bias because we are training algorithms, quite often historical data, and that means biases can be baked in. So I want to turn to Hilary because her company has actually invented a technology that audits algorithms to ensure there is not any bias. So Hilary, tell us a little bit about that. Sure. Happy to. So Pymetrics is a company that uses artificial intelligence and behavioral science to help in the recruiting process, recruiting in other human capital processes. And I would actually argue that machine learning is like any other technology neutral. It's really up to us and how we deploy that technology that makes the difference. It's just like any other technology in the world, genetics. All these technologies can be used either to advance causes that we believe in or actually bring us backwards. And so if you think about current hiring processes, there are lots of ways in which humans introduce bias into that process. And so when we speak about algorithms using bias training sets, those bias training sets often were created by humans. So I think that this juxtaposition of bias algorithms versus the straw man of unbiased humans is just a straw man or a straw woman. So I think that actually there's great potential for machine learning and algorithms to reduce bias. They can augment bias if we're not careful, but they can also reduce bias because a human, all of us here are all bias. There's no way to make sure that Igor or you is unbiased. There's just no test for that. Whereas you can actually audit algorithms. That's what Pymetrics does is we've created a technology called audit AI that essentially looks at the output of any recruiting algorithm that we build and ensures that from a statistical perspective, there are no biases with respect to gender and with respect to ethnicity. Those are the two biases that people care about the most in terms of hiring. And this process that we've developed can really be applied to any algorithmic process. And so we've open sourced it on GitHub. But I guess the broader point that I'd like to make is that I think unlike humans, we can develop technologies that will audit algorithms. And as a result, I think it can actually be a powerful tool for reducing bias that having been said. I think untested or unaudited algorithms can and will often lead to increased bias. So again, it's really about the design of the algorithms and the humans that are creating these technologies and how they choose to both design and then implement them that really will determine whether something is a biased technology or not. That's kind of our position. As a follow-up, one of the things I find interesting about your technology is it's supposed to also be able to identify high potential individuals who may have been overlooked. So how do you train an algorithm to look for high potential and how do you even define high potential? That would be great. I was just going to say that. So I think historically we thought of high potential as being a thing that you could say across the board as universal. We actually don't believe that. I think to be a high potential person in a certain sphere may be very different than being a high potential person in a different sphere. So we actually work with companies and we create custom algorithms using their own employee base to identify what high potential means in a particular context. So again, it's much less about creating a world where certain people have high potential and others don't. It's really identifying what area of life, what area of career will you demonstrate high potential in and matching you to that high potential field. And I think that the reason that we're able to identify it in people that previously have been overlooked is because we're not relying on markers that are often tied to socioeconomic status where you went to school. Other things like that are often a pedigree that's associated unfortunately often times with socioeconomic status. So we try to remove that from the process and in doing that really allow for a broader set of individuals, a much broader set to be matched to opportunities where their ability to have high potential will be maximized. Thank you. So now I'd like to turn to Hilary Silver. NGOs and socially focused organizations are becoming under more and more pressure to come up with data measurements. And I'd like you to talk about, you know, what are the pluses and minuses of this? So I think there is increasingly in my world, in the social world, we're about social change and there's this increasing emphasis on what works, which I think is a good thing. There's not enough dollars to go around. We need to make sure that we're investing in the right way. But obviously the business of human change is messy and complex. And so I think that there are a number of challenges with this emphasis on measurement and a kind of belief in certain types of measures. So I think one of the things, for instance, in the social world is that there's been a recent meteoric rise of the randomized control trial that that is the kind of ultimate gold standard and money is following the RCT. But that's quite complex, that costs around $10 million to do a sort of half decent RCT, which is almost, it was a sum beyond small organizations. And also these kinds of measures can only measure certain things. So we see an emphasis on certain kinds of intervention that are very easily measured. Some of these are good, again. Immunization is a good example. It's not that these things aren't good things, but that we are focusing very strongly on some things that can yield good data at the exclusion of other things. And as you say, not all organizations can afford to pay for this kind of gold standard. And then I think another challenge with the RCT in particular is that to get the right kind of measures you have to have very short term. Your data panels have to be very short term to make sure there's not so much variation. And one thing about human change is the long term, actually, and that humans move backwards as well as forwards and that's kind of part of the complexity. So I think it's interesting to think about what kind of different measures we can have for different contexts and what kind of new measures we might develop for what I would call a kind of complexity paradigm of human development. Is there a danger that we can draw wrong conclusions from data around, especially in the social sphere? Well, I think the actual data itself is often, I mean, so one of the things about human change is that it involves talking about things people don't want to talk about. If I ask you, for instance, are you lonely, you're not going to put in a survey that you're lonely, which is one of the reasons that this kind of global problem was overlooked for so long because there's no data on it. Nobody's going to confess that they're lonely, but this is a very big problem. On the other side, I think that, you know, we could be measuring the wrong things because it's very easy to have data around problems that you can manage. But what actually creates change might be something else. So, for instance, relationships are absolutely core to human change, to encouraging people to change, to sustaining change, and, of course, relationships are extremely hard to measure. So I think it's more that we're kind of focusing we don't have data about the things that matter. And then the other thing, I think, is that all insight and data is not the same thing, is it? And so, again, this goes back to your original question about small organizations. That they should be allowed to be radical to start pushing forward work. A lot of the measurements we have relate to old paradigms. Here, we're talking a lot about the fourth industrial revolution. The world is changing very fast, and we don't have the right kind of frameworks to measure. So a very good example of that would be health, where we have a lot of data around infectious disease, but not very much around chronic disease. So we begin to kind of sort of have measurement systems that force us to look backwards rather than forwards very often. And what we measure. I mean, in my own work, you know, I say that if traditionally we kind of work, most social science works in the kind of big part of the bell curve. And I try to work at either extreme, actually, because I think that if we can kind of design things that work for people who are at the outliers, then it will work for everybody. But again, that's a data challenge, because the data is all about kind of the middle of the bell curve. Yeah. And sometimes, and they talked about how the US should introduce new ways of measuring the economy. They said, you know, the employment statistics and the way we measure GDP were designed at the end of the 1800s. And they don't necessarily reflect the reality today. So it might show we have record low on employment, but it doesn't reflect the fact that you have this working three jobs, they don't have health insurance and they're living in the shelter because they can't afford rent. So, for instance, in the UK we have in-work poverty. We have between a third and half of British families on benefits because they're paid too low to live on. You know, their wages are too low to live on. They're not falling officially below the poverty line, but of course this creates all kind of social and political effects because people actually don't want to live on handouts. But another good example would be that in most countries in the world where people are illegal to die of aging, you have to be put into a category of some form of disease. And of course this totally defines how we look at older people, that you are a body part to be managed, not a human being to be taken care of. And increasingly we're seeing both the kind of social, cultural and financial implications of that kind of categorization. So I think we see it in the economy, we see it socially, we see it everywhere, actually. So let's examine some of the other dangers about the global risk study. And here, Eric, I'd like you to talk about the global risk study. As we all know, the World Economic Forum every year will release reports called the Global Risk Reports. For last year in 2017 it's a 13-year report. And actually this year we just based upon the big data we generated new reports for the Global Risk Reports. And the original report actually measured the five key categories. There were like 30 sub-categories. And the five categories is the politics, the geopolitics, the environment, what else, energies and those different categories. And we actually leveraged like eight millions of data, global data and the seed data. And we took advantage of the machine learnings to learn those seed data. And finally we actually used like 30 millions of data, global data to measure those categories. And we find out seven major differences from the original Global Risk Reports. So same data set, but seven major different conclusions? The original Global Risk Reports released by the WEF is based upon the human interviews. So every year WEF will organize like 700 experts, scholars from different areas to send out the interviews and to investigate those five major categories and 30 sub-categories to have the feedback and to generate that overall report. But we leveraged the big data. What I mean is data from China, the states, it's a global data, it's a multi-lingual data. We let the machine to learn how to do that. And I think when we talk about the extreme weather and how people measured about the terrorist attack and also the uncontrollable inflations. And we find out the five major differences. Which is totally different from the scholars opinion. So I think because consulting companies will release the hyper circle for the new technologies, for the new emerging technologies. And those like a hyper circle is always the same for the past 20 years. That based upon the expert's opinions they said, well, this year the blockchain is supposed to be there and the joints is going to be there. But that all based upon the science judgment. But let's say what if we are able to measure all the data seen in the past 20 years and how many people are talking about the joints in the past 10 years and how many patents, how many companies, what is the scale of the market share and how many new products emerging in that market. And actually nowadays because 10 years ago apparently we were unable to measure those data. We don't know where. But nowadays we are able to gather those data. And we are able to measure those data. And also the data is able to used in the FinTech. For example, we have the market psychology. And we measure the market sentiment to project the crude oil price and also the share price. So that's very much what you are doing as well. So maybe talk a little bit about how reliable you feel the algorithms are and how you are relying on them to make predictions and then investing accordingly. Yes, well there is a reason why we have 19 million algorithms and the reason is you want to maximize diversification because any individual algorithm is going to have a bias. Any individual algorithm will have its flaws. But different algorithms are put together by different people. Different people have different biases. So by the time you mix it all up and look at all of them in aggregate the biases gets small and the signal gets large. So this is how we deal with it. Extreme diversification. Okay. So let's talk about how focus on metrics can distort outcomes sometimes. What is your point of view on that? Yes, I mean I am concerned with the way that data can lead to different outcomes. I am more concerned with the way that certain data leads to certain kinds of programs and actions which then of course lead to different. So for instance one of the things I can give some examples. For instance in Pittsburgh there's a data system which has 131 points that decides for instance how you will behave with your children and whether you're likely to abuse your children or not. So that's, you know what you can't see in that data set is what kind of changes people are already making in their lives, what kind of broader relationships they might have around them. So this would be a very good example where you can use data to try and target families that might be in need. But what you get out of that is you get you get a data set at risk but you get no knowledge of how you might actually then move in to support those families and what are the kind of good productive things that might be happening in order to make that kind of change. So I think that that's the kind of thing that I'm concerned about. And I mean one of the things that I've been doing is trying to develop different forms of metric around people's capabilities that are both measuring kind of external and internal capability so that we can begin to kind of count different things that matter such as relationships which I've mentioned before because we can see definitely in good longitudinal databases that the strength of your relationships is highly determinant of your life but generally those are things that are very difficult to measure. And if you don't measure it it doesn't have a tension and then you don't act on it basically. Did you want to add to that? I mean I would agree I think that if you have an incomplete data set which is really what you're describing then your algorithms or any kind of conclusions you draw are less powerful. I think that's just the kind of basic fact of measurement is that if it's only half of the relevant data that you need you won't get a complete picture. And so I think this is a good segue into the broader topic of we are people are being measured every aspect. We now have algorithmic twins if you will and so everything we do is being measured in one way or another and there's been a lot of talk in the press about the Chinese social scoring system where people are rated on their civicness and there are major consequences about you know where you can apply for a job or if you can have a passport and so forth but what I think has been left out of that conversation is that we have many similar type measurements in the West that are non-transparent. People are scored on their financial responsibility they are scored by recruiters on the way they behave in social media there are many many different ways that we are being measured we have no control over them we don't know who is collecting data about us what they're doing with it and if God forbid there's a mistake made we have no way of correcting it so I think it's interesting to discuss as a society as a global society how do we confront this how do we set some sort of rules or oversight in place to make sure that there is some transparency and some control who wants to take that on well I mean one thing I would say is that you know data is like a kind of modern form of mining isn't it I mean it's highly extractive and it's not just the sort of bigger governmental systems you're talking about and the credit rating data and all of that that we also have in the West it's also that in social programs you are asked to give up a lot of your time and your data so that you can be measured but you don't own your data it's exactly the same as at the macro level so one of the things we could talk about is how this process becomes less extractive how you own your own data whether it is kind of credit data or whether it's data in the kind of work I do which is data around how your health is improving or data around how your children are doing well it's really meaningful to you and you can actually learn from by owning it so in my world kind of owning your own data not only is kind of empowering but it actually means that you learn rather than the people with the clipboards who came to interview you learn and then that kind of promotes sort of exponential kind of social change which I think is really important I would say something in addition to that I have two things I would say in addition to that one is that you know in the West there's generally an indication of data that you've put out there in the public domain like your Twitter feed or whatever it is that a recruiter may or may not be using and again I'm not saying it's a good thing I'm just saying you have put it out in the public domain versus your health information which is highly protected by a lot of regulation so there is that difference and so we should just be mindful of what we're talking about when we talk about data because that could mean a lot of different things and then secondly I think that again this kind of gets into where political institutions intersect with thoughts around what you do with data right I mean in the EU all of the GDPR and data privacy laws that have come into effect really are a response to this idea that data should be owned by the individual and you know as we all know there have been a lot of changes that technology companies small like pie metrics large like Google have had to make as a result of that right so again I think people are responding to these challenges in different ways and I think it does reflect the political climate of different geographies and sort of not just political but the social norms of different countries and I think it kind of remains to be seen I think where this all nets out I think Do you want to answer that Eric? Yeah I think we need a bit more time in the perspective of the data the artificial intelligence and the big data need a bit of time and the companies or enterprises need more time and also the government need a bit of time because that could be a kind of the joint efforts because as we all know that artificial intelligence is booming it could be like last three to five years and when we talk real big data it's maybe in the past five years and due to the chips and the memories calculating computing capabilities but those are the new emerging problems especially for the GDPR released by the European unions but I think for example like the government need to have overall like management for the data privacy especially just now I mentioned medical imaging, healthcare data and also like the news social media the same and for the company the companies need to have common consensus on data privacy protections and also need a bit of joint efforts to the whole process and also then need to be like a the mechanism to coordinate because I think it all comes back to what Hillary said earlier about it depends on how the technology is used so I don't know how many of you have noticed but across the hall there's a demonstration of voice recognition technology that Carnegie Mellon University is working on and the voice recognition technology could play an important beneficial role in the sense that it can detect from your voice certain health issues and so the idea is that you could use this voice recognition technology along with telemedicine to give people early alerts about medical conditions and potentially save lives but if this was used in a different way I mean I don't know how many of you have done it I tested it myself this afternoon but you stand in front of this machine and they have you read like a paragraph and then immediately afterwards within seconds it tells you about what you look like guesses your age can measure different things about your health your mental state at the moment about your personality whether you have leadership potential in a job so you could imagine this being used in different circumstances maybe for recruitment or something else that could be slightly disturbing and especially since at least for now it's not quite ready for prime time and I can attest to that because it told me that I was a man that I'm 20 years younger than I am that I'm very tall and big bones so so there as much faith as we put in technology and as wonderful as some of the predictions now are on global risks or things that might impact the economy we aren't quite there with all the different technologies yet and there are some things that we have to pay attention to some dangers so that I think is a good segue into the next question which is to what extent do the panelists believe that we must marry human judgment with technology in order to have the right context and the best outcomes I think you all certainly that is always going I shouldn't say always for the foreseeable future is going to be the case right I mean a lot of what we're talking about here is not general artificial intelligence it's narrow artificial intelligence which will always be applied by human for a specific use case and usually in combination with some form of human judgment now when general artificial intelligence comes along and god only knows when that happens or what happens when that happens that's potentially a world changing event but until then I think these are all narrow specific use cases that will always be deployed with the context of human judgment that's kind of my take on it I am very interested in how we develop what I would call like a complexity paradigm I mean I would say that the algorithm is a Newtonian paradigm which is great when things are objectifiable linear but human change is not but then the flip side of that and the reason I for one am so interested in measurement is that we have to have rigor like we do have to know what works and we do have to kind of exclude as much bias as possible so I think we need to kind of balance data with human judgment and what would be wonderful would be if we could also put as much energy into the kind of sophistication of thinking of what those methods of human judgment are which you know there's some really exciting developments around participatory methods where people kind of cross check each other for example ground truthing where you spend time looking at things from different perspectives but I think it would be really exciting to develop those kinds of human judgment systems and think about what rigor is in those systems and sit them alongside more quantifiable data systems I think the only other thing I would say is that I think sometimes you really exalt human judgment to be so much better somehow you know impervious to all these problems that we see in algorithms and I would just argue that there are many examples and I won't mention any specific ones but I'm sure we can all think of many examples for human judgment has been extremely flawed leading to very catastrophic events right so I think we just have to be careful when we think about one being you know dangerous and potentially the other being a salve to that but do you think that's because we're limited around humans and we kind of understand bias and we understand cultural judgment and the problem with our kind of sort of almost awe of data because it's so recent is that we don't understand how it's culture embedded and we don't understand how it's biased enough and so that's the problem isn't it's taking that kind of critical perspective now to the data that we might have about human judgment because I agree with you. I think that whenever a new technology comes around people think it's going to end the world you know before I became an entrepreneur and you know when brain imaging came around everyone thought that oh my goodness you're going to scan somebody's brain at birth and be able to like predict something about them and it's going to become this gattaca like future and now the same thing is being thought about with artificial intelligence so I think you know change provokes fear in human beings which is natural and we always fear the worst and then you know and we also dream that these technologies are going to just create this utopian future and in the end it's like a disappointing middle ground where you know things get better but they're not quite as scary as we think they'll be and they're not quite as awesome as we think they'll be because I think the last time the kind of data cracks were powerful was in the last industrial revolution when again there was a similar kind of like bubble about data and you know sort of encyclopedias and collecting and making lists and then people saw the problems and but yeah I agree what's your perspective on this machines will never ever displace humans because one primary reason is that humans and machines are different and anytime you put two different things together you get something better so it's simply not ever going to happen okay and so how does it work in your firm the human algorithm balance you know we have some algorithms made by humans we have some algorithms made by machines we have measures I can say that at this point in time the algorithms made by humans are about 50 times better than the one made by machines but where machines win is the sheer quantity and brute force and the breadth of the approach you know machines are just that machines machines make predictions humans make the judgment as machines bite off a little bit more of what the humans do the human judgment level keeps escalating and going up and that's it's a never ending spiral okay Eric yeah I think it's a process of the convergence and definitely it's a combination of the human efforts and the machine findings I'd like to take the machine translation as an example because like 15 years ago I was a simultaneous interpreter and right now we developed the machine translation and in different like a conference I would say I would say that definitely nowadays the machine translation are able to replace the human translator but at the very beginning you know two or three years ago no one believed it and even when I was telling the stories in the in the school of the translation so no one chose to believe it but the things is that as we all know the Oxford dictionary only have 80,000 words at the machine translation our machine translation we have five billions high quality sentence pair and also people will argue that what if we translate the literature there and I'm telling the truth that what if we put the high quality translation versions into the language model out there whenever the machine translation came across that paragraph or that article it generated the best one it definitely much much better than any one of the translator out there but the truth is that nowadays the machine translation is not replaced the human translator as the machine translation touch upon it definitely the data the information or knowledge never ever been touched by human translators because nowadays the four machine translation for one seconds are able to translate like a 16 sorry 16,000 words for one second can you imagine the human efforts and also another example for the paradox of medical imaging the same even nowadays the medical imaging diagnosis system are able to reach like 99.2% accuracy but only like 1% mistakes could not be tolerated in any circumstances but it's definitely the combination the human doctors even nowadays the human doctors accuracy is only like an average one accuracy is like 70% globally but definitely is a combination what do people in the audience think about this what are some of your concerns what would you like to ask the panelists who wants to start over here I'm from Argentina I run a translation start up no it's fine we are moving into using the data that we have as translators have aligned all these phrases into using that information to position companies abroad because of the use of their own technical language and technical keywords that we are able to identify in different texts and about that combination of human and machine I think about the data that we are talking about a lot about the amount of data but not the quality of the data and it's really important to understand that supervision on machine learning for us worked very well and I don't understand why it's not being used much more what is your take on this I remember mechanical having huge crowdsources of people for the supervision of the results of the machine and I think that should be used a little bit more to make it faster we didn't have enough data and with the supervised information we achieved a quality of results that we were not going to be able to achieve with the amount of clients or volume that we managed what do you think about that actually nowadays when we talk about the big data the big data has two sides because when I think starting from 2016 the neural machine translation came out released by the Google previously it was SMT statistical machine translation at that time the language models need to be trained by at least 100 million but nowadays with the help of the neural machine translation in order to achieve the same quality of the translation quality of the machine translation engines only 10 million high quality is sufficient so I think it's definitely the balance the high quality high quality of the data set no matter the language corpus or the medical images the same type of the algorithm and before the algorithm we need a human doctor I mean the human expert to set up the golden rules for example like medical images and we definitely need a human doctor's expert to to tell the golden rules to diagnosis lung cancer and then we ask the doctors to take each piece of the images out there and tell them the symptoms of those images but on the other side on the other hand the machine definitely have the absolute advantages compared to the human human beings absolutely Viola, I see you shaking your head but do you want to say and please introduce yourself Sure, good job Jennifer and my name is Viola Lualan I'm the co-founder of Avamba Solutions so fintech platform that was created for African realities but one of the few companies in the world that has decided against some western sentiments to involve tribal data into our risk models and built an algorithm the common and concern is that at the beginning your comment about the reliance on big data puts us in a very strange twilight whereby some western investors who use data predominantly to make decisions challenge the fact that where we exist with no credit databases in much of sub-Saharan Africa we've actually had to take the time to send human beings to walk around to speak to different types of people to collect information to turn into data using A.B. testing machine learning algorithms and testing it with our own internal funding and have produced results that are quite indicative of what will we think is the next way in which to fund African SMEs do you ever run into these situations where the emerging markets have to use totally different approaches that are not classic by western standards where you provide the same respect for the veracity of that data and the way it's being used which is very different to what we see in other places Great question Hilary Which Hilary I actually we don't work much in that geography so I'll turn it over to anyone that does have experience I don't have that experience in Africa but I would say that a similar experience would be like 10 years ago saying that we should work on an issue like loneliness where there's no data but where you see by spending a lot of time living alongside people or knowing a community that this is a kind of huge issue that somehow isn't showing up in the data Another issue that I've worked with is in Latin America where children are not registering for school and this is perceived to be a problem that they can't afford school uniform shame because you don't want to register your children if they're illegitimate, they don't have an identity card and they can't go to school So I think socially I see and I think this is a kind of huge issue and ultimately although you can collect different data which is what we do it is then a job of politically lobbying to try and get this on the agenda to collect data in a traditional way around the subject matter but I think it's a very big issue The lack of data and also from what I've read there's also cases of skewing of data in emerging markets where governments will release misleading or even a erroneous information and then that is cited as gospel by everyone else and that's a big problem too so Over here There will be a lot of good before the bad but do you see any scenario in the next 5 to 10 years where this divide that we see both social and economic is actually going to reduce because I keep worrying as human beings first we outsource the memory part of technology then we outsource the processing part and now we outsourcing the decision making part so it kind of begs the question what's the relevance of humanity and if technology has taught us anything it multiplies faster than we can predict I mean in terms of everything it's getting faster, it's getting smaller, it's getting cheaper so none of us in this room can actually predict how fast AI is going to be you know evolved or embraced or anything like that so Amazon trillion dollar Apple trillion dollar like any country you can see the divide is just becoming widened so do you foresee any scenario that this divide is going to get smaller and the benefits of technology actually can reach more people than it is currently just curious would you like to take that? Yeah I can take that I can tell you that you know I've heard this point a number of times but in our company we have been automating we automate things night and day and there's been no job reduction as a matter of fact as you plot the headcount versus what we can do with time we just keeps growing and growing so automation begets more algorithms more people and it's just the things that we need to do keep rising to the next level and it's a level that we can't often foresee at a given point in time I will get to you in just one second I think that your question is a very important one because if I can just add to your comments we look at what algorithms can do to help improve health the kind of thing that you're doing and we get closer to precision medicine where we can really you know treat disease in super effective way we as a society will then have to ask the question well how are we going to pay for this is this going to be available to everyone and that's a question that we have to be addressed seriously how do we make sure the benefits of this technology are distributed evenly and that is not clear today yes over here I actually wanted to speak to the previous question and then ask a question because I had a question before too but actually I am working in the area of inequality and I did want to say actually you're wrong that it isn't in every single country getting bigger globally inequality has been reducing dramatically and that's not only because of the success of China and of India but in Africa too most countries not all but most of the countries inequality is reducing and it is because of technology it is because mobile phones mostly a little bit of the AI but just getting the ICT out to the people and the incredible innovative things that are happening so extreme poverty has radically reduced I mean there are still some people in extreme poverty but it's like 90% reduced it's amazing what's been done in the last 15 years and there's a great credit to the developing world there so it's not all gloom and doom although I agree and I also am working on the very high inequality in the OECD is a problem and it leads to great instability on the global level so I agree with that however that's not the question I was going to ask that when I it was a little bit of the earlier discussion about data and I'm not an expert in data but I know something about algorithms though and I know that one of the the famous bias cases which is supposedly based on data which is about recidivism in America no academic can do as bad as the commercial program that these that the courts were using so in this case you have to say what were these guys doing and were they even really using data and so when you talk about the correct use of data it seems to me that accountability is a big part how do we even know if they're using the data they say they're using I think that there is increasing awareness of the issue of bias in data I mean Kathy O'Neill wrote the book weapons of math destruction I agree but I guess what I'm trying to say is that the first thing we need is awareness of bias in whatever algorithm you're using whether it's in the data or somehow the instantiation of the algorithm and then I think beyond that there are conversations around creating some sort of I don't know if it's going to be an ethics committee or if it's going to be some sort of I don't really know but there's talk in the air of should there be, will there be some sort of a entity that looks at these types of issues I'm not sure that's going to happen if there will be the will to make that happen or if it will be sort of a self-regulated situation but I certainly think that at a bare minimum awareness on the topic is I think at an all-time high and continuing to increase and I think that that's a step in the right direction that's all that I would say and I mean since the question had to do with recidivism you and I were talking last night and you were giving a great example of how data can actually aid in making the case for great social projects and I think that would be a good thing to add. Yeah so I was talking about this I think it's called the Bail project if anyone's ever heard of it it was started by a public defender in Brooklyn and her husband who basically put together a $10,000 slush fund to essentially pay for people's bail because they realized that you know even if you didn't have $500 bail you would essentially stay in jail for a week or however many days and you would you know lose your kids potentially, lose your job a bunch of different things right and so what they did was they amassed all this data around if you actually just loaned people the money what would happen right so the repayment of these loans which you didn't have to repay them you could have just gotten out of jail and then you know gone off was 96% right people were skeptical that it would reach any high number it was 96% and then the most shocking thing was that if you didn't have money to pay your bail I think the rate at which you had a criminal record was like 88% or 90% right because people just pleaded to whatever it was just to get out of jail so they could go back to their job and their kids and everything else if you gave them money to get out the rate of getting charged with the crime was like 2% I mean and again this was a subset I'm sure of all of the people that were out you know that were in there on bail it wasn't you know probably the most heinous offenders I'm sure it was like riding your bike on a sidewalk or whatever it was right but it just speaks to again it's just how humans use data that was obviously an example of somebody who had amassed a great data set to show the power of a very simple intervention around providing a free loan to get out of to pay your bail and it was the data that made the story more compelling yep so we can take one really quick question because then we can have to wrap up yeah actually my question would be for the Fintech area especially the data from China and because my company I'm running a Fintech company for the quantitative trading system provided to the financial institutions in China especially for the to the work quant and founder and I just wanted to know because for the financial data and especially for the work only being over wide like for the worldwide for the investment but for the Chinese data besides the market data and what do you think about the Chinese data provider like are those like Chinese data provider for example WIND which is the official data provider are they sufficient for you to make an investment decision because like we well I used to in the overseas investment banking we create this company for like a facility for Chinese institution but we realized the data the standard of the data and it's not as good as the overseas I would say the European or U.S. standard and it will affect the investment decisions especially for the quant for algo trading system so this is for address like what kind of data you use then we have to wrap up the panel I would say that there is a hierarchy in the quality of data the best quality of data for now tends to be in the U.S. and then it goes goes down from there and that's just the way it's been structurally for historical reasons and so on we are getting more and more data in China but it's a difficult process because the contracts have to be in Chinese and there are a lot of business and legal hurdles to getting it. Thank you. I think we've done a good job talking about the power of big data but also some of the questions that need to be addressed and I did want to mention before we finish that the World Economic Forum is creating a Global Futures Council on new metrics which will be launched in November and so the conversation is sure to continue. I'd like to thank our panelists and also the audience for participating. Thank you very much.