Data Literacy for Managers

Speaker 1 [00:04]

Welcome everyone. Thanks for the introduction. This is actually the name of the company. I'm one of the partners. We also have a booth. Come by. first a little bit of background how did this talk come about so as Christian mentioned I'm around at many conferences so I talk to many data scientists I talk to many experts I think this year with the calls for free conferences I read about like a thousand proposals for conference talks this gives me one good perspective on the expert side but when we work with clients I also hear their side and where they are struggling and what they think they should do and not and this is the perspectives I want to share and my first question is who's data scientist or an expert in Python okay and who's more like in the management or like organizing project management position okay welcome you all it's yeah and I see We very often start, let's first start with, yeah, where would you stand? Yeah, so let's start with the data scientist perspective and let's start with a happy data scientist perspective. So if I speak to a happy data scientist, they say, my work is appreciated. I can live up to my full potential. I'm highly trained. Very likely you're a curious person who wants to provide value to your company and solve problems, improve, and have a high standard also in ethics and everything. But I also talk to many data scientists who are not so happy. They work at a big company and they feel limited to say, I'm a data scientist. I can do deep learning. I can do machine learning. I have many ideas, but actually I'm tied up. I'm working as an analyst, and they're not so happy and torn because they're probably not heard. Since I expected to have more experts in the room, I want to have a little bit broader perspective on what is probably the manager's perspective here. So actually, I found something online, how managers spend their time. So, of course, a lot of it is administration, solving problems, collaborating, strategy and innovation. It's only 10% and developing people and energy and statehood is 7%. So, you see they're busy with many other things where you maybe say I'm doing code review or solving bugs or so. This is probably their part here. um and also quite interestingly uh there was also a survey or how which managers think which skill they will need to success in the next five years and actually i was quite surprised the social the people skills uh in my opinion very underrated because i think the biggest problem in all technology we solve today basically i think the biggest problem is actually human interaction communication and everybody has the same understanding how to solve the problems and where we really want to go i think that's the main challenge uh while of course technology is moving at light speed so okay how if we google managers so i googled oh you cannot read it's um it's uh actually it's uh yeah i googled ai for managers we get like the typical pictures my favorite white android robot this is like uh like the quite false perception of ai but it's pretty sad in culture nowadays but i also googled replacing managers and this is the other perspective uh this is the first hit with ai technology replacing managers so but you know there's also tools like auto ml and stuff like that so what about data scientists replacing data scientists, will AI? It's still a question. It's not a proposal yet according to Google. And if you're a data engineer, you're like super lucky. Yes. But we see there's a lot of buzzwords, the hype promises around and I think people, many people feel like this, like it's like a crazy party, nobody knows we're actually moving, but we need to do something. So let me tell you a story. I recently met an old friend at the train station so the person was very excited to tell me about coming back to the hometown and have a new position in like top level management and we talked hey what are you doing now and I said yeah I'm a partner with people and we run Koenigsweg as a data science consultancy we advise people on data big data and ai and said oh great ai in my new position i have to buy ai as well and i was just like okay context very skilled i it's like a highly intelligent person it's not a dummy uh like and then i just like thought okay how do i imagine I have to buy AI. So is it Aufschnitt oder ein Stück? Yeah, cut a piece. So how do you buy AI in 2014? But we talked longer and our company also understood what the real problem was in the company, which had to be addressed. And of course we solved it. And I want to share some of how we solved it because a big part is actually understanding what the problem is, what the challenges are, and get a better and honest perspective on things and how to operate. So let's start with the definition. Data literacy is the ability to collect, manage, evaluate, and apply data in a critical manner. So if you're a data scientist, you're an expert. This is basically all you know. You know tons of stuff. You go to conferences. You have a lot of input. You're a creative person. You have many ideas how to solve. And very likely, if I talk to myself, I like to solve problems, actually. So, I know a lot. And, of course, I always try to be transparent with my work and my approaches. So, I like to share, okay, this is the background, this is how it works. This is probably also the reason I give many talks. But also, if you talk to your peers, always think. Don't try to explain everything. Just explain what they need to know. remember keep it to the point you know this if you're from germany you know these two guys these two like the mouse and the elephant they'll explain it in a simple way and let me take you to data here there's like four levels of data we have data collection we have to manage the data we have data evaluation and then we finally have the application which is like the nice part with all the machine learning everything so now let's just like walk through here what has somebody in management or on a decision maker have to understand actually about this does he or he or she or them have to understand actually data collection no everything yeah the data is collected great it works that's all they have to care about what about so no so what about data management people work in enterprises every other job usually you work with experts and people who know how to do things of course data management can be complicated but if you know the data is being managed properly in a good fashion and redundant fashion and all you need depending on your needs is managed that's it that's the key message yeah it's been managed now if we come to data evaluation it's not that simple data evaluation basically we can pin it down like to these six parts we have data tools we have data analysis we have interpretation and understanding we have data visualization we have presenting data in an oral way so this is not the same as the visualization and we have data driven decision making and which parts actually somebody has to understand better to make a decision i think the data tools the tools don't matter it's they matter to us we want to have great tools when we program and solve problems but it's it's irrelevant which tools actually are used to solve things we have data analysis We need to have some knowledge about this. Very important is interpretation and understanding of the data. What can we do with it? Data visualization is always nice, but actually it's more important how can we present the data in like a presentation? How can I explain it to somebody just maybe in a meeting or a phone call without Seaborn and all the cool tools we have? and also it's also of course important how to make a data-driven the data-driven decision-making process so so basically we have a mixture but we see basically data evaluation is the first part we have to dig a little deeper in some ways and what about data application this is basically the most talked i mean this is the most talked part data application and this is where all the work is actually what most of the work is um let's look here we have data ethics data citation data citation if you don't know it is like if you cite from for example scientific data from somewhere else the the data culture data sharing this is more like sharing data between your peers critical thinking and the elevating process this is evaluating the decisions and if you look here of course data is super important you have to understand how you work with the data um various data sets i think we have a lot of the heard about us at the conference and previous conference also in the keynote yesterday um it's important but maybe not the most important data culture is important critical thinking is actually what people trust the experts to do the critical thinking so you have to know there's critical thinking involved but that's the short message here so that's what needs to be known data citation is irrelevant the data is just like they're also like the sharing how you share the data is not relevant for the decision maker and but the evaluating decisions how but that's basically the outcome that's the predictions and everything of course this has to be understood deeply but you see tackling down maybe from So everything I put here on the slides, maybe you have to get other people in your team or decision maker maybe only on board with like 10, 20%. Okay, so here we go. Just in short, I just want to point to this. Also, if you talk in teams or companies or with clients, there are different personalities in the field. And I think Peter Baumgartner, he works in NLP. He gave a talk at Spacey IRL earlier this year here in Berlin. it's referred he narrowed it down to these six personalities you can also run while running projects with them and I just want to show like for example let's show off Sarah who just wants to do AI he stereotyped it a little in a very funny way just look it up later I think this is very useful input because there's these different types of interests here and I think he narrowed it down really good of course the decision maker also has to know it's not the IT department you have to talk to. We are not doing an IT project. The IT department of course will operate everything we produce but at first data science and AI is research and development. Once we go to production we have to involve the IT department or to get resources to do the research but we need to have an open space and an open mind to address the problems, to bring the value to the clients or to the company you work with and uh and this is very often overlooked they see okay yeah okay it's data it's programming okay let's talk to the it department the department in many big enterprises the it department is actually just like mostly configuring tools from other companies it's not they're probably not even super capable of programming uh maybe they have just some basic skills uh but basically they operate more um some misperceptions one misperceptions also are bigger is better. So I just buy the Hadoop cluster and then I'm set up for the next 10 years and I bought also like a Let's Buy Watson and we set up with all the knowledge but that's also say okay, we don't need the big resources we should invest probably in other areas. And another misperception is like a data lake because this is also like nice pictures, data lake, so we're like nice clear lake. It's not. A data lake for example is a complex system which involves many other systems. Basically, it's more like a thought concept than a nice lake you can just pull the data out. So these are all things you have to point out. Very often, I see ourselves confronted that the problem seems just to be simple to the peer we talk to or the person we talk to because there's no experience. So you hear all the hype and it's easy. We just throw deep learning at it and we'll solve it. And we have to point out, no, we should use classical statistical methods first because they're stable, they're proven, and we can explain the outcome. Deep learning, we cannot. But this is very often. So this is very often you need to stand up for this and say, no, we have simpler methods to use this. We don't need the latest and fanciest tech just because we read about it. Of course, it's very important to look into this. But it's not the first thing to look at. and the other thing is also like senior management maybe say oh we have all seen this before it's just like a hype and actually they're right i did some research uh i have an mit technology review has all past issues back to the 1895 i guess online in the archive of the pdf if you're subscriber so i actually for a talk i never did i had another project but like uh but i saw all the while i was looking at the old issues i saw stuff like this so in 1998 we have can computers create literature okay oh i just gave a talk last year at pi data berlin about text literature like not literature but producing text so they are and yeah and i wasn't that successful um what about But beyond artificial intelligence, one year later, in 1987, we see how to keep mature industries innovative. This sounds very much like electrifying automotive nowadays. And also other issues like automation. In 1985, there was also a lot of interest in the industries. There was a lot of automation going on. Also like agile, lean management, caban. This goes back to the 80s. not invented by computer or like IT and tech. Actually, this comes from production industry. We just think we're like super cool because we know agile and lean. No, it's old concepts. And also like this, I thought it was really funny and from 1986, after 25 years, artificial intelligence has failed to live up to its promise. Okay. But also since 1988, we know, don't believe the hype. but I think we have to look at the pattern here because with AI, AI history is all 70 years, so we have to see there's a constant progress and challenges and things to solve so we see there's progress here, so it's probably you also see sometimes wave, golden age, dark wave, ups and downs, so maybe deep learning will hit another bound soon, we don't know maybe not, that's nice recently I saw Kasparov when I grew up Kasparov was a rock star Kasparov was the best chess player in the world so and he gave a talk recently in Lausanne and he was on stage and he actually he played about against the computer and all these people here who basically built the theoretical groundwork for what we work on computers today said once a machine can beat a human in chess they take over they have won so okay casparo he was the brain's last stand in the 90s i remember that we still live computers still have taken over the world i'm not saying uh we see also like progress here so of course uh all the stuff from the deep blinds is is better than the players here um it's it's quite quite interesting uh but i think this is the best point he made actually he was citing Pablo Picasso because he said computers are useless they only give us answers and let me also introduce you how we basically address like getting literacy a little bit better in the head of clients and decision makers so this is now about demos and magic and one really good approach from GoDaddy Driven actually was I stumbled across you know Vincent He also just gave a talk earlier today. He's from Amsterdam, and they did a game where basically the data scientists have to play and create value for the company. There's a blog post about it. Just read about it. We did Jupyter Notebooks with the Titanic. Titanic is super boring for everybody who's here in data science, computer science. But think Titanic is still super instructive if you are new in the field or if you're just curious how this works. So I put an executive summary here. Tidetic data is instructive in sample. You can learn about how to make predictions, how to make bad predictions, like for example, hey, all the women die, all the men, no, all the women survive, all the men die, we already have 80% accuracy. We all know that. If you're not, you know now. And this is very instructive about domain expertise and it's a really nice example. People can just, everybody understands Titanic. And so it's very nice to also present this in just like one hour with not too much code to management people. um or something like this upon a time there was a little mermaid named siren who lived with her stepmother under the sea she didn't get to go out of the sea like any other okay this is a synthesized voice i was was trained on tacotron too and uh what's the main takeaway here it's not what we can do is be synthesis nowadays no the main takeaway is i'm trained it on this macbook in the egpu in nine days of training it can read english text without cloud service it's probably not the super production ready but this is like come on this is this is no budget for some output like this uh it's not a million budget my macbook is yeah maybe well s plus the time and then also you can explain this is scalable and you have an instant application like reading text and many people don't know this they think oh it takes all a long time uh yeah this is what i would skip this is just like this is the network i would explain it to you but the executive here the summary here is high quality data plus a cleverly designed neural net produces this outcome at a very reasonable budget the same or similar i won't go deep here is like data signal and predictions so actually did you know like style transfers this is actually just like a prediction of this picture using this style it's a convolutional network you can just like we simplify the data and apply another style um i have to speed up here a bit so this is like like a short video how a network trains but it's very instructive to see people oh okay this is how the network is learning and how it's flickering you see we're just like evening things out it's people can feel how it works people don't want to read math or hear too much powerpoint stuff and you can explain things take longer this is just like some examples say hey all fancy style transfer all x everything in my slides is that transfer fancy fancy fancy but what i also like to show people it does not really work all the time so after it's like it's just like a picture from the beach but you always see like this hole here this is not working well we can reason about why so it's a very good explanation people can see you just not throw ai at it and it will always work um because one great thing is we have a lot of open research papers plus the internet plus the news so there's at least one news about oh some researchers somewhere found this you can do this with ai and will change the world. So, but the summary here is papers demonstrate impressive findings, but following progress is important, but only 95% of the research actually survives in production. There are some other reasons. So it's not all you read about will actually work out in the field. And also people should be aware about adversarial attacks because the network is not just like better, like a perfect machine. Actually, not so perfect. This is like a really nice example how some researchers fooled Yulu here. So because just like of having this picture in front, he's not recognized as a person. Where we actually see, we think like object detections and identifying people is basically really soft and scalable. But it's not. It's like just take this picture and you go look it up in video. So the summary here is new networks are not perceiving reality as we humans do. Newer networks may be fooled, and they are better and faster at many things, but they are not really good at transferring knowledge as we do. So I think to actually solve it and how to get it going, I think, to sum it up, I think it's important to team up, to establish an open, honest, and respectful communication culture. So I think, I very often now tell the clients, come to a Python conference, see the communication culture, the respectful communication culture we have established, because the Python community, I think one thing it makes us great is you can talk to an astronomer, you can talk to a web developer. We have many different perspectives to look at things. And I very often reach out to astronomers. For example, if I have really tricky questions, can you maybe have a solution? Because astronomers deal with big data for a very, very long time. It's not just the internet. it it's it's very open to and don't don't expect to this to work overnight with one meeting it's all about trust it's respect very often i see a physicist or somebody at a client and i probably know we should improve the coding skills or thing but take take the time and take the space to communicate it and then for example part of my job or our offering is like to train enable people how bring how can we bring ideas to production how can we experiment how can we make all this fly and uh yeah if you like the pictures in these slides because they were all made with style transfer that's why it's really easy to take all the pop culture pictures just throw style transfer at it and it's here so also we had some happy people here you our booth is just in the main hall we have a photo box you can just like go there and take a picture send it to yourself have fun with friends and yeah thank you very much we do have time for questions rachel uh thanks for your talk what types of um resources or additional learning do you generally suggest to managers that you work with if people want to learn more i think i think it's very important to give them an oversight this is the technology this is where we come from also like to like to give them better definitions because for example there's big data what is actually big data many people think with two gigabytes this is big data you say no it's not well try to not to put it in figures just like I'm always trying to always like establish other pictures so big data is you basically if the handling of the data is a problem before if you don't have if you can save it on a drive or in your center it's not a problem I think it's very good to give them an overview and always for example if you prepare a training for this we always keep the schedule flexible because you never know what the real interest is because i always we always ask hey what are your real problems you want to solve where are your pains which where's uh where how can we help basically so and then usually during the training the entrance move a bit and then we go a little bit more into the machine learning space or we address deep learning or we very often say okay we can just solve it the stats actually and and very often we see okay the suggestion is very often okay you have to establish a better communication culture you you're facing data silos and your company and we we show how we can solve it with open source so we always try to push for open source solutions as well which is in Germany not that simple because it's still a new thing for many enterprises. Yeah, that's how we tackle it.

Speaker 2 [26:36]

Thank you for the talk. So we experienced that often there's a translation problem between data scientists and managers, where data scientists obviously know a lot of the data techniques, but managers have a difficult time coming up with how data science can help them in their daily work. And then they come up with things like, what is the revenue of this customer in two years if Brexit comes? Yes or no, right?

Speaker 1 [27:00]

how do

Speaker 2 [27:02]

ideas that are just not feasible and so we're thinking about educating them and we don't have such a good experience with with your idea of giving them the Titanic data set because if you make it easy enough and and and then they they just have a good time and I think all this is cool but they still don't understand what data science can't do which I think is often so I would like I'd like to have your opinion on educating them more about the concepts and the structure, but don't really do hands-on data science.

Speaker 1 [27:40]

Actually, we don't. It depends. It depends on the team and how we evaluate first who are our peers.

Speaker 2 [27:41]

No.

Speaker 1 [27:48]

Is somebody with programming skills? No programming skills. We do no programming. And I was simplifying with the Titanic sampler before. Here, the main message was, okay, it's boring to us, but probably not to other people. But of course, and the rest is basically flexible. It depends on the group we work with. So sometimes you just need to explain the Titanic example. and you see, okay, there's other opportunities, but very often we come back, okay, how's your data pipeline? How's your data stored? Is the data even accessible, which is like, I would say, the majority of the cases, yeah, we have all these ideas in our head, but we cannot make it happen before we have data pipelines, production ready and accessible in the company. But usually we don't try to go too deep into tech and code because many people are just like afraid of code because we are used, like even you can like think about just show a JSON file, which is like super simple data structure. Just show it to your parents and they will say, well, what is this? This is just like gibberish. This is just like random characters on an empty page. We very often forget because we scan these patterns, JSON code, and it says something to us. And even very good code is like just gibberish if you're not used to it. We have time for one last question. Hi, thanks for the talk. It's great. I recognize a lot of the things you mentioned. I wonder, do you have any experience with, like, so we are in data science, so what we can do is for those businesses make predictions and all that kind of things. I always have a very hard time explaining that many of the things we find are not causal relationships. However, as a follow-up, they do want to take action using the models. Okay. and they always hope to see like okay if we can like tweak this button then this will happen but often the data does not allow you to make these kind of conclusions so do you have any experience with that actually it's we try to it's always like it's how can we say it's like a balance because very often we are faced with clients as well they have big expectations of course as it does if they hire us to help them build the models and get work of course we want to deliver but sometimes there's nothing to deliver it's just like not enough the data quality is not good enough it's not we cannot really say and we just like we are just honest and I think another example can be so many people I want to use the latest tech so for example yeah we need a kubernetes cluster then yeah we just like prototyping where's the kubernetes cluster for me if we go to a production and if there's many many machines we probably will use it and sometimes it's not easy even for me as a senior person to stand up because of course you want to also win the client you want to work with the people and you want the business but on the other hand we'd rather say okay no we say this is how we do it this is how our perspective you don't need this it becomes much smaller we are your advisors and yeah and very often to honesty is trust building and then you can have to follow up business as well and if for us that's basically in how we see also our businesses. I mean, if the client just wants to have the hype and have some results to present something, it's probably not a good fit for us. Then we'd rather not work with the client because our mission is to enable them. That's how we see it. And then you sometimes have to say no, although even if it's hard. Thank you, Alexander. Thank you very much.