Do we really need Data Scientists?

Speaker 1 [00:03]

I'm Setare, and I am working as director of data analytics and data science at Deconium. And, I mean, you can check on LinkedIn and stuff, like, that I won't waste time on that. But Deconium is actually an IT company working, like, it's owned by Volkswagen. So most of the projects that we are doing is with Falvey or Volkswagen, in case you are interested to know about it. Yeah, so I actually gonna talk about do we really need data scientists today, which actually I hope that I can answer that question during this 20-25 minutes that I'm going to talk. But first I will talk about actually what is the problem and why am I actually going to talk about this question and where it came from. And then I would like to give a little bit of history of the data science as a job, where it came from and how it emerged and the background of it. Because I think it's actually helpful to understand a little bit what is data science, where it came from, and what do we need to expect from data scientists. And also at the end I would like to talk about what is data science really and do we really need them. And yeah, so let's just start. So as I said, I will talk about what is a problem. So the problem is actually that in the last two years, let's say, that there are a lot of online articles that I have seen, and many of you probably also might have seen, that they predict actually the data science job. And they talk about the future of data science, is it going to vanish or not, or is it dying out and for example I put now a screenshot of one of them here on the left says why machine learning engineers are replacing data scientists but there are actually a lot of examples of them why she should become a data engineer and not a data scientist or as I said like actually many other as well this third one actually yeah I don't understand I mean I I read the article, I understand the content of it, but data science feels like a fake entrepreneur coach on YouTube telling me that the future will be bright as long as... I don't know. If you understand it, come to me later, explain me, please. But anyhow, as I said, there are a lot of titles like, will data science become absolute in the next 10 years? Is data science dead in 10 years or not? So that is actually why I decided to talk about it today in this conference, because I also assume that many people who attend the conference are working either as data scientists or close to data scientists, and that's actually the question that's going around. I actually read those articles and tried to summarize the arguments, and I could summarize them in three categories. I think a lot of people in these articles, actually, they gave the argument about automated machine learning so that they think data science and machine learning tasks are now becoming automated and there would be no need for data scientists, but more for ML ops or data engineers. Another category of the arguments is that oversaturation of job market, that job market is flooded by the people changing their job from other background or jobs to data science, but also in the past years, the graduates of data science, graduates like from bachelor master also entered the job market and that is also one one argument about saturation of the job market the third category is that there is something greater which is happening ai which we have been heard about it for many decades actually like um that it is happening and now we actually need more application of ai in a set of data science so i would like you actually to to have these three categories of arguments in your head because i come back or refer back to them as well in the talk so these are the arguments and to answer actually to these arguments and to think if they're correct or not i decided to put a little bit of history into our actually perspective and specifically i mean i think that history is in general very important about anything any job or anything actually when we are trying to analyze it but specifically for data science I think it is very important because of that the job is actually not very old so looking at the history of it is not like looking I don't know how many years ago it's just like couple of decades that we have this job so I started actually since actually from 1960s when the power of computers started to become like changing and like it was a rapid like like, involvement, actually, in computer's power, or computational power. And also it was in 1962 that J. W. Tukey wrote about, like, an article, the future of data analysis. So it was actually the time that the whole shift for mathematicians and statisticians happened, like, on data analysis. And I actually really loved that photo on the left because of many reasons. I mean, this photo is like showing us from 1962 and it's showing four boards of like parts of like four early computers from basically left to right. It's like the four generations of the NEC that the size of them getting smaller, the computer boards. But there's actually another thing interesting in that photo for me is also, of course, that A couple of decades ago, the people who have been holding the boards, it was a very female job, working with computers, and it's very interesting how in a short time it changed when the money and power started to come to the field of IT and computer, that it became a male job, but it actually started differently. This is not the topic of my talk, but I found this photo very interesting. It was in 1974 that Peter Nauss, who was a Danish computer science pioneer, wrote a book in Concise Survey of Computer Methods, and he mentioned in that book, when he's basically referring to what we think of data analysis and then data science, he referred it as science of dealing with data. So till this point, the term data science was not being used, but it was the first time he mentioned the science of dealing with data when he was talking about statistics and analysis of data. And then it was in 1976 that programming language S was developed at Bell Laboratories, and till then Fortran was being used for statistical computation. I don't know if there are people old enough as me to know Fortran. I'm actually much older than what I look like. Yeah, it was actually, Fortran is a powerful also programming language. But anyhow, S was developed for that, to offer an alternate and more interactive approach. Ailey design decisions that hold even today include interactive graphics devices and also providing easily accessible documentation for the functions. So this Bell Laboratories actually should have it. Keep it in your mind. Like it's actually they played an important role in over like a couple of decades in the in the field of statistics and also like industrial statistics and they're owned by Nokia now. So now I think that it's called Nokia Bell Labs. I don't know if they're that important now anymore, but but in the past, like they were actually, and a lot of great mathematicians worked at Bell Labs. This photo is also from Bell Labs. I don't know if it is from the 70s. I think it's from the 80s, but I find it also very interesting. I really love looking at old photos and analyzing it. I don't know if they posed for that photo or not, but if they didn't, it seems that he's talking about something which he's very passionate about. One of them is very interested. The other one is a little bit suspicious. is the guy on the left, I don't know, or angry even, and the woman, the lady there, oh, my gosh, she's bored. She's really bored. I have a feeling that she's like, okay, leave me alone, I want to work. But, yeah, anyhow, I find this photo also very interesting. Yeah, so in 1991, R, program language R, which was an open source implementation of the S was developed by statisticians Ross Ihaka and Robert Gentleman from New Zealand University of Auckland. And as we know, probably most of you know, that R is still being widely used in the field of data analysis and data science. And, yeah, it has been used actually much more. It is now like a lot of companies or a lot of research centres prefer Python, but still R is being used also a lot. And in 97, the term data science is being born. Basically, Professor J.F. Jeff Wu calls for statistics to be renamed data science and statisticians to be renamed data scientists. Professor Jeff Wu, born in 1949, is still working as a professor at Georgia Institute of Technology. He's from Taiwan, but moved to the US after his initial studies in Taiwan. And he's known for his work on the convergence of the EM algorithm. I don't know if you know it. It's very interesting. It is an exception maximization algorithm. And resampling methods such as the bootstrap, probably mostly heard of it, and industrial statistics. I actually put this information there because to just not forget that the person who used first time the term data science is still not just alive, but also still working. And that is actually like a point for me for this like a slide that actually how young is still this job as like data science job. In 2001, Cleveland published data science and action plan for expanding the technical areas of the field of statistics, and to Cleveland, this was also very interesting for me to read that, because nowadays, actually, we are a little bit different, like on the definition of jobs. To Cleveland, actually, data analyst is good at programming, but has limited knowledge of statistics. A data scientist, on the other hand, comes from a statistical background, but has to work more closely with computer specialists. This is actually like, yeah. He also worked at Bell Labs, as you can see also on the screenshot of the top of the photo, the paper, and yeah, so that was also a very important article in its time. In 2012, probably, you have heard of like DJ Patil, a very actually important person, and also Thomas Davenport. They wrote this data scientist and sexist job of the 21st century, and this article was being circulated a lot for years. From the article, I quote that, more than anything, what data scientists do is make discoveries by swimming in data. It's their preferred method of navigating the world around them at ease in the digital realm they are able to bring a structure to large quantities of formless data and make analysis possible make analysis possible that's actually also um interesting to see like where actually when when it was the hottest job or sexist job um what how it was being described by the experts so um now to have like a little bit of history in in our head, I would go to the next section or last section that, what is actually data science nowadays, and do we really need data scientists? For that, I have a lot of ideas myself. But for preparing this talk, I decided to ask to make a very small survey and ask people in my network who work either as data scientists or are related to data science or data science managers. Is there anyone here who answered my survey, actually? OK, two people. Ah, nice. I'm sorry if I interpreted your answers wrongly. Come to me afterwards if that happens. I just sent this very short, I don't know, you needed like three minutes to fill out like a couple of questions. I know that, I mean, as a data scientist, I know that I kind of like generalize the answer of this survey from actually 65 people to what it exactly really is involved. But I think it gives a little bit of idea to us. So 65 people answered my survey. 40 of them were working or are working as data scientists, 20 of them as data science managers and team leads, and five other working with data scientists. So I actually split these five people based on their job. For example, one was a data analyst who had worked as a data scientist before, so I put in the category of data scientist. Or one was a data science teacher. I put in the category of manager. So just, like, try to split these five people in the answers. All of these people have worked, like, or 63 of them work for European companies and mostly for German companies. So that is also, for me, important to mention here because I don't know how it's in China. I don't know how it's in the U.S. It might be different. What I know from my experience and people around me is mostly from, like, actually Germany and European companies. So that I also want to also emphasize here. So, the first question I asked actually, what are data science tasks to you? And here I want actually to, it was a multiple choice question as well, I tried to write very long sentences to get like an idea of like different fields. I didn't want just to put data analysis or I don't know what, just to, because then it was maybe, yeah, to really think of like, to see how people think of the task. scientists who answered this question the most popular answer like which around like a bit more than 80% of them chose is extracting information from data using statistical knowledge and programming language like example Python and R it was in parenthesis the second answer was supporting business owners translating between business problems and data solutions oh well fair and the third answer data analysis with scientific and research background then in designing models and deploying them. I emphasize I'm deploying because we have also afterwards designing machine learning algorithms and validating them. And then we have implementing ML models and producing an API endpoint of the model for the engineers. So the first three, which are the most popular ones, I try here to focus on as the answer of the people who work actually as data scientists right now. And to compare actually with data science managers, Most of the data science managers, I'm also one of them myself, we worked as data scientists. So I was very surprised to see there is a very big difference of the answers of the people who lead the teams of data scientists and the people who actually work as data scientists. So as I said, for data scientists, these three were the most popular answer. Extracting information, using statistical knowledge, programming languages, supporting business owners, understanding, like translating between business problems and data solutions, and then data analysis with scientific and research background. But then, however, the first three answers of the data science managers is like, yeah, they were just clicking on everything. Basically, as you see, their percentage is also high. But the first one is designing models and deploying them. So for them, it's the main task of data scientists, designing, but also deploying them. And then the second one is actually, like, supporting business owners, okay, fair, designing machine learning, algorithm validating, I think also fair, and then the rest. I want to focus on this one. This is the biggest gap. So it was the main task by actually stated by data science managers, but also, I don't know, fourth one by the people who work themselves as data scientists. And I think that's one problem that we have. Then I ask, skills needed to get hired from data science managers. So if they choose their managing teams, they go to this question. What are skills you're looking for when you're hiring? And they answer machine learning and algorithm, okay, for mathematics, statistical, analytical thinking, great. Deployment of models, again, as one of the main tasks. Data analysis and data visualization, field knowledge or domain knowledge of the business, familiarity with the databases and data warehouse and all that, but still like the ones on top that I want to focus. and then i asked them okay these are the skills that you're looking for do you actually find these skills easily when you're hiring and um just six percent of the data science managers said that they find these skills easily like more than 50 percent of 53 percent said no i can't i can't find it and 41 percent of them said that um well sometimes i have to pick all this finding them That's actually also one thing that I wanted to pull out. Then I asked, where is the skill gap? If you cannot find these skills, what do you think the reason is? Where is the main skill gap? Then we again come to that point. Because I'm looking for data scientists who can do the deployment as well, or in general, have some skills in engineering. I don't go through the rest because I think that is what I want to pull out today. Again, come back to it. Then I asked data scientists, is it clear what you need to do? Is your task clear as a data scientist to do? More than 50 per cent of the data scientists who answered to that survey said that it's not clear for them what their task is. And just 47 per cent of them said that it's very clear. I mean, can you imagine any other job, like carpenter, electrical engineer nurse that will say oh, it's not I don't know what to do. I I don't know what my job is Yeah, as Germans would say it is catastrophe I think it's really catastrophe Why isn't it clear so I asked him why is it this was not a multiple choice question So they could just I wanted to understand what is the main pain point so they could just choose one And so the ones that they said it's not clear, they go to this question. And the first answer is, there isn't an unrealistic expectation from data scientists where I work at from analysis and engineering. And then the managers don't understand what data scientists can exactly do in the company and work at. There isn't much data. I don't comment on that. There isn't any data science use case. So these are like the how. Yeah, so I just go fast forward. So summary of the survey for me is that difference of perception in data science tasks. So data science managers see deployment. I want to, again, make this one bold as a main task of data scientists as data scientists, they don't do it themselves. Data science tasks are not clear. More than 50% of the data scientists find their tasks unclear. Skill gap between the requirements and the job market. 90% of the DS managers have problems finding the skills they're looking for. And there is an unrealistic expectation. I mean, just to come back to the skill gap, to go back like that, the argument over saturation, over saturation of the job market. I mean, if we do have over saturation of job market for data scientists, how data science managers are now finding what they want to 90% of them. Yeah, so. And then unrealistic expectation, about 40% of the data scientists who find it has a clear thing. There's an unrealistic ninja expectation from data scientists, from everything, statistics, data analysis, engineering, business, everything, presentation. So is there science dying out? I would say no. I think that if there is an unrealistic expectation from data scientists, it means data science is not dying out, but needs to be redefined, reformulated as a task, as a job. If you realize you're looking for a ninja as a manager, so maybe you need more than one person, maybe you don't need a data scientist, maybe you need other titles as well. And I think data science is not dying out. It's time that data science is becoming mature and we should define it, all of us, where we work at as data scientists or as manager. What is data science? So I think that data science, of course, has overlap with other fields, like it has overlap with business, with data analysis, with MLOps and engineering. I mean, if you have an answer, well, mine probably go, oh, what is that overlap? it's smaller than the other one. I mean, forget that. I would just put four circles there to give a message that there are overlaps and it is fine and it's normal. Every job has overlap with some other jobs. But I think that as a data manager, you shouldn't look for a ninja who can do all of these things. So that shouldn't be the focus. And I think you shouldn't focus on the specific parts and overlaps, if you are just like actually concentrating on the overlap of engineering and data science or business and data science, then maybe you don't need a data scientist or maybe you're looking at the wrong, like you're just very narrow, looking very narrow. And then define who you exactly need and if data science title fits the role. And as data scientist, I really think that you should define what kind of data scientist you want to be because there are data scientists that they are actually more closer to the business. They can work with business stakeholders, understand the problem, translating it. There are data scientists that they have engineering skills, and they can work more closer to that overlap. And I think it is fine. Or data scientists have fantastic mathematical and statistical knowledge. So I think that you should define that for yourself, which kind of data scientist you want to be. And I think in interviews and hiring process, you should communicate that with your with your interviews and also from both sides of course from the people who are hiring but also this should be discussed in interviews okay what do you understand as data science job what data scientists should do what team you have do you have engineers there do you have data analysts there do you have the product owners and product managers there so this is very important to understand what would be your task as data scientists i want to finish the talk like like here. Thank you a lot. And we are actually hiring data scientists who should do my MLF. No, I'm kidding. But we are hiring data scientists, but also data analysts in my team. So contact me if you are looking for a job. Thank you.

Speaker 2 [24:17]

that's really good and actually I was a data scientist before and I do have like some kind of some of the thing resonate with me and I really wonder like um do you have any opinion about the hiring process now for for data scientists like do you think there is what like anything that you think is not a good practice or anything that's good yeah any opinion yeah

Speaker 1 [24:40]

Yeah, opinion, of course, I have everything, I'm kidding. But I think that we have not just for data science in general and tech industry, we have this very frustrating long process of hiring via a lot of, I don't know, tech challenges, many interviews, and I think it's not okay. I think it's a lot of wasting time of the candidates but also the people who interview and hire. And I think that's not just specific for data science. I personally don't like it, and I think it's not healthy and good. But in general, for data science, I think, as I said, the problem is that it's very unclear in hiring process when I hear from other people going through hiring processes that what is really their task and what is needed from them. And I think that's...

Speaker 2 [25:26]

Yeah. So we have a question from the audience. Should companies hire statisticians instead of data scientists? Yeah. Statisticians? I don't know how to pronounce that.

Speaker 1 [25:41]

statisticians. So I think what I try to make as a point here is that you should really see what you need. I don't know what that company needs. I mean, statisticians not necessarily should be able to code in Python. Not necessarily. I mean, some do, but not all. So I think that really depends what the company needs.

Speaker 2 [26:10]

Yeah. So I think that's it for the questions. Thank you so much.