#51 Inclusivity and Data Science

Hi, I'm Maggie, VP of Curriculum at datacamp, and I know firsthand how overwhelming it can be to keep up with the rapid changes in data and AI. But staying ahead doesn't have to be hard. That's why our curriculum team has curated the best learning content to help you and your team stay ahead, covering everything from data and AI literacy for busy executives to cutting edge generative AI for developers, and everything in between. Whether you're upskilling yourself or your team, we make learning practical, engaging and flexible. It just works. Power your data and AI transformation journey on datacamp.com.

Welcome back to Data Framed, friends, the weekly Data Camp podcast exploring what data science looks like on the ground for working data scientists and what problems it can solve. We are more excited than ever to jump right into season two and to start off 2019 with a bang. This week I'll be speaking with Dr. Brandeis Marshall about people of color and underrepresented groups in data science will talk about the biggest barriers to entry for people of color initiatives that currently exist and what we as a community can do to be as diverse and inclusive as possible. Brandeis is an Associate professor of Computer Science at Spelman College. Her interdisciplinary research lies in the areas of information retrieval, data science, and social media. Other research includes the Black Twitter project, which blends data analytics, social impact, and race as a lens to understanding cultural sentiments. I'm your host Hugo Bown Anderson. You can follow datacamp on Twitter @datacamp and me hugobown. You can find all our episodes and show notes@datacamp.com Community Podcast this is Dataframe. In the show Notes, we've also included a link where you can suggest future guests, so please make as many suggestions as you'd like. The majority of suggestions from the community so far have been of white males, so I'd encourage you to make suggestions that reflect the diversity both in the data community as well as the diversity amongst all of our stakeholders. As Brandeis says in our conversation, and I quote, I think it's so important that for anyone that's in the data world, if you can call yourself a data scientist or any of its variations, is that you look around the room and see who is represented and how they're represented and include other people as part of that conversation. Asking questions is a necessary component of being in the data space. Grab hold of that and continue to ask questions.

Hi there Brandeis, and welcome to Data Framed.

Well, thank you. Wonderful to be here.

It's such a pleasure to have you on the show and I'm really excited to have you here today to discuss the representation and lack thereof in data science of people of color, along with your work in data science education at Spelman College, in particular about the black Twitter hashtag, among many other things. But before we get into all of.

These issues, I'd like to find out.

A bit about you. So I thought maybe you could start by telling us what you're known for in the data science community and what type of work you do.

Well, I guess it started in graduate school when I took database. So I'm really in the database realm of the data science continuum. So I'm what they call now a data engineer because I love the organization of data and the ability in order to retrieve it in structured ways and sometimes even unstructured. And then I moved into information retrieval because I love the web and understanding misinformation and noise was part of my PhD work. And then I kind of continued on the continuum there with loving data. So I'm really a computer scientist who loves data and I kind of stumbled into data science about five years ago. Information retrieval is one area of data analysis. And so I like algorithms, I like databases. And so that led me to information retrieval. And then there was this notion of how do you make good decisions based upon that data? So you're trying to move data to information. And for me, that gap between data and information is credibility. And that's where you have these algorithms and these processes and how do you narrate what's happening front of the data to give you the informations to make the good decisions. And so that's how I got into data science.

Great.

And what type of use cases do you think about in terms of these questions?

I think about what is the in client, so what is the stakeholders or going to do with the information that they receive from the data and what is the quality of the data so that they can make those good decisions? I'm more exploratory than explanatory in my data analysis. I want to make sure that there is good value in the data. And of course I want to make sure that the information that they receive is sound. And that's very hard to validate. Very, very diffic to validate because you can always use more data, you can always use more processing time, you can always do another algorithm on that data. But it really is about what story you're trying to tell and what's that narrative that you're trying to expose the stakeholders to.

I was just going to say this.

Is such an important part of data science which is often overlooked when the conversation happens in general.

Right.

These kind of foundational thinking about the data generating processes, getting your data into a form that is actually usable in order to impact decision making or research, right?

Yes, very much so. I mean, I consider data science to be five lanes. The first one is, as you mentioned, is the data collection and cleaning. The second one is the storage and management. The third one is the analysis. The fourth one is the viz. And the fifth one is the storytelling. But I think everyone needs to understand that storytelling is to me, one of the biggest crux because if you do, if you ask these questions, you go through the full five lanes. If your story doesn't make sense to the end user, you fall completely flat. It makes no sense in order to even to be churning, to be wrangling, to be manipulating that data.

Absolutely. So you're also heavily involved in data science education, right?

Very much so. I think it's imperative that as many people know about how to handle their own personal data and, and as well as how to handle data within their own domain. I think every domain uses data in some aspect. Either you're collecting it, either you're manipulating it, or you're using it. Doesn't matter what discipline you're in. So you need to know how to manage it. You need to know when you have enough data to make sound decisions. You need to know when you don't have enough data to make sound decisions. You have to know how to ask questions. It's about being curious. It's about trying to understand what that data is trying to tell you, not what you're trying to force the data to tell you.

Right. And currently we live in a, in a world where some people have data skills and they're heralded as data scientists or data analysts. Other people who may be their colleagues don't. And there's kind of a gap there existing. And I'm, I'm wondering if you see a future where this gap still exists or one in which kind of data skills and data literacy are spread throughout organizations.

I think we're almost at this point where data skills is essential, as important as reading, writing and computer programming. I mean, I know coding, everyone calls it just coding, but I say computer programming. Data skills are so essential. We carry around computers all day long with our cell phones and our mobile devices. We need to know how much data we're ingesting. We need to understand what that data means to us individually and what does that mean to everyone else. For instance, you get an email. It has an attachment. That attachment is, let's say it's 17 megs. Why do you have an attachment that's 17 megs? How does that impact the rest of your system? How many times has that email been sent out and how many people has it been sent out to? That is putting strain on systems. That's putting strain on yourself because you have to download this file to even be able to view what's inside that particular file of that size. And that's just a small example. Data skills need to be universal. It needs to be taught from the time a child can pick up and use a device.

I love it.

And that was going to be my next question. How early do we need to start educating people around this? And my next question, I mean, you just answered that. So my next question is, how long will it take for the education system to catch up with these needs?

Well, that's a tough question, Hugo, because.

I'm glad you didn't ask me that one.

I went to, I would say, how long do you think it will take? Because right now we're still in CS for all, or we're trying to infuse computer science skills in K through 12 education. And that is something that we have seen to be a bit of a challenge, to say it nicely, but there are a lot of momentum in getting that done. But how can we make sure that it's equitable? How can we make sure that there is participation of all students in computer science education? I think a lot of the lessons learned from that environment can be therefore translate into the data science environment. And I want to make a particular point here. A lot of people consider data science and computer science to be essentially the same thing, and I beg to differ. Data science is a combination of several different disciplines that includes mathematics, statistics, also computer science. But then there is a domain context that needs to be considered here, infused. You have to consider data science completely separate from computer science, just as computer science is completely separate from mathematics and statistics is separate from mathematics as well. That makes sense.

It makes perfect sense. And I think one thing you're speaking to there is, you know, developing data analytics skills and data intuition and statistical intuition and all of these types of things, which you don't necessarily need to write any code to do, for example, but learning how to like the touch of data, the feel of data, the smell of it, how to work with it.

Right, exactly. If some student is going to build a YouTube video, have their own YouTube channel, create these videos, they have to know how that Information is going to be stored and who has access to it.

Absolutely.

And what the analytics are around that video, how many views, how many likes, how many comments, what is the nature of those comments? Those are the analytic skills that that individual has to understand. They can't just put up a video. There's so much richer and or robust information that needs to be understood by that particular individual.

Agreed. And so I know something else you're heavily invested in is practical solutions to, in thinking about and working towards diversity, equity, and inclusion in data science in particular. For example, you educate at Spelman College.

Which is historically a black liberal arts.

College for women in Atlanta. Right?

Correct.

So I thought maybe you could say some words just about how you think about this in general.

Well, educating black women at Spelman College is a whole different experience because not only do these students look like me, I am in some respects a model, in other respects a mentor. But I think it's important to understand that since I'm in computer science, a lot of the computer science literature is centered around a very homogeneous type of background. What I try to infuse within the classroom is the fact that people of color, particularly black women, are part of the conversation, have made advances and innovations in computing. Therefore, the students, I feel, now have a better attachment to why they are represented and why they are welcomed inside of this discipline. Maybe they don't feel welcome in all spaces, but at least they have a certain amount of context so that they can feel as though they can do anything they want in the discipline. Makes sense.

That makes perfect sense. And I'm sure places like Spelman and kind of the deep foundational tradition they're built upon. It's interesting having new disciplines such as data science being taught at places such as Spelman because you can build on kind of the old traditions of educating, in this case, black women, but in a totally new discipline.

Right, Exactly. I think it's important that people of color are first to be thought of inside of a new discipline. They're not an afterthought. It's not going to be scrubbed without people of color. I want to make sure that when we talk about this emerging field of data science, that we continue to be inclusive and continue to make sure there's representation in all areas of the discipline. And for the students that I teach, I want to make it very clear, here is some examples on how representation is lacking in the discipline and how there's an opportunity for representation to therefore get elevated. And so, hopefully, with the classes that I teach and the lectures and the conversations that I have with students, that that is something that they have a takeaway from.

Interesting. And so what changes in your pedagogical approach when teaching, for example, black women at Spelman?

So what I did in the very beginning, I joined Spelman in 2014. And in the beginning, I would kind of skip over the history of computing because it was very homogeneous and all of the innovators and all of those type of individuals. So I used to skip it. Then I decided to create their own little hashtag. I called it Black Computing. And it was once a week I would find a black person, sometimes seasoned, sometimes a little younger, sometimes with advanced degrees, sometimes without, and have the students learn about that particular individual just for a day or so, and then they would share what they learned over Twitter. Then I decided that it was important not to just make this a once a week type of engagement, that it was important that as topics were introduced, that they would see people of color. So I would take time in order to find people of color, black people, women, black women, who added to the innovations of the discipline in which the topic that I was talking about in computing, I tried to do the same thing within data science as well. I think it's very important to have the cultural environment so that students know that they can too add to this particular discipline. And data science is a bit more difficult because it's so new. But definitely in computer science, I was able to make that happen.

Yeah, absolutely. And I do know that if we consider data science as the confluence of statistics, data analysis and computer science, which is one at least zero thought or approximation to it, I think, particularly for women, the statistics part of it, I think is incredibly helpful to see how many women are active in the statistics community.

Yes, very much so. I think it's important to connect students, connect myself even. I'm still a learner within data science to continue to look for those individuals who have made such huge strides before it was deemed data science, before it was really deemed statistics, or before it was deemed computational math, that we look back, know our history, and be able to bring that history forward and give it new life.

Absolutely. So then, through the lens of all the work you've been doing to bring broaden participation in data science, what are the biggest barriers to entry in data science for people of color?

I would say there are two. The first one is awareness, and the second one is access.

Tell me about awareness.

So with awareness, it's unconsciously using data without necessarily knowing how to use it. So not being really aware of the Power of the information. So let me provide a little bit of an example. So we're in the day and age where everything has a credential attached to it. You have to give a username and a password, and with that username and password tends to come along with that. Is your demographics, whether you're male, whether female, what happens to be your ethnicity or your race? And so my students are in this generation where they're used to having everything behind a username and a password. I'm not of that generation. So when I have conversations with them about, well, you're downloading this app and you're providing these credentials and you're attaching it to some other social media, do you know that they now are being able to get all this other information about you or even to understand how you might speak or how you might use your language in order to figure out whether you're male or female or possibly what demographic? That type of awareness is something that students, a lot of students I speak to, a lot of people that I speak to don't really understand. So having that conversation is very important. That's what I mean by awareness. What are you giving up? Do you find it to be giving up, or is it just a normal mode of transaction?

Yeah. Why is this fewer barrier to entry for people of color in data science?

I consider it a barrier because if you don't understand or don't have conversation about the context of what information that you're providing, you don't understand its value. If you don't understand its value, you have no idea that it's important for you to dig deeper. If you don't understand, it's important for you to dig deeper, then you won't dig deeper. You almost as if there are things that just happen to you versus you being proactive. So this whole reactionary versus proactive kind of thought path. And so if you, for data science, if you don't know how important data is and how you being who you are is important to whomever that business or organization is, you therefore don't have the questions to ask.

Right. That makes perfect sense. And do you think this is a bigger barrier for people of color than other demographics?

I think it's a barrier for everyone. I think historically, people of color have unfortunately had some very difficult incidents recently and in recent past that have made having these type of conversations a little bit more difficult. And so I think people of color, especially black people, because that's the community I come from, needs to really think about how their transactions are Happening and what does that mean for themselves, for their families and for the next generation?

Absolutely. And I think there's such a trade off there. Right. I mean, what the example that springs to mind when you talk about giving access to certain aspects of your data and social media is when I sign in with like a lot of websites offer me to log in via Google. I can just use my, my Google account and I've got. I honestly should probably do a bit more due diligence and figure out what I'm giving away there.

Right. Google is one example. Right. Because if you also use, you use Gmail, then of course all of your email exchanges are therefore in somehow being processed and analyzed. So there's just a lot of information that you're giving that you don't realize you're giving. You just think it's just data, but it's really information about you.

And we're starting to see kind of the downstream effects of that more and more. So I get emails from recruiters every now and then and say, no and Gmail will make suggestions for you now. And I got an email from a recruiter saying, would you be interested in a chat? And the two things Gmail suggested were no, I'm not interested and no, thank you. And I was like, oh, you've learned pretty well. And in fact, I joke that if one of the options had been yes, I would have been like, oh, Google thinks I might be interested, so perhaps I should.

Right. It's this notion of power of suggestion that is very subtle and that subtlety could be dangerous. But also the fact that you can write an email now and there is an autocomplete is based upon hundreds of millions of scrubbed correspondence in order to figure out what you're going to say. So is that really you saying it or is it that power of suggestion? It's a more philosophical questions, more on the right creative brain than my analytical left brain. But I think that's what data science does for me is that it sits at that intersection and lets those synapses between the left and the right brain kind of fire. Right. What happens with that logic part and what happens with the creative part? What does that mean?

Fantastic.

Yeah.

We'll jump right back into our interview with Brandeis Marshall after a short segment. Now for something completely different. I want to introduce a new data frame segment called what Data Scientists Really Do. I'm here with Emily Robinson, a data scientist on the growth team at Data Camp.

Hi, Hugo. I'm excited to be chatting with you on the first episode of season two of Data Frame and many congrats on a great season one.

Thanks Emily. We were fortunate enough to have so many great guests on the first season, yourself included. And that's actually what I wanted to talk to you about today.

What in particular?

Well, a question I get a lot is Hugo, you speak with so many data scientists for long stretches of time about their practice. What are your general takeaways about the discipline as a whole, what they do, what the space looks like, where it has come from and where it's going.

And how do you respond?

Initially I'd have some sort of brain splosion as it's a non trivial task to synthesise 30 plus hours and growing of long and discursive conversations into bite sized chunks. It was actually a really difficult task, but I thought it would be worthwhile. So I decided to try to write an article about what the data science landscape looks like and where it's going through the lens of the podcast.

Right?

And that was eventually published in Harvard Business Review as your article what Data Scientists really do according to 35 data scientists.

Right.

And I've also given related talks such as what Data Scientists really do according to 50 data scientists at the PI Data NYC conference and the NYC Open Statistical Programming Meetup, for example. One of the great things about this form of synthesis for me is that different media allow for differing ways of telling parts of the story. So giving talks, writing articles and discussing it here on the podcast naturally allow different emphases. Another great aspect for me is that the content naturally evolves as a function of how many data scientists I speak with. Imagine at the end of season two that we could talk about what data scientists really do according to 100 data scientists. I also joke that something I love about giving this talk is that I get to state other data scientists opinions so I can't be held accountable.

So how do you go about synthesizing 35 or 50 or 100 plus hours of conversation?

Hugo, let me first say that by the time I attempt this, I've spent a huge amount of time with each episode. So perhaps it's worth digressing slightly to tell you and the listeners a bit about the process of constructing each episode. Would that be okay?

It's your podcast man.

Well, guests are either people in the space that I know of and think our listeners will find interesting data scientists who are recommended by others, or data scientists who reach out to me and make a pitch for an interview. In any case, I spend a substantial amount of time looking into their work talks blogs, GitHub, repositories, articles, and more in order to see whether their interests would align with those of our listeners. Then we jump on a call to brainstorm what a conversation could look like. As a function of this call and any other resources the guest shares with me, I draft leading questions for the interview in a Google Doc. The guest will generally throw down some bulleted responses. We definitely don't want a script for the conversation, but some talking points jotted down or a skeleton go a long way in my experience. Then we record. Last season after recording, I would edit the audio myself using Audacity. Then I'd get the audio transcribed to text, write and record my intro outro and the segments, the construction of which for another time and place. I'd then arrange all the pieces together using GarageBand. Or do you say garage? What do you say?

I generally say GarageBand.

GarageBand for our North American audience, and I'd give the episode one final listen then in its entirety while editing the transcript, which were generally okay with the understandable exception of technical language. So tangentially, one of my favorite regular transcription errors was Psycho Learn instead of scikit Learn.

I remember a lot of that process from when I appeared on the pod, but I didn't quite realize how much work went into it after recording.

Yeah, and I haven't even mentioned the nuances of working with communications and PR teams of large organizations when my guest represents them. So this is always a very rewarding process. But such teams need certain checks and balances which add significant steps to the workflow. So anyway, all this is to say that I've spent a lot of time with each conversation, so when it came to synthesizing them, I had a lot of thoughts racing around my head.

And I'm sure it's difficult to access high level trends from memories of so many hours of dialogue.

Oh totally. And what really helped was leaving Datacamp hq, going for long afternoon walks in downtown Manhattan and reflecting on my memories of these conversations. This was almost a mystical first step of ideation in reconstructing the general trends. Only after this did I go back and scour the talking points in G docs, reread portions of the transcripts, and start to really reconstruct the higher level trends from the podcast.

Thanks Hugo. This has been a great introduction to.

Whoa, whoa, whoa, whoa.

I haven't even told you any of my takeaways of what data scientists really do.

Unfortunately, we're out of time. Let's chat later in the episode.

Sounds great. And who knows, Emily, perhaps We'll even need a recurring segment. After that interlude, it's time to jump back into our chat with Brandeis.

So that's kind of a very nice introduction to thinking about awareness in data science. The second point you mentioned was access. So can you tell me a bit about access?

So a bit about access. Okay, so there is this conversation that's been had over the past, probably decade or so about the digital divide, is that people of color don't necessarily have access to the Internet as people not of color. Okay, so I will take that point that yes, there is a digital divide, there is a gap. However, I will say that there is newer research that's coming out that says actually African Americans do have access to the Internet, but it tends to be on their mobile devices. That access on a mobile device is very different than the access on a laptop, very different than an access on an iPad or any other type of tablet device. So when it comes to access, you want to be able not only to access the Internet, but you want to be able to access the tools, and you want to be able to then create and innovate. And the access that I'm speaking of is in the creation and the innovation component. Data science is a field that is very open. It has many different definitions and threads and educational pedagogies, instructional strategies. But when it comes to access, who is delivering that information and how are those learners supposed to then get expertise and grow their expertise? For example, there is a surge of boot camps sprouting around the country. These boot camps are sometimes a week, sometimes two weeks, sometimes they're 12 weeks. But these boot camps tend to cost money. These boot camps tend to.

And a lot of money.

And a lot of money, yes.

I mean, yeah, we're talking like tens, twenties, thirties of thousands of dollars potentially.

And there are different models, many different models. And they tend to be all day at 9 to 5. So if you're gonna say, let's do a two week boot camp, that's 9 to 5 every day, who is gonna have time in order to deal with that boot camp, deal with all the exercises, learn what they have to learn as well as work, as well as possibly taking care of a family that doesn't provide access. Correct?

Absolutely correct.

It's part of the problem. We want to create data professionals, but creating data professionals takes time. Every company realizes they need data professionals. Call them data scientists, data analysts, data engineers, data visualizers, but there's no time. So then someone is going to have to quit their job in order to take A boot camp and then hopefully get a job thereafter and make a lot of money. That's one possibility. The other one is they still have to feed their family, they got to feed themselves, they still need a roof over their head. So how do they do it? And then if you go to that particular boot camp, do you have the machinery that you need, the computing devices that you need in order to to completely absorb all of that content? Can you download Anaconda? Can you make sure that you understand what Python or R is? Do you need some pre learning before you get to the boot camp in order to therefore be successful on top of then having the money in order to pay for it and the time in order to dedicate and invest in yourself in order to become a data professional? So there's a lot of different stumbling blocks that I believe really are opportunities, but that access is really providing that access to the information and the time and the investment and the money in order to actually pursue this particular track. And this track doesn't have to be all computer coding. It can be on the VIZ side and the data stories telling side without a problem. But there needs to be a better mechanism for access.

You're mentioning that these stumbling blocks can be turned into opportunities. One takeaway or one thing that comes to mind there is that as educators we need to meet learners and potential learners where they are as opposed to where we want to teach from.

Exactly. And that's very difficult because there's a certain strand where people believe that you have to know how to compute a program in order for you to enter into data science. I don't believe that's true. I think that there's a lot of methodology and practice that does not include computer programming. But if you want to computer program, I think it's important that you have that foundation which now means for an educator they have to choose a programming language. Is it going to be Python, is it going to be R, Is it going to be something else? Right. Is it going to be Julia or something else? But you have to choose a programming language to switch between programming languages. As a computer scientist, I am not a fan of.

I'm not even a.

Computer scientist, I'm just not a fan of it. So switching from one language, going from Python, then going to R and then going back to Python and going to Java, I mean, it's just not something that I think builds competency and confidence in the learner. Stick with one language, let the learner get one language in for a full year and then you can introduce New languages, it makes more sense. And then of course, the learners could be non traditional. You can have people who have been working in the tech industry or in the non tech industry for a decade or more. There are different types of learning. You have to meet them where they are, what are their objectives, what do they want to accomplish, what jobs do they actually want? At the end of the day, some people just want to be in data visualization. So sculpt the learning around data visualization. They don't need to know all the things about data acquisition and collection and cleanup. They're going to be pulling data down from an Excel spreadsheet or from some database somewhere, and they're going to be visualizing it and writing reports. It's just so important to meet them where they are.

I couldn't agree more. And this idea of access is something we think a lot about at datacamp. And actually, you know, it doesn't solve all the problems by any means. But one of the many reasons I first joined datacamp was the fact that, you know, I used to do a lot of in person training. And one of the toughest things when training people, as you know, is installation at the start of a workshop. Right. And so the fact that at datacamp, when I met the datacamp co founders in their exercises, we spin up images where people can get coding straight away. And of course they will need to do installs on their own systems downstream. But you can get people kind of being functional immediately without getting them just.

Bogged down for like three hours in.

Getting it up and running.

Right. I mean, I think it's important to have the ability to ramp up quickly and then go back and let's. Okay, now let's take three hours at the end of the day of week two of a ten week program and let's now try to put it on your own machine so that you are able to work independently post training. That's very important. I think that piece doesn't happen. I think there's a lot of pieces that will ramp people up very quickly. There will be two hour, like hours of code and then it falls flat. Right. There needs to be a concentration on making sure someone can work independently. Because as part of being in data science, you will work in teams. Because data science really is done as a team, It's a team sport. And it really is something that is done individually, just like mathematics, just like computer science, just like every other discipline that exists on the face of this earth. There will be individual work and there will be group work and so how do you make sure that you translate those learners from the group think and ramping them up quickly to get them excited about what they're learning to being able to work individually. That particular gap analysis is where an in person educator comes into play. That's why I think it's so important to have people face to face, at least for some segment of time. It's very, very important. Certain things you just cannot quite do online that can be done much more seamlessly in person.

Agreed. Because we're inherently social animals that are part of a community as well.

Exactly.

So these issues of awareness and access I find so, so interesting. And I'm wondering what initiatives exist for underrepresented groups in particular people of color and black people that can get them up and running, coding and doing data science and AI and all of these types?

Yeah, well, I know of two initiatives that have been around for a little bit. I mean a little bit meaning like two or three years, maybe a little bit more for the second one. But I know there's a number of them kind of sprouting up. But the first one I know of is Black in AI. And this is a couple of individuals I think mostly are graduate students and newly minted PhDs have come together and put together some workshops. One of their workshops I believe this past month was actually at nips.

That's right.

And then the other one is Women in Machine Learning. And those have workshops have been going on for about five, six years now, I believe. And so these are women of all different background, all different pedigrees that are coming together to talk about machine learning and their work and their work in many different domains. So those are two places I would definitely go for the listeners to just get a start on what exists within the data science community for at least people of color as well as for women in this discipline.

So we'll definitely include those links in the show notes as well. But anyone listening? Blackinai GitHub IO if you want to go there right now to check it out. I've also seen on Black AI's website there are related organizations such as Black Girls Code, which looks incredibly interesting.

Yeah, so Black Girls Code has been around. Oh, I do not know exactly how long, but they provide learning camps in the summer and during the school year at different schools all over the country. I believe they have an engagement in New York that's pretty heavy. As well as in the California area. I believe it is in the Southern California area. They have been around trying to get students, especially black Girls in order to understand how to code. As far as I know, they look at Ruby as the introductory language and continue to try to add new curriculum and add new trainings. And so they have students that have been part of their trainings to come back in order to help out with new girls, to actually teach them something of how to code. I think it's a good initiative now, translating what black girls code does into something that is more formalized in the classroom and making sure the students can therefore continue their learning. Well, that's where I come in.

Fantastic.

So, generally, as a community of data.

Scientists and people who kind of talk about it in public, how can we do better with respect to being as inclusive and diverse with respect to people of color in particular?

This is actually a hard question, and it's hard because we are trying to right a wrong from generations ago. So as a community, we have to be intentional about our inclusion. It has to be at the core of every engagement. I think there's four principles. I call it pair. It's P is for participation, A is for access, I is for inclusion, and R is for representation. So I think as a community, we have to take these four principles and we have to ask the question, who is participating? Who has access? We also have to ask who is being included and therefore who is being excluded. And then last one is who is being represented? And that corollary, who's not being represented? If we take a moment to think about this pair principle, then I think we will get to a place, but we will be intentional about including individuals from all demographics. When it comes to our data sets, when it comes to our teams, when it comes to our conversations, when it comes to our algorithms, when it comes to our testing of these algorithms and of these systems, we have to be very, very intentional and purposeful, and I bet resourceful, because the conversation is always had. Well, we can't seem to find people of color in X or Y discipline. It depends on where you look and if you're really looking.

And that, once again, brings it back to intentionality.

Yes, yes.

And you actually spoke to something else there. Not only about how we, as a community of data scientists, can be more inclusive and diverse, but also in terms of thinking about end users and the stakeholders in all types of algorithms. Because, of course, algorithmic bias is a very huge conversation at the moment. And one of the most, I think.

Public examples of algorithmic bias was ProPublica's.

Expose of the recidivism risk model that was biased against blacks and used for parole Hearings.

Right, right. Yes. And there are several other examples. Right. I think Amazon had to step away from their algorithm when it came to hiring practices because their algorithm had some bias within it. Yeah. I think that there's a lot of examples of where we need to really think about who's in the room, what are the results saying and what the results may be excluding intentionally or unintentionally. I think a lot of things happen unintentionally just because who happens to be sitting in the room don't necessarily think about certain perspectives. And I think all those perspectives need to at the very least have some type of representation.

Absolutely. And I actually, at the end of season one, I had Kathy o', Neill, author of Weapons of Math Destruction, on the podcast and she discussed this idea that she's been developing with other people of an ethical matrix for algorithms whereby you have a row for each stakeholder and a column for things from efficiency to false positives to false negatives and.

That type of stuff.

And she gives the Compass recidivism model as an example, where you'd have certain demographics as stakeholders and, and the company as well, and society and all of these things. And I think that's one step in the right direction.

Yeah, I completely agree. If you have not read the book, get it today and take a little time, just read a little bit about it, just start leafing through it, just pick a chapter and start reading. I think all of it is good.

Very much so.

We'll jump right back into our interview with Brandeis Marshall after a short segment.

Welcome back, Hugo. To talk about what data scientists really do. We're here to talk about what you see as the broad takeaways you've discovered so far in your 50 hours of interviews with data scientists on Data Framed.

Yeah, we are. Emily and I just want to reiterate the motivation. So I'm really fortunate to have the opportunity to speak with all my incredible guests. And I think it's important to do these conversations justice as a whole by trying to decipher the emergent themes.

Would you like to tell me at a high level what you see as the themes?

Oh, absolutely. So firstly, it's important to note that the themes are more generally about the.

Space as a whole, rather than solely.

What data science, which happens to be the first theme. The fact that Data scientist is a catch all term for data savvy, quantitatively minded, coding literate problem solvers, as Cathy O' Neill and Rachel Shute wrote in their 2013 book Doing Data Straight Talk from the Frontline.

I do recall Cathy o' Neill pushing back on that definition in the final episode of season one of Data Framed.

Yeah, you're right, Emily. Kathy's new definition is data science doesn't just predict the future, it causes the future. And this forms part of a key conversation around the of data science on individuals and society and the importance of ethics, which we'll get to.

So if data science is a catch all term, what do data scientists actually do?

I think we'll need at least a.

Whole segment for this, but you can slice it in many ways. For example, you can think about a task based or a function based classification. When looking through a functional lens, you can coarsely break data science down into business intelligence, machine learning and decision science. I'd like to discuss these in more detail in future segments. In a task based classification, you can break it down into data collection and cleaning, building dashboards and reports, data visualization, building models, whether it be statistical inference or machine learning, and communicating results to stakeholders.

And it's also key to note that data science is an important input into the decision making function in any given organization, right?

Totally, Emily. It's essential. Data science doesn't exist in a vacuum and a huge part of data science is supporting decision making in a business and we need to recognize that it's only one of several inputs into the decision function. Also in episode 40 of Data Frame, when I interviewed Renee Tate, who's a data scientist at higher ed analytics startup Helio Campus. She's also creator and host of the Becoming a Data Scientist podcast. Renee and I discussed the importance of translating business questions into data questions, solving for data answers, and then translating these data answers into business answers. So this is a small teaser for more that we can chat about in upcoming segments.

So what other emergent themes will we chat about in the future?

Well one's that great strides are being made in industries other than tech. For example, we're seeing a huge impact of data science and analytics in health, retail, news media and the North American trucking industry, to name several.

What else?

That it isn't all just the promise of self driving cars and artificial general intelligence, which we can all be forgiven for thinking when we often see headlines such as an AI God will emerge by by 2042 and write its own Bible. Will you worship it? So that was a real headline. Related to this is that the skills data scientists need are evolving and perhaps experience with deep learning isn't necessarily the most important one.

Related to that is that specialization in data science is becoming more important.

Totally and that's another theme that emerged, which we'll cover. Last, but definitely not least, ethics is among the field's biggest challenges. When data science and machine learning are used in decisions around insurance payouts, as inputs to parole hearings in hiring, and to decide whether people receive their health benefits, we need to think long and hard about how our processes reflect our ethical frameworks.

There's a lot to discuss in all of these Lucky.

We're just starting this season off.

Thanks, Hugo. I'm looking forward to diving into all of these again soon.

Me too, Emily. Thank you. Time to get straight straight back into our chat with Brandeis Marshall.

Pivoting slightly. One of your research interests is the spread of social movements on social networks. And I was wondering if you told me a bit about the black Twitter hashtag and what you've learned from it.

Yes. So about 2015, I started to really get more engaged within Twitter. And as a way in order to engage my students, I developed a little project that I wanted them to work on, which is to delve into some of the social movements that were happening at the time. So there was a lot of celebrity disagreements, they call them beefs, back and forth. There was a lot of conversation about social injustice with the death and the murder of many black people and people of color. And so as a result of that, I had the class in order to put together a number of different hashtags that they found within some of these social movements and wanted them to look at Twitter as a way in order to kind of sift through that particular data collected and sift through that data. It was part of an information retrieval course. Long story short, the class did pretty good on the project, but I decided that I think it really should be a research area of mine because the social movements were continuing and the predominance of black people on Twitter was growing. And about 2014, 2015, there were about 23, 24% of African Americans were on Twitter. And then it increased as of last year to about 26 to 27%. And I think in 2018 it's about at 26% as well. And black people tend to be on Twitter and other social media more than any other Democrat. I think Hispanic Americans might be up there as well. Anyway, so when it came to this particular work, I had two students and we were talking about the work of Andre Braque and his conversation about black Twitter and Twitter as a cultural conversation. That's one of his works, as well as having conversation about the work of, of black tags or racialized Hashtags coming out of Sara ferlini. Then late 2015, early 2016, OscarsSoWhite hashtag came up as a conversation of how the Oscars, the National Academy of Arts and Sciences did not have any black nominees. So my students, understanding this conversation also around the time of the Grammys, decided, well, well, this is crazy. Why is there no black people nominated for movies? And why are there not that many people nominated, not black people nominated for Grammys. And so we put together a little Python in order to collect some data. We found some keywords and some hashtags that we culturated ourselves, we used ourselves. And then that's where the Black Twitter project was born. And for the past two and a half now, almost three years, it has been an ongoing project in order to collect in real time the tweets when the Oscars are being broadcasted. 2016, 2017 and 2018. We recently got a paper published and we talked about this movement, this movement from Oscars so white, where there was no people of color that were nominated for the oscars, to the MeToo movement, to Oscars less white on all these different hashtags. In between, the predominance of who was the host to what was the conversation happening within the Oscars and what the commentary was and how the language was. So that's what we're looking at. We're looking at the language of Black Twitter in respect to this social movement within the Oscars. And then more broadly, that's happening around the time of the Oscars. So. So it's been a wonderful engagement as a little bit of black background to black Twitter. I forgot to even mention that. So Black Twitter, the conversation kind of started in 2008 with a little blog post by Anil Dash. You can put that in the notes.

Will do.

And then there was another post, I believe on Medium by Chris Wilson, you know youw Black Win was the name of his. And then the published work of Andre brock happened about 2013 or so, 2012, 2013. And since then, there's been a lot of conversation about what is the cultural and the racial undertones of Twitter, what does it mean? How do groups identify themselves? How what are the characteristics and the behaviors of the group? But understanding that the group doesn't necessarily represent the whole culture or the whole people. And that's the very interesting component, is that Twitter was created. And it is, as an undertone, is something that's very much a white space. But Black Twitter is a subculture that's very much a doctrinated with, with black culture, with the conversations with the social justice, injustice, the black appreciation that exists within the black narrative and what that narrative looks like and how the conversation amongst that community is something that is very much akin to conversations that would happen analog world, but is represented in the digital world. So it's a very interesting construct, it's a natural evolution of Twitter, but something that was not necessarily the intent of Twitter. But it has definitely been a wonderful platform for at least black people and advocates and allies of black culture to have conversations, to have serious conversations and sometimes not serious conversations, and how hashtags and trending topics therefore have been elevated with the conversations that are happening within black Twitter. And there's actually a lot of work about how the conversations happening within the black Twitter space is somehow changing the language that's happening within the non black Twitter space. So I know I've talked a lot, but ask any question.

Oh no, there's so much in there. And one thing that sprung to mind was it must be so rewarding to work on this with your students and see their responses and have them learn all the techniques of network analysis, sentiment analysis, all of these things in a context that actually matters to them.

Yes. And so that's one of the wonderful things, is that because algorithms in computer science is something that's highly theoretical and pretty abstract and some students have difficulty in understanding it, I can easily now put together a little talk about graph theory that talks to them about users and followers and edges that makes it so concrete for them, they then understand, okay, now I know every time I send a tweet, what happens, or anytime I'm on Snapchat, this is what happens and then how the information or that tweet propagates. And what does that mean when it propagates? How do you deal with retweets? How do you deal with likes, how do you deal with comments? And what does that mean in this context of analysis? How do you rate the importance of each of these elements on Twitter in order to then be able to say, okay, this particular tweet is what you would classify as something in black Twitter, because black Twitter is not a. Is a very much a nebulous environment. It's not. Doesn't have boundary lines, it doesn't have a list of influencers. It is in the digital ethos. And so how do you then say something is part of black Twitter versus something is not part of black Twitter? So that's what my students and I are talking about and having these conversations about and really helps them because they are using their creative brain as well as their analytical brain in order to answer these questions. So we read things that are very technical on the computing end and then we also read things that are very much on the social science end as well, which I think is a wonderful blend.

Absolutely. And is there anything in the use of Black Twitter that you'd like to.

See change for right now? I don't want to see anything changed for right now, but if I were to hope for one thing in the next few years, that would be that we would use Black Twitter in a way to be proactive. There's a lot of situations that have come up where Black Twitter has been reactionary. There's the Black Lives Matter movement, say her name, I Can't Breathe. A number of other hashtags and social movements that have come out of the Black Twitter space and the users of Black Twitter. But I want it to be more than that. I want it to push success of black excellence, push education, push geekhood and geekiness as being something that's championed and valued. I wanted to push healthy living and self care. I want it to be something that black Twitter can really be part of changing certain aspects of the culture to be a lot more positive and taking care of ourselves as black people.

So we're going to have to wrap.

Up in a minute, but I'd love to know, you know, because you have a very technical background as well, what one of your favorite data sciencey techniques or methodologies is.

Oh yeah, I have a lot of them. I'm very much an exploratory type of data analysis person. So I always try to ask myself questions that I hope to be answered by a data set, whatever data set that I'm collecting. So of course I have to ask questions before I create a data set or look for a data set. But after I ask those questions, I then want to perform some very rudimery elementary statistics. I want to know mean max and things like that. That's an easy method. Call in Python. Right. Python Pandas. But then the other thing I love to do is to use Seaborn's pair plot.

I love pair plot and I love.

It because the diagonal will give you a histogram of. Okay, now what is, is the span of these values for each element. And then I like the correlations that are on the triangularization. Then I can kind of see as there's clusters that are forming and it gives me a good visual to know what to do next. That's when I'm able to Reevaluate whether or not my questions can even be answered by this data set. And then I always spur off into different questions. And I think that is very important as someone that works with data, is that you're never locked into the original questions that you ask. You have to be able to answer those questions that you originally asked and then be able to migrate and pivot to do questions based upon what the data is telling you.

And as you suggested or alluded to, the amount of information in Seaborn's pair plot, being able to see all those things and see different qualities jump out at you is really so much fun.

Yes, it's a lot of fun. And then you can of course do different visualizations. Then you know whether or not you need to do a linear analysis. You know whether or not you need to do some type of a sentiment analysis or semantic analysis. You can then figure out where to go next. Because if you see a correlation, okay, maybe you want to do a quick least squares linear analysis. Maybe you need a multi, multi regression. You don't really know where to go until you kind of see what the data is revealing to you. And a pair plot is where's my goal to. It's just my go to method. So then I know what to do next and know what questions I really should be asking of this data.

So my final question, Brandeis, is do you have a final call to action for our listeners out there?

Final call is be intentional and be resourceful. I think it's so important that for anyone that's in the data world, if you can call yourself a data scientist or any of its variations, is that you look around the room and see who is represented and how they're represented and include other people as part of that conversation. Asking questions is a necessary component of being in the data space. Grab hold of that and continue to ask questions. The call to action really is, yeah, be intentional, ask the questions, get the answers and ask more questions. That's the only way we're going to broaden participation in data science is to do the work, do the good work in data science and make sure that who's doing the work is as diverse as possible.

Thank you, Brandeis.

It's been such a pleasure having you on the show.

Thank you for having me. This has been a lot of fun.

It really has. Thanks for joining our conversation with Brandeis Marshall about people of color and underrepresented groups in data science. Two of the biggest barriers to entry for people of color in data science are awareness and access. Awareness in terms of knowing how data is collected, used, transferred and exploited when interacting with the world. Access in terms of the digital divide, access not only to the Internet, but access to the tools of data science. We discussed initiatives that exist, such as Black in AI and Women in Machine Learning, along with how to be intentional about inclusion by applying the principles in the acronym pair Participation, Access, Inclusion and Representation. Now, if I stated everything I took away from this conversation due to my verbosity, it'd take longer than the conversation itself. So I suggest you go back and listen to it again and or read the transcript. Having said that, I'd like to wrap up by re quoting Brandeis. I think it's so important that for anyone that's in the data world, if you can call yourself a data scientist or any of its variations, is that you look around the room and see who is represented and how they're represented represented and include other people as part of that conversation. Asking questions is a necessary component of being in the data space. Grab hold of that and continue to ask questions. Okay, so make sure to tune in next week for a conversation with Gabriel Straub, the head of Data science and Architecture at the BBC, where his role is to help make the organization more data informed and to make it easier for product teams to build data and machine learning powered products products. We'll be talking about data science and machine learning at the BBC and how they can impact content discoverability, understanding content, putting the right stuff in front of people, how Gabriel and his team develop broader data science and machine learning architecture to make sure best practices are adopted, and what it means to apply machine learning in a sensible way, along with so much more. I'm your host Hugo Bown Anderson. You can follow Data Camp on Twitter @Datacamp and Meugobound. You can find all our episodes and show notes@datacamp.com community podcast.