Podcast with Dr. Bjorn Mercer, Program Director, Communication, Philosophy, Religion, World Languages and the Arts and
Dr. Zona Kostic, Program Director of Computer Science
Big data is collected and processed to provide insight about nearly every facet of modern life. In this episode, Dr. Bjorn Mercer talks with Dr. Zona Kostic, a computer scientist who specializes in data visualization and machine learning, about the benefits of artificial intelligence. Learn how AI is being used during the COVID-19 pandemic and how new systems like Explainable AI (XAI) are being used to help data scientists understand how AI systems “think” and come up with decisions.
Listen to the Episode:
Read the Transcript:
Dr. Bjorn Mercer: My name is Dr. Bjorn Mercer. And today we’re talking to Dr. Zona Kostic, Program Director of Computer Science. Our conversation today is about artificial intelligence. And welcome Zona.
Dr. Zona Kostic: Hi. Hi everyone, nice to be here with all of you. And thank you so much Bjorn for having me.
Dr. Bjorn Mercer: Oh, of course. No worries. I’m excited about this conversation. Of course I know zero about artificial intelligence besides articles I’ve read. And so I’ll go ahead and jump into the first question. Can you explain what artificial intelligence is?
Dr. Zona Kostic: Yes. Okay. Well, that’s a really tough one actually because we always find a hard time to explain what AI is, given that there are so many different terms right now. There are so many different models, there is also terminology such as machine learning or big data.
So, all of these terms kind of point ours the same way, which is how we are using data, massive amounts of data to resolve certain problems or to actually expedite coming up with certain answers.
So, in terms of artificial intelligence, it actually uses machine learning as modeling principle and artificial intelligence should be seen as an application of ML rather than exactly the same term.
I also mentioned big data and big data in all this family of machine learning and AI applications is basically focusing on large amounts of data that couldn’t be processed with current machine learning technologies that we are using right now.
So for example, if we want to process a very simple data set, figuring out an average income, we can do it using ML applications. But if we are talking about average income, for example, the entire United States, or even if we want to come up with much bigger discoveries, we will need to employ big data applications, which are of slightly different than AI. But as well as all of these machine learning big data is part of the same family. So we are processing data in order to come up with a certain asset.
Dr. Bjorn Mercer: And that’s wonderful because honestly, for me, when I heard artificial intelligence, machine learning, I didn’t know how those two related or the same. And of course we’ve heard about big data for a while. And that totally makes sense.
And it totally makes sense that only recently, within human history, that computers are only now starting to get fast enough to see some results. So as a follow-up question, how far are we from seeing some real tangible benefits from say AI, you know, because of the advancement of computers?
Dr. Zona Kostic: Yes. I also forgot to mention that having a computational power is one thing, having large amounts of data, it’s sort of a trick here because we’re fighting right now, misuse of data. Yet at the same time, we need large amounts of data in order to be able to come up with certain results.
So, on a side, having a computational power or working with large amounts of data, we can see some interesting, even fascinating results with artificial intelligence. Not only when it comes to gaming, because this is one of the most interesting field where we saw that artificial intelligence or automation is capable of winning most of the games, actually winning in the competition with world champions. For example, AlphaGo is one of them.
But what’s also interesting is that we are seeing a lot of improvements in certain fields, such as medicine, or as well as education. We are, at the APUS here, we are working 100% online. How we can use AI to actually help our own students, not only computer science students, but all the other students. So these are all of these domains where we can see breakthrough in AI.
Dr. Bjorn Mercer: It’s great because as myself in what I do here at APU, AMU, is I would love to have a crack at big data so I can help better craft the learning experience for students. But it seems like at universities, and of course, most schools out there, there’s just so much data that somebody like me has no idea how to sift through it all and to really get usable material.
And that’s where I’m sure a computer scientist, like you, helps come and write the code or the algorithms that help us make sense of just so much information, even something like adaptive learning.
Dr. Zona Kostic: Or automatic assessment, for example, this is where we all can benefit from.
Dr. Bjorn Mercer: Yeah. And there’s just so many things that I’ll say currently, higher education is still lagging because it takes some real capital, of course, to develop this. And so that leads me to the next question. What are some benefits of AI for the average person? And what are some of the benefits of AI for say corporations and government?
Dr. Zona Kostic: Well, it seems like we have a completely different needs and different benefits. So actually corporations are benefiting from average people. Government is still fairly closed. So the biggest issue that the government has is they’re trying to reach this goal and become at least powerful when it comes to AI to compare themselves to China, because China has this big goal of becoming AI world leader by 2035.
And this is pretty far away in the future, even in these tech years. 2035 is not something that we can actually predict right now with enough certainty. But the government has a pretty big issue right now is that they’re closed.
They have a really big established bureaucratic system and change happens really slow. So I see a certain shift that government started employing corporations to help them resolve certain issues, but also government cannot share data as easily. So there are certain problems over there.
An average person, you know, average person can benefit from many different perspectives and can benefit from many different applications. For example, I’ve discovered recently an application for this new normal. So, it’s like a COVID-based application where users can just switch it on and go outside walk. And the application will actually try to figure it out, is there any person nearby who might be having COVID? And they’re going to do it based on coughing.
So right now the company’s asking kindly to submit as many samples as possible so their own application could be as accurate as possible. And of course, all of our coughing samples will be anonymized, but I can see that as a really good example right now.
Dr. Bjorn Mercer: Yeah. And as far as COVID, any kind of way in which computers can help track and predict a possible spread of COVID is of course a benefit. So a follow-up to that question is what are some potential privacy issues, because to be able to have that kind of usable data, I’m assuming, say, our cell phones or whatnot would have to ease drop on us, which the benefit is if you have it set up correctly, you instantly know potentially who’s sick. The downside of it, especially for civil libertarians, is somebody’s listening to you.
Dr. Zona Kostic: True. There are applications that are also trying to help. Basically they’re bringing benefits to the average person while making sure not to violate privacy and to keep the data safe.
Right now, there is a new domain called Federated Learning in which we are basically training this main machine learning algorithm on servers using completely anonymized data. And then if we want to benefit from the data of our own users, then we are going to retrain model, but not using their own data.
We are going to just retrain model on their own cell phone and benefit from so-called weights or a new intelligence and update the major model with new intelligence. So I believe we can all benefit from this type of AI and still be sure that our data is only with us. And it basically stays on our own cell phone.
Dr. Bjorn Mercer: That’s fascinating, and I’m glad we talked about China briefly, and you said they have a goal of being a leader in AI by 2035.
Dr. Zona Kostic: Yes.
Dr. Bjorn Mercer: China is so interesting because if they put their mind to it and if the government supports it, then the full weight of the Chinese government is behind it, which you know. Being an authoritarian communist government, means that everybody’s on board.
Of course, one of the challenges of the US is that everything is decentralized. The government has power, but limited power, which my personal opinion is a much better way than an autocratic communist government.
But what do you think some of the challenges of the US will be to be able to compete with China? Is it, say, the free market and allowing for innovation with corporations and businesses?
Dr. Zona Kostic: Yeah, I think the biggest challenge, again, is going to be education because we are actually providing the best possible AI education, including a fairly big number of Chinese students.
So these students are educating in our own schools and we are not really taking care of them. We are not making sure that they are going to stay. So I think this is going to be a biggest challenge.
So are we going to ring our own education? Of course not. We need to figure it out. How my personal idea, and it might be not something realistic, is to think of how we can collaborate, how we can basically benefit from the best of both worlds, which I’m pretty sure is possible.
Dr. Bjorn Mercer: Right. And to me, again my own personal opinion, there’s absolutely no reason why anybody can’t collaborate. Especially AI, yeah. Assuming it comes down to bad faith politics, if somebody says, “I can only win.” And the other side says, “I can only win.” And then that just creates conflict, but humans. Humans have conflict and have always had conflict.
Excellent. So the next question is, are there any AI applications specifically useful for the new normal?
Dr. Zona Kostic: Oh yeah. For example, when it comes to COVID, I mentioned the coughing app. There are applications who are actually employing computer vision machine learning systems in order to understand, based on the first MRI images who might be a congested person.
We all know that there is a pretty big sub domain of AI in drug discovery, as well as gene sequencing. And this is where probably AI is also making another breakthrough with this new vaccine that is going to very, very soon, if they didn’t do it.
On a side of medicine, so maybe, we do have COVID is a really important problem to solve right now. And of course the most interesting topic would be to talk about the applications of AI in medical domain, but I’m not a medical expert. So this is why I’m like, “Okay, I know that there are applications. I do have a lot of fellow post-doc researchers who are working in this domain.”
I’m also interested personally, as a researcher, as an AI researcher in applications that are coming from some other domains, such as VR or AR applications. Education, of course. This is going to be always my first love and my primary goal to apply AI in an application domain.
So we have Zoom. So we didn’t actually mention, but Zoom is one of the most appropriate AI applications. And we think it’s just the application for teleconferencing, but it is backed up with a lot of machine learning algorithms. Let’s just use for an example, all of these virtual backgrounds. So they’re using machine learning algorithm to actually help us stay within a nice and comfortable room.
We also have, I mentioned, automatic assessment. Cell phones, these days, they’ve been used for so long, but even right now, we’re seeing so many cell phone-based applications. And when I say cell phone based applications, I do mean AI as a backup of this app.
So we don’t have libraries anymore, for example. This is one of the research projects that I was working on. And how are we going to bring libraries back? I’ve seen in my own neighborhood that there are so many of these small house libraries. People put like a small house with a couple of books, so you can freely go and just take a book and bring it back after you finished reading.
So right now, there is this idea of if I touch too many books, maybe I’m going to put someone else in danger, but I really want to take a look at this book and figure it out. Is it according to my own personal taste?
We developed the application in which a person can just take their own cell phone, take a picture of a book spine and obtain all possible information about this book. So, they don’t even have to open it and then, take a look at it, and type the title. All the information can be taken, it can be actually retrieved based on a book spine.
Right now, we are working on, if we have multiple book spines, we can compare all of these book spines, and recommend books that are good for this specific reader. So I believe a lot of cell phone-based applications are those that are going to be interesting for this new normal.
Dr. Bjorn Mercer: Oh, for sure. Just speaking about education, it’s amazing how people use their cell phone for a lot of what they do as far as learning goes. For me, that’s very difficult. I need a computer. I need something larger, at least a tablet to truly kind of delve into information.
But I know for a lot of people, when it comes to learning, they do the forums or discussions on their cell phone, which is fine. They do a lot of reading on their cell phone, which is, of course, great. It’s maybe extremely difficult to do actual assignments on a cell phone.
Dr. Zona Kostic: That will be tough, but iPads are actually possible for it right now.
Dr. Bjorn Mercer: Yeah. And iPads and Chromebooks and things like that, which are not as powerful as regular computers, but they do enough for education. So since we’re talking about education so much, what are some ways in which you try to encourage people to go into computer science?
I mean, to me, it’s an easy sell because the future is built on computers, but I think some people might be hesitant or a little intimidated by computer science. So what are different ways in which you talk about it to people?
Dr. Zona Kostic: So usually I believe the best possible way is to start with examples. People might be interested in computer science or artificial intelligence as a field because it is something attractive right now. People are talking about it, but they don’t really know, will they really feel comfortable in this area?
So something that we did as part of the School of STEM, for example, is we created a first course that every single student, no matter what kind of program they are in, first course called STEM 100, where students are going pretty much through all of our programs. Each week, they will dedicate themselves to one domain.
So first week is going to be about computer science. Then, we’re going to talk about chemistry or physics or math. And according to lectures, labs, as well as assignment posted at the end, students can really understand, “Okay, this is something that I’m interested in.”
So first explaining what computer science is. Not only what computer science is, but what computer science is possible to do in order to make certain change. Right now, if we talk about currently attractive roles, this is also one of the interesting topics that I like to talk about with my own students, like what industry wants right now? But what industry wants right now, they might not want in in four years.
So what will be these future roles is more interesting and then asking them, “Would you be interested in really tackling this problem?” And also I do like to collaborate with industry a lot. So I like to call industry partners and usually industry partners can see slightly further away in the future and they can tell us as well, “Okay. These are some potential roles. Okay.
Data science is interesting right now. There are so many data scientists on this market. Should we take a data science as a new math? And then add one more domain on top of it?” That’s one of the ideas too.
So collaboration with partners is what I believe it’s opening all of these future roles and helping students figuring out are they really interested in this field or not?
Dr. Bjorn Mercer: No, it’s excellent. And that’s interesting going from math to data science and adding another domain because as things advance or as we discover, or life becomes a little more complicated, say the math of yesterday is still valid, of course, but for the world of today, we need to add a little more, a little more to of course be competitive.
I guess another question for you, do you find when you talk to younger women, that they are interested in computer science? Do you find that they’re not? I mean, it’s such a challenging question, but there’s always the representation of females in Google and engineering and that it’s a very low percentage.
What do you find is hesitancy of some women to go into something like computer science? Is it just because it’s generally mostly men? It’s obviously more complicated than that, but usually when you listen to media, they distill it down to the simplest possible explanation.
Dr. Zona Kostic: Well, maybe I actually live in my own bubble, but I don’t see it as a problem anymore. First of all, there are so many girls, almost half, if not even more than half, as part of the computer science, undergraduate and grad schools.
Whenever I go for a conference, for example, I see the women present and not only on certain roles. So right now we have sort of an equal spread, but I’m, as I said, I’m definitely in my own bubble because for example, the country I’m coming from, which is Serbia, it’s a country in Eastern Europe, we are still not seeing women, not even as part of college, not to mention as part of industry over there. So I do see change.
I don’t see any sort of fear, which is great news, actually. Great news is that there is no fear on neither of two sides. Neither girls are afraid anymore. I don’t think boys or men, they were ever against it, so to say. So I’m seeing a really nice collaboration.
I, personally, in my own research team, which is not really nice, but I don’t have a girl yet. Oh, actually yes! I do have a first girl. So most of research collaborators, grad students of mine, are males and they don’t have an issue with having a female of a leader.
We assembled this team accidentally. I mean, I just figured it out. I’m a bit biased. It’s not really nice. I have to change that, but it’s problem I didn’t really think about it because I don’t see it as a problem anymore, which is great.
Dr. Bjorn Mercer: That is great. And I mean, the one thing that, as we progress in time, I guess you can say, is the goal is to there not be a problem. Of course. And today, if one had a conversation, it’d be really hard to imagine somebody being like, “Oh, you’re a girl? You can’t be in computer science.” I mean, that’s almost unimaginable.
Dr. Zona Kostic: One thing that I haven’t mentioned is that the only thing that it’s really hard, given this topic right now is for example, if we take a look at job postings, we see, “Oh, we really want to include more women. Oh, we are giving the advantage to women.” And I’m asking why?
We don’t have to do it anymore. I mean, the doors are finally open. It doesn’t even matter, are we going to have more women or more men within a certain team, as long as this team is going to work. And I also do believe in balance.
Dr. Bjorn Mercer: And just going off of something you said a little while ago, I was a little surprised that in Serbia, you said that women aren’t really part of college, which Serbia having been a communist country for such a long time, it seems like at least the one thing communist countries did was try to make education even. But was that not that way in the former Yugoslavia?
Dr. Zona Kostic: True. But unfortunately in former Yugoslavian countries, if we exclude Slovenia, for example, in former Yugoslavian countries we are there, still there in this communist period of time, which is we haven’t changed anything in the past 30 years and I can speak about the entire region. So, it’s not about Serbia only.
We used to be part of the same beautiful country you mentioned called Yugoslavia, and I’m saying beautiful country because we used to live in the same country with so many diversity, different languages, different people, different religions, the Catholics, Orthodox, Muslims. We all used to live in peace.
Right now when it comes to former Yugoslavian countries, I could definitely say Slovenia has a really good educational system. Croatia is trying to catch up. I’m not sure they’re still they’re there yet. When it comes to everyone else, we’re still stuck. And there is not any promising future, at least for right now.
Dr. Bjorn Mercer: Serbia and the entire area, for me personally, is a very fascinating history besides the, of course, tragic wars in the 90s, being part of Yugoslavia and being communists. But outside of the sphere, I guess you could say of the Soviet union to a point, and then even before that, being the powder keg of World War I. And then before that, all the Balkan Wars with Greece. And then even before that, the Ottomans.
Dr. Zona Kostic: We do have a very, very big history, but I have an impression that every single time we talk about Balkan it’s a Balkan chaos or Balkan war, you know? I would really like to hear there is this Balkan initiative when it comes to ML, AI, bringing more women, inclusion.
Dr. Bjorn Mercer: No, I completely agree. I mean, that area is such a crossroads of so many different, putting big quotes, “empires” that seeing that area truly prosper is of course the goal. It’s the goal of every country. Every country just wants to be left alone and just prosper because of just the way history goes in politics. You know? I think it’s tough for some countries.
Dr. Zona Kostic: Well tougher for some than the others. I mean, what kind of cards do we have in our own hands? And then based on that, you are making your own first move.
Dr. Bjorn Mercer: Exactly. And we can have an entire philosophical, historical conversation about Serbia, and that area. That could go on for hours. Now, your background is art or graphic design, correct?
Dr. Zona Kostic: That’s my first bachelor. Yes.
Dr. Bjorn Mercer: Yeah. So how did you get into AI coming from an artistic, and say graphic design background?
Dr. Zona Kostic: My second bachelor was in computer science. I was always having a need to bridge these two domains. That was my ultimate goal. And I believe that I’m finally able of doing it.
So first trial was during my graduate studies, as well as in my own PhD pieces and my own PhD project mostly before the thesis was about how we can create virtual environment, which is strong CS background needed for it. In order to help on this, then students learn graphics design. So, this still is a computer science to support graphics design, was still not the combination of both domains.
When I say finally these days, I’m able to combine this domains in a completely new sub domain called Explainable AI. So there is this pretty big problem with artificial intelligence is that algorithms are really easy to be deployed and used.
So pretty much everyone, even if a person is not a domain expert, they can go online, upload their own data easily, use machine learning as they use to install any program using next, next, next, finish. I want to do this. I’m going to use my own data, upload data, predict, finish. Without even knowing how machine algorithm works. Right now, this is one problem when it comes to AI.
Second one is that not even AI scientist know how machine learning algorithm came up with its own decisions. So, we pretty much have to figure it out how to get into this inner stages of AI and understand how AI thinks and Explainable AI as a new sub domain is really interesting right now. Why? Because it employs a lot of visualization.
So this serves as a mediator, visualization as the mediator between AI and a human being in order for us to understand how AI thinks. But also we are using visualization to give input to AI. So as soon as we have this input and AI is changing the direction, how it’s going to come up with the next decision? So finally, we have the combination of arts, of visualization and computer science in one sub domain called Explainable AI.
Dr. Bjorn Mercer: That’s very fascinating and being able to visualize data, it’s a very palpable thing for people to understand. And there’s been data visualizations for awhile now. And I think people really respond to them, even if they have no idea how those visualizations come about, but the sheer amount of data that some visualizations.
Dr. Zona Kostic: True, And I think New York Times did a great job when it comes to it because there are so many interesting articles, but sometimes we really need to be domain experts in order to really grasp what’s going on behind this article. And then of course we can have time. And then we want to just easily swipe and be able to understand and answer certain questions in a shorter amount of time.
And what New York Times did, they impose data visualization. So even if I’m not domain expert, before reading the article, I have a small map. I can interact with this map. I can get engaged. I can learn more by filtering and clicking. And then, once I kind of get the big picture, then I can proceed with reading the article. So, that serves as most like an opening, powers a domain that we, as a general audience, might not be experts in.
Dr. Bjorn Mercer: Oh, that’s great. And then, this is a final follow-up question. With the election having completed at the time of this recording, was machine learning used when trying to predict voting trends? Because one of the things that really surprised me was just personally Pennsylvania, with the night of the election Trump was ahead a lot. And then they said, “Nope, it’ll swing and probably go Biden.”
And for me that was like, “He’s up by half a million?” But I could see how it’s crunched big data so much to know that every single precinct will vote a certain way. And that’s obviously much larger than just having a really big Excel chart.
Dr. Zona Kostic: True. And I think everybody would pay a lot of money in order to have such a big data, AI algorithm that’s going to help them understand, “Okay, with this, with 99% accuracy, this is the expecting straight, and this is how many votes we can expect in every single country.”
The problem is, and yes, there are so many machine learning algorithms involved in election process as a whole. And when I say process, I’m thinking this as a greater picture than we have Facebook, for example, that uses AI to maybe influence certain votes. It uses AI for the advertising. We are also using AI in order to predict certain amount of votes.
But the ultimate question is like, “Why are we all failing in it? Why do all of these magical, great big data algorithms fail?” Is because they ultimately depend on data and pollsters actually failed to come up with with good samples. That might be one of the reasons. And second reason might be that somebody didn’t really want to go, but I will stick with the pollsters.
Dr. Bjorn Mercer: Yeah, the pollsters especially and it makes sense that the the margin of error is much larger than just two and half percent these days. And in that time that person might actually might not tell the truth, or just because of the election of 2016 and 2020, Trump is a figure, which is a little more complicated than just straight D or straight R, straight Democrat, straight Republican. And so it’s just more complicated.
And especially when looking, like you said, with the data, if you’re looking at historic data that doesn’t mean current data.
Dr. Zona Kostic: True, true. I mean, we can definitely benefit from historic data in certain percentage, but right now, in order to come up with an accurate predictions given this current elections, is we really need to come up with good samples. We cannot go and interview everyone. Then, we don’t even have to vote, right? We will have this final decision.
What we can also do, and you actually made a really good point, sometimes we cannot be certain about it because whomever is interviewed, he or she might change opinion. So maybe they’re going to say, “Okay, I’m definitely going to win award for one side.” But then they change their own mind at the end. Or they just don’t want to say, they just don’t want to be honest in this respect.
So what we can do? We can also use AI to actually basically measure, for example, face, or actually have a facial recognition algorithms that they’re going to tell us. This person is telling truth, or it’s not telling truth, but this is definitely something that we shouldn’t even think of as a responsible engineers, right? Because what will be the influence of it? We all see negative implications of this side of algorithms.
Dr. Bjorn Mercer: Right. And it’s the final thing, that makes me think of China right now where they’re doing their social currency. I believe? I can’t remember exactly what it’s called, but each person gets a score.
Dr. Zona Kostic: That’s right. That’s right. It’s like a very Black Mirror-type of score in which not only they’re following people on the streets and using facial recognition algorithms. But if they go to a liquor store, for example, they’re going to get negative marks. And if they want to go to the bank and ask for a loan, their interest rate is going to be much higher because they had such a bad behavior recently.
Dr. Bjorn Mercer: Yeah, and again, from a Chinese authoritarian communist state, it makes sense that you would give everybody a score. It makes things very simple. This person has a score. This person has a lower score.
Obviously the complications are what happens when people get such low scores that they’re now outside of society? The state still has to deal with those people. And the other thing is, I remember watching an interview with one of the developers in China, and the person asked the question, “Do you see any ethical issues with this?” And the person was like, “No.”
Dr. Zona Kostic: Probably not.
Dr. Bjorn Mercer: Yeah. Over there, which even if over there, maybe sometimes they’re hesitant to say it actually on television, but there’s huge ethical issues with all of that. Again, facial recognition, different things like that, giving somebody a score, I mean, again, most people just want to be left alone and just live their lives.
Dr. Zona Kostic: Yeah. It is all happening, as we like to say, behind great China’s firewall.
Dr. Bjorn Mercer: Yes, for sure. Well, excellent. Well thank you for the absolute wonderful conversation. Any last words?
Dr. Zona Kostic: Well, thank you so much. Thank you so much for everything. And I really hope that we are going to see more and more, very good and useful beneficial AI applications, as well as that our own future generations, not only when it comes to computer science students, but I mean, all of our students will be able to tackle some of these challenges that we’re facing right now. And thank you so much.
Dr. Bjorn Mercer: No, for sure. And today we are speaking with Dr. Zona Kostic about artificial intelligence.
About the Guest:
Zona Kostic is an experienced faculty with a demonstrated history of working in the higher education industry. Her professional interests are at the intersection of data visualization, machine learning, and digital realities. Zona holds two bachelor degrees, in fine arts and computer science, a Master’s degree in computer science, and a Ph.D. in engineering systems (University of Belgrade).
She conducted a postdoc research in immersive visual analytics with the Visual Computing Group at Harvard University. Zona has published six books, and a number of research works at scientific journals and conferences. As an innovation fellow at Harvard Business School, she co-founded ArchSpike, a startup that integrates data science and visualization with market modeling, allowing users to design buildings that better respond to future market demands. Her professional contributions were awarded with the Distinction and Excellence in Teaching Award for the course on Advanced Methods in Data Science at the Harvard John A. Paulson School of Engineering and Applied Sciences.