(Intro Music)

Host:Thank you so much for coming today. So, to start off, can you tell me a bit about your journey to STEM and just a bit about yourself?

Kevin:Sure. My name is Kevin, I am a CS (Computer Science) student at UWaterloo, I’m working part time right now as well. I had one co-op experience before and I got a lot out of it—I’d love to talk about it. Regarding how I got into STEM, I think from a young age I kind of wanted… I was just curious about a lot of things and I wanted to do things that would have a lot of influence. There were a lot of opportunities in STEM, and I was also, academically-wise, more STEM-focused: I did mainly math competitions throughout high school. Towards the end of high school, mostly during the COVID era when people were all at home—it was partially because we didn’t have much to do, but also because for a long time I had a lot of coding projects I wanted to do, and I was thankful for the opportunity because we were all home and there weren’t many other things to do at home. Coding projects, in general, are online, so there’s no requirements. It’s also easier to do alone, especially for smaller projects, so that’s why I thought it was the perfect opportunity to jump into these things, and that’s how I had more and more experience with CS. That’s how I ended up in STEM.

Host:So you mentioned that you’re working part time. Can you elaborate a bit more about what you do on a daily basis?

Kevin:Yeah, I work part-time at a startup called Skinopathy. They’re a medtech startup: medical technology. So what they do is—some of the products they provide are regarding apps that can basically take a photo of your skin and use screening. Usually it’s for, say, acne, scars, etc. and other kinds of skin lesions. The AI tags it and rates it at high or low priority on a scale of 1 to 5. If it is a higher priority, it just notifies the physician as soon as possible to give them notice of it being high priority. As for me, I do full stack. Currently I’m working on a service that builds billing. They also have [other medical services], so I’m trying to set up the billing side of that.

Host:Thank you so much for elaborating on that. Just considering how much time your part-time job takes, and also being a university student, how do you juggle those two things together?

Kevin:[I’m] not really sure. I think I just—spend less time at school, and sometimes I try to min-max school, which is probably not the best way to do it. Maybe a few pieces of advice: the things I’ve heard are generally “work smart, don’t work hard.” I think that’s a bit overused. I do know a lot of people that study a lot but they don’t get too much out of it. They just cram, which is one thing I highly recommend not doing. Another thing that I’ve generally found helpful is—I guess for some [projects] it’s worth studying in groups and others are worth studying individually, and all courses are a mix of both. In general, that’s a very effective way to optimize how much you’re getting out of each unit of time, if that makes sense. For every hour, you want to get as much out of it as possible, because you only have twenty-four hours in a day. One thing it would be helpful to focus on is how much you’re putting in—studying individually.

Host:Yeah, of course. Having peers to rely on and team efforts [allow you to] work and optimize things.

Kevin:Mm-hmm.

Host:So just going back to your high school, how did you find the transition from high school to university and how did you adapt to some of those changes?

Kevin:For me I think content-wise, the difficulty went up a bit, but there was overall less work. I found that for high school, what I struggled with was the sheer amount of work. I did IB (International Baccalaureate), and it was more ‘quantity over quality,’ honestly. I think some of the things are—they just want you to do more work rather than learn more useful, practical, or even cool theoretical concepts. In uni it was more that you had less to do, but overall harder. On the side of actually living and [university life], this doesn’t really apply to STEM, but for most of us—especially at Waterloo—because Waterloo is kind of in the middle of nowhere….actually, it’s relatively close to Toronto. Most of us live [on-campus] so compared to say, UBC, or anywhere there, SFU, from what I’ve heard most people live [off-campus] or they live from home. For us, even if we’re off-rez, it’s like student housing. We’re pretty close to each other, so it’s overall a very different vibe. I think as a consequence of that, there’s a lot more ‘adulting’ that’s required: living alone, dealing with things here and there, and I think that for me it was a bigger change than school. [The living changes] probably deserve more attention than the school content, because most students have been studying for—what, ten, fifteen years, but they haven’t been living alone for that long. I think that’s something that I think people should be paying more attention to.

Host:There’s definitely a big social change once you move to another place and have to realize that you are by yourself and [it requires] individuality. So given that we had the pandemic and that was a big social change, how was university online and how did you adapt to that?

Kevin:I think it was okay, actually. It was honestly less work, because it was online, but it’s just harder to learn certain things. For me, for 1A—my first semester—I had my two math courses in person and everything else online. Then for 1B, it was basically in person, after about one month of online classes, because in January [of 2021] everything shut down. 1B in my opinion was pretty much all in person. We did have that month online, but midterms, finals, and the main lectures were all in person afterwards anyway, so I guess I won’t discuss that too much. I think 1A online—one thing that was really helpful was that I was on-rez and I had people to talk to, because I think that’s a bigger part. Again, the studying: a lot of us had already been studying online, with the last year of high school, and I don't think it’s too different from that. There might be hard concepts occasionally here and there that are harder to pick up online because they aren’t delivered in the same manner, and you can’t just put your hand up and interrupt. Well you can, but it’s harder online. I think regarding that, maybe it would be worthwhile finding a group of friends in the same classes and finding time, an evening off [to work together], or whatever.

Host:I’m interested in the skin detection [technology for Skinopathy]. How does it account for bias when you have, say, darker skin tones, and how does it recognize when there’s a flaw within that tech?

Kevin:I mean generally, if you use any of the modern algorithms for skin lesions, classification detection, etcetera, I don’t think that should be a problem because—well, a few things. I guess I can imagine, for example, things like scabs, right? Usually they’re a different colour, so I guess you could say that a darker colour of skin could make it harder to see the scabs. But usually for classification detection, everything to do with images, you use CNNs (convolutional neural networks) and they adapt very well, like they’re basically just pattern recognizing algorithms. The background colour isn’t really a problem when it comes to that because you still see the contours, right? Even if you do something as simple as just drawing out the contours—also, I need to clarify, I’m actually not working with the skin detection algorithm itself, I’m working on the development and that side of things. But anyways, I think that it wouldn’t be a problem because even if you just draw out the contour lines or give some sort of contrast to the algorithm, it would make it very obvious that for [something like] scabs, no matter how dark your skin is, you will still be able to see the scabs. I don’t know if that makes sense. I think there is still a very obvious difference in darkness between the scab and your skin. Even then, there are also other patterns that you can recognize. For most people, from the perspective of an image, their skin is just one flat colour, and then compared to features such as scabs, lesions, everything like that—there are fewer things there. One thing that is more complicated than dealing with biases such as skin colour or skin tone is size. If you’re moving your camera, depending on how close your camera is to your skin, that could become a huge problem. They are doing things to counter that, I’m not sure if I’m allowed to talk about them [laughs] so I’ll just leave that.

Host:Alright, thank you for elaborating on that. I was just curious about the technology.

Kevin:Yeah, for sure.

Host:Okay, why did you choose computer science out of all the sciences out there?

Kevin:For me, just from a young age, I thought it was something that was really cool to do: coding robots and other projects. I started learning to code early and everything along those lines, doing little projects here and there, and I think it was maybe towards mid-high school where I found that it was also a booming industry. There’s a lot of opportunity there, and that’s kind of how I settled on this field. I also tried to self-learn it and ended up really liking it, so it worked out in the end. Well—so far [it’s worked out]. It’s only been a year and a half of uni CS.

Host:Thank you. Good answer, I like it. You mention it’s a booming industry. Do you think because of other machine learning [tech] or AI per se, do you think there’s still room for humans in technology? We all have that conceptualization that we’ll be replaced by robots, so what [kind of role] do you think humans play in that shift, or perhaps not that shift?

Kevin:I think firstly, it’s kind of more short-term, and there’s a lot of little things there that machines absolutely suck at. For example, like, one simple thing is walking. There’s so much research in robotics about getting robots to walk on two feet, and it’s actually—especially fast, right? I think the world record is almost human walking speed, basically. This is on cement. There’s no stairs, there’s no obstacles, there’s nothing complicated. Machines just suck at it—well, I don’t think machines suck at it. I don’t know why, in general, bipedal animals are so good at not falling over, like just logically, I think the main reason is that we have a bunch of receptors and everything. We have stuff like proprioception, which is like… say you close your eyes and raise your hand. You know your hand is raised. You don’t see that your hand is raised. If you consider the five senses, [proprioception] is part of feel. Usually people think of touch: I touch my monitor, I feel it. But nothing’s touching your arm. Maybe your clothes, but you still know your hand is risen [regardless]. Things like that, we haven’t classified or implemented into robotics to make them robust enough to handle complex actions like walking on stairs, around or over obstacles, stuff like that. I just mean that most things so far won’t be replaced for a while, I think. I’m not familiar with the robotics industry, but I guess this is still very much on the manual labour end, despite ideas like “are we going to get replaced?” In my opinion—I’ve done a lot of research in CS on natural language processing (NLP), which is like language AI. So I think some of the things that might get replaced or at least heavily intervened with are things to do with very mechanical natural-language writing. By natural language I mean [languages] like English. That’s a natural language. Any human language, really. I can see, for example, those writers being replaced. You’d be surprised at how good the essays written by these bots are. One thing that researchers emphasize with NLP is that these machines, so far, all they do is rearrange words, change the meanings, etcetera. But all of that is done automatically, so there’s still a huge element of intelligence. It’s not hard-coded, where the human decides “let’s move this around,” it’s more under the hood. Because with all modern ML (machine learning) models, nothing is really hard-coded—or not much, these days. It’s mostly that machines learn to do certain things. For example, GPT3. You give it an input and it gives you an output. It’s called sequence to sequence, which means you give it input text and it gives you output text, and usually it involves something called prompt engineering. That’s where you tell it what to do, like “write this paragraph,” and then give it a paragraph. The AI summarizes it. You can give it a question, and say “this is a question,” give it an answer, say “this is an answer,” “this is a question, this is an answer,” and then you ask the question you really want to ask, and [the AI model] would give you an answer. You can get it to do things that are similar to question and answer logs. There’s all sorts of things you can do, but those are some of the examples. That’s why I think more of the mechanical things can be done [with AI]. Obviously not creative writing—I mean, it does pretty good creative writing, but I think it’s not very genuine. Because it’s trained on the Internet, [the model has] probably just copied off some parts of stories here and there and combined them. Although you can argue that regular fiction authors do the same thing, kind of just take parts and pieces of fiction and make it their own. Of course, there are parts that are more genuine, but I guess the point is that the more mechanical things: summarizing, maybe writing abstracts, little things here and there will start to be replaced. I’m trying to think of jobs. I doubt that jobs like journalism will be taken over, but some of the things I read look worse than the bots’ writing does. It’s just so mundane. Like “this is true”—there’s so much repetition. This person—by the way, he’s 22—has done this, and then later on they’re like “oh, this guy goes to university” or something. And…I don’t know. I think the writing style is very boring. I’ve seen bots write better things than that by a long shot. I don’t know what profession writes stuff like that, and even if it is, I feel like it’s already been automated by bots given how mundane some of the articles are. But that’s one field where there will be a lot of machine intervention. One of the common things I’ve heard is the blue-collar jobs of the future will be humans doing recognition for machines for training data. So, basically reCAPTCHAS. I can see that. It’s kind of—I’ve already seen similar things. For research, I find this kind of unethical, but honestly I don’t know how we can get around this. They use a tool called Amazon Mechanical Turk, or AMT. It’s for generating test data or evaluating their machine learning models. For example, for NLP this is what they usually do. Say I want to make a better creative writing bot. I try something new, and then I give the original state of the art and our state of the art and ask [people via AMT] which one is more creative. Or we give them individually and ask “rate this out of 5: how creative is this, how fluent is this,” everything like that. People are paid to do this. You have people going on AMT and filling out these surveys, getting money for them. I was reading about how ethical this is, and I guess on one side there are people who need that extra money and it’s a quick way to gain some money, but at the same time it feels like it’s pushing more towards making such things a full time job, which would be very concerning. I think that’s the bigger thing that I find [worrying], than maybe some of the other automations. Also, self-driving is another thing that a lot of people are concerned about. I’m not too sure regarding how close we are, but I’ve talked to people and they said it worked. So far, there are still cases that machines can’t handle. So, like a stop sign. The O, if someone vandalizes the O: draws something on it, puts a sticker on it, etc. the machine can’t recognize that stop sign. Not always, but sometimes. Issues like that, the machine handles really badly. That’s why I think self-driving is really slow. Although in my opinion they should fully transition soon, because it’s already significantly safer than humans. If [everyone] switched right now we’d save so many lives. I’d rather have the one or two occasional mistakes where someone vandalizes a stop sign than the millions of deaths every day.

Host:Alright. Sorry, I have a quick little anecdote. You’re right about the creative writing. They had a human audience read poetry by robots and poetry by humans and they could tell the difference. So you’re right with that.

Kevin:So you’re saying as in, it’s very noticeable—

Host:It’s noticeable: the difference in how they write.

Kevin:Oh, yeah.

Host:Given that you were part of Millennium STEM, how do you think that helped you progress within your career and your life so far?

Kevin:I think most of what I was doing was working on the website and whatnot. I think the billing service as well. I think that just gave me a lot of experience working with this type of thing, so that was really helpful. Regarding co-op, I knew a bit more than everyone else, so it made the job search easier, I think. That’s one thing. And more experience is always better. I know what I like, I know what I don’t like to do. I know what works regarding web development, and I think that’s the main thing for me.

Host:Because you had that opportunity, what would you recommend to other youth who are wanting to get into STEM but don’t know really where to start?

Kevin:Okay, I’m a bit more biased with CS. In my opinion, it’s probably the most learnable thing on your own. You don’t need classes, school, whatever. If you want to do physics, for example, you kind of have to go all the way to, like, a grad school PhD. Otherwise, it’s not very possible to go into academia or any research without that. I think that’s true for sciences more generally—although engineering is [moreso that] you can get a bachelor’s degree and work immediately. For CS it’s very self-learnable. I think if you go on Stack Overflow there’s a survey—I can’t remember exactly, but a huge fraction of current developers are self-learned. It’s an impressive amount. I guess how to start would be to go on YouTube and learn Python. I can link [you to] the YouTuber that taught me. His name is Corey Schafer. He explains it really well. He gets the fundamentals down quickly and slowly at the same time. That’s what got me into it quickly, and then you can look through some cool little projects to do. The thing that Python can’t really do is web apps. If you know HTML, it would make it easier to do small projects with it. One of the projects I did just in Python without using all the other stuff is a scraper for COVID cases. I made it so that when COVID cases in BC hit above a certain point it would notify me through SMS (text message). That’s one thing I did. You can find a lot of similar projects, maybe do the same thing for your province, city, region, whatever. I think there’s also a lot of room for improvement with that project but I just never really got to it because COVID kind of died down by the time I was finishing up. Just start with little projects, then build into the bigger ones. After that, probably learning Node Javascript. Then learn the general full stack apps. It’s kind of a thicker learning curve, but I don’t think it’s that bad. You get a lot out of it, because you can apply it to ninety percent of full-time jobs after learning those three things. That’s where I would start.

Host:Thank you so much for [discussing] those opportunities and some ways that are accessible for everyone to join pure science. I remember at the beginning of the interview you mentioned that coding can have a big influence. I want to know: what inspires, or what excites you, about CS?

Kevin:There are two things that I really like. The first is that after you finish a product, there’s a big [sense of] satisfaction, and it feels really nice to see what you spent your time building. It’s a great feeling. It applies not just to CS, but to any form of engineering, anytime you’re building something. So there’s that satisfaction. The other thing I think is that there’s so much potential right now. There are all sorts of automations you can do here and there. I guess it’s kind of against what I was saying earlier about taking away jobs. It does take away jobs, but it does save lives: autonomous driving, stuff like that. That’s what really excites me. There’s also the aspect of—I like the idea of automation because there are certain things that, in my opinion, should be automated, so that people can do jobs that they want to do rather than things like this. There is an argument that it takes the jobs away from them, but I read this book called “Economics In One Lesson,” and one of the things they said was that there’s something that’s often missed with [the argument of] taking away jobs. Usually when you take away jobs, in terms of automation, everyone on average would be spending less money on this particular thing that they were using human labour for rather than automation. So everyone has more money, then they’re going to spend more, and every other industry is going to get more funding, essentially. And in turn, those people are eventually going to find jobs. There’s going to be a gap where [the employment rate] dips, but I think eventually it’s going to go back up and higher. It’s going to be for more valuable things, anyway. For me, what I’m working on at Skinopathy is billing. To put it into context, usually now how doctors bill is that they write down everything they did for the day, they record all their appointments, whatever, and they send it off to the biller. The biller figures it out and uploads it to their cloud and everything. It’s just kind of ridiculous that you’re paying someone just to upload documents, and it’s not your receptionist, it’s just a separate person that handles billing. Part of the reason is because the Ontario health web system is so bad, in my opinion. I’m a developer working with this, and there are errors in the documentation, they use outdated technology from twenty, thirty years ago—they need to step up their game. And I think that would solve a lot of problems. To put it into perspective, these billers are making about two percent of what the doctors are making. If the biller bills for ten doctors, that’s twenty percent of how much a doctor makes. You can imagine that’s a really good salary, because doctors can make up to half a million a year. So just by billing for ten doctors you can make roughly six figures. I find that a little bit ridiculous. Doctors are losing a lot of money to things like this too. Sometimes your bills get rejected and doctors have to handle that. If there’s a better system there, it would be very beneficial to doctors so they can focus on patients more than paperwork. That’s one of the things that excites me about this project. I think uploading documents as a job sounds kind of ridiculous, and they have better places in society.

Host:Alright. I think that’s a good place to end our interview. Thank you so much for joining us today, and thank you for helping us unlock STEM!

(Ending Music)