Ethical AI and Mentorship in Tech with Alison Cossette – Techstrong Unplugged EP29

Hey everybody and welcome back to Techstrong Unplugged. I'm your host, Daan Solomon, and this is episode 29 of our series. In this episode, we're taking you to AI Deb Summit 2024.

There our co-host Cassandra Chen met up with Allison Cassette and they discussed ethical AI and mentorship in tech. Allison is a developer advocate at Neo four J and she shares her unique path from healthcare to data science. Without further ado, let's head over to AI Dev Summit 2024.

Welcome back to Text on Unplugged. My name is Cassandra Chin. Today we're here with Allison Cosette.

Yes. Can you introduce yourself? Sure.

I'm Allison Cosette. I'm a developer advocate from NEO four J. And should we mention where we are today?

Yeah, yeah. So we're here at the AI Dev Summit and we're talking a lot about Gen AI and all the latest in technology for developers. Do you wanna talk about like how you actually got started into tech?

Sure. So actually I came from healthcare before I was in, I came into data science. And so I knew that as a healthcare practitioner I could help one person at a time, but I'd always loved math and analytics.

And so for me, I was able to say, but if I go into data science and I work in healthcare, then I can help many people at one time. So that was my transition really was directly from the healthcare side to healthcare analytics and it's been a love affair of data science and AI ever since. And you're still doing data science?

I am, I am. So my specialty at NEO four J is around using graph algorithms and graph data science on network structures to understand the data science of relationships. And like you said, used to like mentor.

I did and I do. So for many years I taught for boot camps and at universities, um, everything from basic Python and math and all the way through more advanced deep learning courses. I taught for Northeastern and I taught for Metis.

I taught for University of North Florida, a number of different colleges. And it's actually one of my favorite things to do is to teach, because especially when it comes to data science, people come in and it seems that this the big magical technical thing. And then as they progress, there are these moments where they realize, oh, I can do this, I can do data science, I can work in ai.

And those moments to me are always my favorite moments where someone sees themselves as a technologist for the first time. So that's always like been one of my favorite things. And even now, you know, we'll be at conferences and there'll be students that come up and like, here's my LinkedIn, reach out to me anytime.

Because a lot of times when you're getting started in technology, even if you've taken the courses, you don't necessarily understand the industry. And so having people that you can reach out to and just ask some basic questions of, of how do I put my resume together? Or what do you think is a good company for me?

Or how do I figure out what's a good company for me? And just having someone that you can talk to that has that experience, I think is really, really helpful. So I definitely encourage folks who are either early in their career or you know, are in school to really make those connections early because this is a thing what I found, people love to talk about what they do and they love to share their story.

So I think there are more people that are open in tech to being mentors, not an, even if it's not anything formal, but just someone that you could kind of touch base with and ask a question of. I just think it's such an important part of that growing process. And you know, we've learned a lot AI's changed significantly in the last 10 years, and so the more we can bring in that next generation and support them, but it's their future.

So how do we, how do we help them make it what they want it to look like? I think it's really nice how like you'll connect to students and help them out. Yeah, we had a couple of students here today, um, that are in for the summer from the east coast.

And so I was like, here, like, here's some questions, here's some things you could do, reach out. Um, we were at a meetup last night and there were actually two high school chil children, high school men from Germany and they were in San Francisco vacationing with their parents and they hit one of our meetups last night. So that was really fun too.

So you never know where those folks are gonna come up. So I just encourage people in the industry too to just reach out and just be present for others. Yeah, it's all really exciting.

Yeah, for sure. Um, do you wanna talk about like more about your passion for a data science? Yeah, so one of the things that I really love about data science for the longest time, I, and I think it might still be my Twitter handle, was stories are, uh, numbers are the best storytellers.

Um, and it was kind of a funny number story. My husband and I used to do triathlons together and back in the days when you had to actually connect to your watch to download the data, it tells you how long ago we were doing triathlons. And there was one day he comes home and he says, and I, I downloaded his data and I was like, really?

And he's like, what? I'm like, come on. And he is like, what?

I said, will you please shift on your bike on the hills? And he's like, how do you know I wasn't shifting on the hills? I was like, listen, the altitude is increasing.

Your heart rate's going through the roof. Your RPMs are going down, like, and the power is off the charts just shift on the hills. Um, and so just the way that numbers tell stories has always been really interesting to me because I think one of the things we find is our perception of what's happening around us isn't always in line with that reality.

And so data science was a way for me to help people ground themselves in what is the truth of the numbers, right? So that's where a lot of that comes from. But in AI specifically, I've got a really strong passion for what does it mean as a society, right?

We live in this algorithmically curated world. Everything from our newsfeed to what jobs come up when I do a search, everything has been curated by an algorithm that someone like me or o others have put into place. And so what is it that we as an industry need to take accountability for when we're building those algorithms and how we can, how we can do that as responsibly as possible.

You know, the intention of AI is to support humanity and make it better. And so we just need to always keep that in mind whenever we're working on projects. You were saying that data science is about numbers, but when I think of numbers, I think of math.

Yes. Is there like a relationship where they're completely separate? Oh, it's all math.

It's all math. I mean, all AI is math really. So when we talk about data science, we're talking about building models and we're building predictive models.

Um, but it's also just what we call exploratory data analysis, which is understanding what is the, I call it the topography of the data. So if we think of a data set like housing prices, right? What is the spread of how many bedrooms are in the community?

Oh, they're mostly three to four, but there's a couple that there's a, there's a tail, there's like a statistical tale of a 12 bedroom house or something like that. So it's really starts with some basic statistics. And then as we move into model building, you're going to get into more of the calculus in the linear algebra and maybe get into deep learning.

You're certainly going to get into those areas. Um, so it's all math all the time and when we're working with really big numbers like, and like it becomes a bit hard to do hand calculations. Yes.

Is there like a code element to this? Yes. So all of the work that I do and that I've done in data science has all been done in Python.

Some people will use r but there are libraries that you can use in these different technologies that will help you run those summary statistics and help you understand what's in there. So, um, Python and Pandas is really the best place to start when you're trying to analyze sets personal recommendation. And when we're talking about data sets, like are there any like data sets you've handled, like specific types?

Yeah, so I mean, I've been around for a minute, so I've had lots of different kinds of data sets, but what I encourage folks to do when they're getting started is to go to Cagle because Cagle has all different kinds of data sets available and it's a really good place to kind of see what does data science look like, what are some of the things people are working on. And a lot of them, there'll be an opportunity for you to look at the notebooks that other people have used. So it's just a great community learning source to go to Kegel.

And Kegel is actually a competition site technically, um, where you can participate in different challenges and sometimes there's monetary value. Um, but it's just a great place to be able to get started and play around and just be part of the community. And I think you're also giving a talk later today.

I am giving a talk today. So my talk today is called Beyond Vectors. So, um, when we talk about generative AI and the GPTs of the world, one of the things that we know is that the information that they create isn't always exactly accurate.

It's something called hallucinations. And so what I'm talking about today is how do we actually ground that data in like your own database and how can we make sure that what's in that database is really robust and what can we learn from that? So life beyond the Vector, is this like intermediate beginner?

I would say it's pretty intermediate. You have to have a good understanding of what retrieval augmented generation is. Yeah.

Um, but if you do, I think people are gonna find that, uh, they'll definitely pick up a new trick or two. Have you worked with retrieval augmented generation a lot before? Um, mostly over the last year as have most, uh, but yeah, like quite a bit.

So a lot of the work that we do in developer relations is creating different demos and bringing things out to folks. So I've spent a lot of this past year working in that area. And what I really want to bring to it is what is, how can data science actually support that?

So a lot of what I'm talking about today is understanding the data science of what's in that data set that you're using to ground your outputs and what is it that we can know so we can get the best possible output. Are there any demos you've created, which you've found really interesting there? I do have some really fun demos that I like.

Um, my coworker Jason Ku and I have done a couple of Star Wars demos that I really enjoy. So we've done some, um, about like how we can use centrality algorithms. We basically built ways for the Star Wars universe.

So, um, being able to, if I wanna go from Alderon to tattooing, what's the most efficient way to get there? Uh, leveraging Hyperdrive. So we've done that.

Um, and we've got, we actually built something where you, we, you are able to determine if I want to take down the empire using centrality algorithms, which are the best planets to focus on, um, depending on what your intentions are. So those are some of the fun ones. Also wanna highlight like you're starting your own podcast program?

Yes, I am. Which I'm super excited about. We're just starting it here at this conference.

It's the AI community cast that we're doing at Neo four J. So similar to what the conversations we're having now I sit down with folks who are speakers and other thought leaders in the industry and just find out what's on their mind and what they wanna share with the community today. And Nick, you'll be a really great podcast host.

Thank you. I appreciate that. We have also, um, like an audio podcast from you four J called Graph Stuff fm.

So I've hosted that audio, but this is gonna be our first video interviews, but I'm excited about it. And you get a little bit of practice now, practice today. Yes, we'll definitely get some practice.

I spent a lot of time public speaking and in the booth, so I think it'll be all right. Um, are there any other, like topics you wanna talk about? You know, the big thing for me always is just around this idea of ethical ai.

Um, one of the things that causes me personally some concern is the, the way the training data that's been put into the large language models and the copyright question. So that's something that I think we as a, as an industry really need to take a look at. Because I think for me, I believe in something called a pattern, right?

Which means at the moment of creation of any digital asset, there is an inherent right to determine whether it's used in a model and how that pattern extracted from it. And so I just think as an industry we need to take a harder look at the data that we're using, where that's been sourced from, and have we really gotten appropriate opt-in, of extracting that pattern? So think of, um, you know, coded bias from Joy bini with facial recognition issues or, um, you know, the different copyright issues from the Writer's Guild and things like that that have been going on.

So that's really a big concern for me is what are the digital assets that we are creating, even if it's just the behavior on your phone or on an app, and how are, how is that being used in AI and what obligations do we as an industry have to society at large to be mindful about those things? I think that's definitely a tricky one. Yeah, I've seen like age limits on AI because of the ethical issues.

They don't, don't want children to be at risk. Yeah. So one of the things I'm looking forward to is I'm looking forward to like, one of my dream things is that there's like a K to eight or a K to 12 foundation model and something that's really just built on like language that is the safest for that community.

Um, it's not an easy problem to take on and I certainly wouldn't necessarily be able to put in the computational time to do that, but I think that it's something that's really important and I think, you know, somebody's gonna do it sooner than later. When you describe the problem, you mean the ethical AI issue? It's somewhat the ethical AI issue.

And I think one of the things that's happened when we look at foundation models is that because of the requirements of how much data you need to build it, the net was cast very wide. And so, um, you know, there's just a lot of things that may have fallen into that that aren't the easiest. And so you have all of these, um, sort of like requirements and guardrails that people put on after the fact.

But my question is are there things that we can do, can we put those guardrails in at the training data level so that then we don't have to worry about it coming out on the other side? 'cause if it's not in the training data, then the mathematics of those word combinations isn't gonna come out on the other side. So is there something we can do like in the training data that's building these models that can maybe mitigate some of this?

So again, I'm not the AI researcher. I'm sure somebody is working on it, but those are the things that sort of I ponder. Yeah, I can definitely see it.

Like we wanna solve it at the root cause. Y'all just put like reels at the end. Exactly.

Exactly. So those are some of the things that I contemplate. It's all really interesting.

Yeah. So thank you for talking with me today, Allison. Thank you so much for having me.

It was really a pleasure.