Reducing Barriers to Complex Data Science Entry by Leveraging AI

Author: Beverly Wright

In this podcast, Dr. Beverly Wright, Vice President – Data Science & AI at Wavicle Data Solutions, engages in an insightful conversation with Hua Ai, General Manager Predictive Modeling at Delta Air Lines, focusing on the topic “Reducing Barriers to Complex Data Science Entry by Leveraging AI.” They discuss the critical questions merging data science and artificial intelligence, such as the types of data science tasks that can be automated and effectively managed with AI, the specific steps that can be streamlined through AI, the transformation of necessary skill sets, the evolving landscape of data science jobs in the context of AI integration, and valuable advice for using AI to overcome barriers in data science. 



  • Dr. Beverly Wright, Vice President – Data Science & AI at Wavicle Data Solutions  
  • Hua Ai, General Manager – Predictive Modeling at Delta Air Lines 


Watch the full podcast here or keep scrolling to read a transcript of the discussion between Beverly and Hua:



Beverly: Hello, and welcome to TAG Data Talk with Beverly Wright. Today I’m with Hua Ai, from Delta Air Lines. She is general manager of predictive modeling, and we’re talking about reducing barriers to complex data science entry by leveraging AI. 

Thanks for being here.  


Hua: Thanks for having me. 


Beverly: Absolutely. Let’s start off with a little background. I know that everybody at Delta thinks you’re super cool. So, tell us, why are you so cool? 


Hua:  Well, look at my last name. Exactly. It’s exactly how I got into grad school. 


Beverly: Yeah, with the last name like AI. I like that. Maybe I should change my name to Beverly Data Science. 


Hua: It’s been a fantastic journey with Delta. I’ve been with the company for six years. I lead a data science team and I’m embedded in the operations side of the house. So, if you see that the airline is like the commercial side, where they figure out pricing and revenue management, there’s a very complex revenue management system for the airlines. And on the operations side, we run the flight schedule, we make sure that every flight will get out on time, with bags, with great customer service. So, my team look into: if things are delayed, what do we do? Given the weather, do we expect to delay? And how can we be proactive and serve our customers to get them where they need to be? 


Beverly: Well, I’m a big fan. I really liked Delta, and I like to fly Delta. So, what did you do before Delta? 


Hua: I did consulting work. I was curious to know where the best place is to work in metro Atlanta, and consulting is a quick way to find that out. So, I had fun with a lot of great clients in the Atlanta area and different account teams and all that. And then I just found a place to settle. 


Beverly: Nice. I like that. Yeah, there are a lot of great employers. I mean, it did my rounds too. So, this reducing barriers to complex data science by leveraging AI, this is like an “AI for all” vision. This the dream, you know? AI is going to help with all these sorts of things. So, let’s start off with: when we talk about AI helping reduce barriers to data science, are there specific types of data science work where AI makes more sense? What kinds of data science work can leverage AI more effectively? 


Hua: Right. First, I want to make a comment that AI is a very generalized term, it means different things to different people. And nowadays, especially generative AI. I’ve had people talk to me like, “hey, Hua, we have just generated AI models that can help you to forecast,” and when we talk more, I was like, okay, so what you really mean is a forecasting model trained by big data, and had some pre-trained phase where you can transform whatever model that you learned to a new problem. It is not an NLP type of generative AI or a vision recognition type of cognitive AI system that people talk about. So, I find that to be pretty interesting. Now it’s just AI. AI is the general term of anything machine learning, analytics, or whatever that you throw into the data science fields that now it’s under AI. So, that’s pretty interesting.  


But I think to your point, your specific talk about the type of systems that help the data scientists to do their work faster. If you think about data science workflow, if someone calls you and says, “hey, Beverly, could you help me think through analytic solutions to this problem?” You probably need to understand the business processes first: what exactly they’re trying to solve? What metric are they trying to optimize? And, then you start from there, you collect data, you understand the processes. Once you actually have the data in one place, then you’ll sit down and do your like data impute, cleansing data, engineer your features model, modeling process all that. Once you get some numbers, you’ll get back to your client to review the model performance and see how do you interpret those and see how they really like it, is it good enough or not. You go back to your desk and to your computer and do more.  


So, there’s still a lot going into the data science practice, and it means different scope if you’re a different company, I think there are definitely a lot of steps that can get accelerated, amplified, and scaled by technology. And it’s not a scary thing at all, we were already doing it. Who claims they don’t have Stack Overflow open while they are working? 


You’re already actively using technology, using knowledge out there on the internet to help you to build better models to do things better. What if it’s more convenient? What if you have a copilot? There are a lot of entry level tasks that can be largely automated. For example, data cleansing, feature engineering, model tuning, model selection. Those are tools that can help data scientists to do that more effectively. 


Beverly: I see. Okay, so the process that you just described sounds a lot like the scientific method applied to model building. Once you’ve identified a problem, you go through the steps, all the way to solution development and communicating to the client. What about the other sort of window dressing, kind of the outskirts of the model? Because data science, AI really, is more than just model building. There’s all the other stuff. Can you speak to that a little bit? Can AI help with some of those things like understanding the business context and framing the problem?


Hua: Yeah. I’ve actually heard about the use case, the other day that when a company that does image recognition first introduced a technology to a hospital, they first thought about if they can use the new technology to help to read X-rays and to add into the diagnosis. But when they work more with an AI solution company, they actually found out that there’s the piece that’s very available to be used to use the technology to collect patient history right there. And they totally transform the use cases from a much higher entry point that some people may not feel comfortable to know that right now you can use machines to read X-rays, even if we can assure them that it will be looked at by an expert, but people might have different opinions about whether you should do that or not. But after knowing what AI solutions can do, they actually find a better way of using their AI technology. I think that’s going to happen in a lot of data science fields.


First, you’ll probably only be confident to use the technology to deal with a lot of the technical aspects of it. But, once people understand a bit more of what automation can do for them, what AI solutions can do for them, there is a lot of opportunity. For example, to understand the background of the project, to understand the history, there might be a lot of data that’s already been collected, but under a different project or under a different context. How do you tie these different pieces together? And especially when we use machine learning more in an enterprise concept, you’re not just talking about one model, you’re talking about a forest of models.


So, how do you keep up with the different models that different teams are building? Are you putting another programming organization architecture on top of it? Do you leverage AI to know that this model has been developed by this team, and all this has been stored in a knowledge base that now by talking to the AI system, you can easily know: what are the models they’re building? What’s the background? What are use cases? Whether that’s related to your use cases have not, so you can be more targeted to connect with other teams that are doing similar things.


Beverly: Do you think there is a stronger appetite for certain types of tasks for AI to enable? I’m not really as much “hands-on keys” anymore, except for some of my nonprofit stuff, but I can tell you right now that if I could get rid of data engineering, I would. We all want to build models. We don’t want to spend all the time doing janitorial work, so there’s probably an appetite there. But are there certain pieces that you feel the data science and AI community is going to say, “no, don’t take that from the humans, that needs to stay with us” and other things where they’ll say, “oh, please take it.”


Hua: I think the amazing part of this is that experience can be very personalized. So, think about having an assistant, and it will take whatever part that is not exciting for you. There are engineers out there that just want to make the most solid and robust things they can build. They think actually that the modeling process is pretty standard. Because you’re not inventing a new model, you are applying someone else’s invention, like Google came up with the transformer technology, but it’s really the application that has been standing out and taking the company to success.


So, some people will focus on scaling up machine learning solutions and various standard mature solutions. I think it is more of a personalized experience; if you love to understand basic processes and don’t care so much about coming up with a quick model overnight, you can have your AI technology help you to do that. If you are an engineer, you pick the fun part you want. If you’re a scientist, you can pick your part. We have a lot of diversity and inclusion discussions nowadays, and this technology is so profound that it’s going to change our relationship with knowledge and our tools.


Back in the day, you possessed some knowledge; you went to school for years to possess that piece of knowledge. But nowadays, it is ad hoc, like you borrow a piece of knowledge because you are in this domain and use case today. And if that’s something you’re interested in, you can become a so-called expert very quickly using AI assistance and do something interesting that’s beneficial for your solutions. But then you can jump out of it and do something else. This technology is going to re-level-set a lot of the entry requirements for jobs. 


Beverly: That leads me into the next piece, which was how this has changed the skills we’re going to need. But let me back up for a second because you said something profound: AI has the potential to change our relationship with knowledge itself. And if I hear you correctly, you’re saying instead of being required to hold this knowledge in ourselves – to: learn, and absorb, and keep it in so I’m ready and have knowledge – it will be less about that and more about knowing when to borrow it. So, what sort of skills are different in that completely different paradigm?


Hua: Yeah, it will make some people uncomfortable. Because you have an identity in the professional world, where what you know is your identity. But it might be transitioned to what you care about and what you deliver, it is no longer about what you know or are trained for. So, it is going to be a change in our office. But that’s upcoming, and that’s not up to us. It’s clear we need to adapt and change ourselves. 


Beverly: So, it’s the knowledge of knowing when you’re going to need different pieces of expertise that is probably going to trump having the actual knowledge because it’s impossible to have all the actual knowledge that an AI system would have. But understanding when we need what we need and how to apply it is going to be more important. We’re not really writing ourselves out of a job; we’re changing the way we operate. Is that right?


Hua: That’s how I think about it. I get a lot of questions from my friends because I’m working as an AI fellow in this case. “Hua, do you think that AI is going to take over?”  It needs to be a complete sentence: “AI is going to take over what?” It’s going to take over certain aspects or really take care of certain aspects, but I don’t think it’s going to take over our lives.


Think about it. Now, everyone can be a CEO, and you can have a very effective assistant that helps you to achieve your goals faster and more efficiently. So, how are you going to interact with your assistant? You’ll probably not ask your junior assistant to write a statement of the company mission or strategy for the next three years. Especially nowadays, or maybe in the next five years, the way we interact with our AI systems will be breaking things into tasks.


I heard one of the CS talks this year. And the discussion was mostly about jobs being broken down into tasks. AI might be taking over certain tasks, but the intention behind the tasks and the sequence of the tasks need much more understanding of why and how we’re doing it. Those are the strategic things that the boss is driving, but there are more of us who will be our own boss; we will figure out what we want to achieve and effectively use these new tools to achieve it.


Beverly: Wow, very interesting. To wrap up, what final piece of advice would you give our listeners about this movement, where AI is reducing barriers to entry and in data science? 


Hua: What you said already, embrace it. It’s here. When someone invented a calculator, you no longer needed to do those calculations anymore. When someone has such a great tool for you to free up and think more about what you really care about and what you want to do, bring yourself up one level. Think about what matters and makes you satisfied with your job, and care more about specific tasks; there are lots of tools, platforms, or AI that can help you do that. So, embrace the change. 


Beverly: Embrace it and really get into it. Yeah, I love it. Thank you so much to Hua, General Manager of Predictive Modeling at Delta Airlines.  


Hua: Thank you so much. It’s been fun. 


In a data-driven world where AI is reshaping the landscape of analytics and insights, Hua and Beverly’s conversation about reducing barriers to complex data science entry by leveraging AI sheds light on the transformative power of AI in data science. Explore the full catalogue of TAG Data Talk conversations here: TAG Data Talk with Dr. Beverly Wright – TAG Online.