Q&A: Is AI making it harder to learn? What a former executive wants every student to know

Q&A: Is AI making it harder to learn? What a former executive wants every student to know Swati Mestri Scientific Editor Andrew Zinin Chief Editor When Miriam Friedel started working at Capital One in 2020, generative AI was still largely a research curiosity. By the time she left the company as vice president of machine learning engineering in 2026, AI tools were writing code, building applications and reshaping how people work. We spoke to Friedel, who recently joined the University of Virginia School of Data Science faculty, about the realities of navigating school and work in the age of AI. In this interview, Friedel lays out the AI skills employers are looking for, explains how data science has changed and describes how this technology can be used to augment, but not replace, learning. In the midst of the AI boom, is learning data science still relevant? The hardest part of data science is understanding the right problem to solve. If you have a massive data set, AI can help you with some of the data exploration and modeling, but it can't tell you if you are solving the right problem in the first place. You need to understand why we are doing what we're doing. What are we trying to accomplish? How is this going to impact actual humans? Data science also requires a ton of business context, which isn't often available in a form that can be effectively used by AI. What we teach at the School of Data Science is foundational to understanding how AI works. Loss functions, linear regression, gradient descent—these are the building blocks on the way to understanding transformers, the architecture on which LLMs are built. Ensuring that our students really understand the technical underpinnings of this technology is going to make them more able to use it with both technical rigor and skepticism. Why is skepticism important as we integrate AI into our work and learning? Blindly relying on LLMs can be problematic because they're so sycophantic. They often just tell you what you want to hear, and that's actually really bad for making impactful decisions. You need another human who's going to push back and make sure you are solving the right problem for the right reasons. Generative AI systems produce output based on data they've seen before, and it could be complete garbage. The underlying model is saying, "This is my prediction. Here is what the user wants based on what they are asking and what I've seen." LLMs do this on a more complex scale. Not understanding the nature of this technology can lead people to blindly trust what is coming out. You need to have a healthy skepticism about what you are seeing. Part of that comes from your own knowledge and experience, and part of that comes from a more foundational understanding of data and how these models and systems work. We are working hard to instill this understanding in our students. How can students use AI in a way that doesn't negatively affect their learning? One thing that has been a big shift for me coming from industry is how to use AI when you have years (or even decades) of expertise versus when you're learning something for the first time. Many companies have associates using AI on things they are already experts on, so they have a good sense of how to gauge the output, and it can be a real accelerator. For students learning foundational concepts for the first time, they haven't yet built up the knowledge of how to best leverage AI. If you're a student using AI, it should be for things that you can verify the answer to. If students are just putting things into Claude or ChatGPT and assuming the answer it gives is truthful without having any way to validate it, that's not a good use of AI and it can short-circuit your own learning process. Being able to think through hard problems and get frustrated and push through—that's how you learn, and that's how you innovate. New things are created because some things are hard. Use AI to refresh things in your mind that you know. If you've never built a linear regression in Python or R and you're trying to learn how it works and what the outputs are, AI can be incredibly helpful in getting you started and getting you unstuck, but you should do the hard thinking. Use AI like a TA, like a sparring partner, but you have to actually do the work. If it turns into, "I'm just going to have Claude tell me what to do," it's replacing your learning. How is the AI boom changing the way data scientists work? Claude Code really raised the bar in terms of the quality of the code that an LLM can generate. It can also help with things that can feel like drudgery or are not central to your learning process. Lately, I've been using it to help generate plots to share with my students. I can never remember plotting syntax, and AI helps free up my time to think through more foundational concepts I want to teach so I'm not getting bogged down in making a plot look exactly how I want it to. At Capital One, we liked to try to automate everything but the creative thinking and problem-solving. Use these tools to free you up from the tasks that are repetitive and don't require a lot of human input. This gives us time to think about more complex and interesting problems. What do employers mean when they say they want candidates with AI skills? I usually like to answer this question by telling students what employers are not looking for. Typing a prompt into ChatGPT, waiting a few minutes and cutting and pasting the answer without review is not a skill. Employers are looking for people who have the discernment to know both when an AI is an accelerator for pushing innovation and when it is getting things wrong. When I left Capital One, if you were committing code that was being written by an agent, you were saying, "I, the human, am responsible for what this code is doing." When there is risk involved, they are looking for people who are going to have good judgment and can own that risk. This requires knowledge of coding best practices and systems design, and the ability to identify when things can fail. Having a deep understanding of technical fundamentals is going to make you much better at using AI effectively. If you're a data scientist, employers want to know how you can effectively leverage AI to build pipelines or speed up the model development process. This might look like an agent that builds a model with certain knowledge or guardrails you have put in place. If you are a researcher, employers are looking for someone who deeply understands the architecture of LLMs and how you might be able to contribute to the frontier. If you're an engineer, employers are asking if you can build systems that leverage an LLM as one part of a broader ecosystem. They want to know that you can use AI tooling to both write code and build agents, and that you can properly test the generated code and be sure it's doing what you think it's doing. Across the board, if you can build an app that uses AI or write a skill for repetitive tasks, you will be at an advantage. What was your biggest challenge working in industry? The hardest problems that I had to deal with as an executive were not technological challenges; they were figuring out the right problem to solve, executing on it and working with other humans. Technical skills matter, but effective communication and a willingness to solve problems are paramount. That meant dealing with disappointment (both mine and others'), giving and receiving difficult feedback and managing conflicting viewpoints. Sometimes you need to make decisions for your team that might upset someone. If you are interested in leadership, it is important to ask yourself, "Can I work hard to solve hard problems, and can I deal with other humans in a way that is fair and gracious?" Provided by University of Virginia