Beyond the hype and toward real impact

The field of data science is evolving at an unprecedented pace, with artificial intelligence and machine learning reshaping industries, research, and daily life. But as we accelerate toward automation, are we losing sight of the fundamentals? To explore these concerns, we spoke with Martijn Willemsen, an associate professor at Eindhoven University of Technology and the principal investigator of the Recommender LAB at the Jheronimous Academy of Data Science in 's-Hertogenbosch.

The path into data science: a problem-driven approach

Martijn’s journey into data science did not begin with machine learning or neural networks. His background in Electrical Engineering at TU/e led him toward a deeper curiosity about how people make decisions and interact with technology. While working on his PhD, he developed empirical studies to understand decision-making, ultimately shaping his work in cognitive processes, statistics, and user experience:

"I just liked doing experiments—better understanding how systems work for users."

When metrics take over: the limits of AI optimization

One of Martijn’s biggest concerns is that modern AI systems prioritize performance metrics over user satisfaction. Recommender systems and AI-driven decision models increasingly focus on optimizing one key metric—whether it be engagement, retention, or clicks—without understanding whether these numbers reflect meaningful experiences:

"It’s not about the metric itself. A single metric doesn’t tell you if people are actually satisfied."

This shift has also influenced how companies use data. Many organizations start with data and then look for insights within it, rather than starting with a problem and determining what data is needed to solve it. This is a mistake, according to Martijn, as true data science is problem-driven, and should start with a clear definition of that problem:

"What happens often in companies is that they have data and want to ‘look for something in it.’ But that’s not the point. We should start with the problem we want to answer and then see what data we need to address it."

At JADS, the master's program in Data Science in Business and Entrepreneurship emphasizes this applied, problem-solving mindset. Unlike other TU/e master’s programs that focus on developing new algorithms, this program trains students to apply existing techniques effectively in real-world scenarios. Initially, Martijn worried that students might not be learning enough technical depth, but feedback from industry partners proved otherwise:

"Companies tell us that our students are real problem solvers, and that’s what they need most."

The importance of industry collaboration in data science

A key aspect of impactful data science research is collaboration with industry. While academia provides theoretical depth, companies provide access to real-world datasets. This relationship is crucial, as demonstrated by Spotify’s API, which has enabled researchers like Martijn to conduct realistic experiments on user behavior and recommendation systems:

"In data science, you cannot do good research without collaboration with companies. They have the data, and to conduct realistic studies, we need real users."

Music, in particular, presents an excellent field for recommender system research. Here, users can determine their preferences within seconds. The availability of real-world data from platforms like Spotify has enabled more naturalistic studies than traditional lab-based experiments.

However, concerns are rising over the increasing restrictions on API access. While many companies once embraced open data access, the trend is shifting toward closed systems, making research more difficult.

"It’s true that companies make money from this, so they’ll keep APIs available to some extent. But we’re seeing a growing tendency to lock data behind corporate walls, and that’s worrying."

The ethics of AI and the risks of uncritical adoption

Beyond access to data, Martijn raises concerns about how AI research is conducted today. The rapid pace of AI model development, particularly with large language models (LLMs), has outpaced traditional scientific review processes. Whereas academic research goes through peer review and validation, AI companies are releasing powerful models without sufficient scrutiny:

"Companies are developing new algorithms at an extreme pace without fully thinking through their consequences. As scientists, we have a process—peer review, validation, critical discussion. But now, that process is being bypassed, and that worries me."

This rapid development also affects how AI is used in business. Many professionals are adopting AI tools without understanding their mechanisms, leading to potential misuse. AI-generated business plans and automated course creation, for example, often rely on repurposed internet content rather than original insights:

"People are using AI for things like business planning, but often it’s just a fancy way of searching Google. That doesn’t mean it’s producing truly innovative insights."

Rethinking AI’s role: beyond automation and toward understanding

As Martijn puts it, he is not a "hype person." While many celebrate the endless potential of AI, he is more cautious, emphasizing the need for a balanced approach:

"I hope we go back to simpler models. The hype is making us run fast in certain directions without really questioning whether these things work as we want them to."

During a discussion at TU/e, Martijn was surprised by how students engaged with large language models (LLMs). Expecting them to be more enthusiastic about the technology, he found that because they understood its limitations, they were more skeptical:

"I thought students would be fully embracing LLMs, but the more they knew about them, the more cautious they became. They were aware of the risks and limitations, which is different from how people outside the field perceive these tools."

This highlights a key issue: those who understand the inner workings of AI tend to be more hesitant in its adoption, while those with little knowledge often embrace it uncritically. This gap in understanding poses challenges in how AI is integrated into business and daily decision-making.

LLMs are often glorified for their reasoning abilities, but Martijn remains skeptical about this topic:

"It’s not surprising that they outperform humans in some tasks, but the energy cost is enormous. For simple queries, a Google search is more efficient, and you actually learn more along the way."

His biggest fear is that we are losing critical expertise by relying too much on automation. In some cases, deep learning models have been shown to perform worse than traditional, simpler models. Yet, the industry continues pushing AI-driven solutions, often at great computational cost and with little real benefit.

"I worry that we are no longer learning. We need experts who truly understand these models, not just people who know how to use them."

Final thoughts: a call for reflection in data science

Martijn’s perspective serves as a reminder that data science is not just about automation—it’s about understanding. Whether in academia, industry, or AI research, the focus should not be on blindly adopting the latest models but on critically evaluating their impact and effectiveness. The next generation of data scientists must go beyond technical skills to become problem solvers, ethical thinkers, and responsible innovators.

As AI continues to reshape our world, Martijn urges us to pause and reflect:

"We need to do more than just say ‘we’ll see.’ The internet brought many good things, but also some bad. With AI, the process is happening even faster. People are adopting it without thinking, and we need to ensure that we don’t lose sight of what really matters."