How do we restore trust in technology in the age of AI?
As AI systems become more powerful and more pervasive, the question is no longer only what these technologies can do. It is also whether we can trust the knowledge that underpins them. That’s where ITU Associate Professor Aske Mottelson comes into the picture.
Aske MottelsonResearchartificial intelligence
Written 17 June, 2026 10:27 by Theis Duelund Jensen
For Aske Mottelson, one of the most urgent challenges in AI is strengthening the scientific foundations for how humans are studied, represented, and evaluated in relation to technology. This work lies at the center of Human-Computer Interaction (HCI), the field concerned with how people use, understand, and are affected by digital systems.
While HCI is often associated with design, creativity, and critical exploration of computational artifacts, Aske Mottelson’s work highlights another, critical dimension: the reliability of the scientific knowledge produced in computing research.
“Technology should not only be efficient or functional,” he says. “It should work for people. But to ensure that, we need to study real people interacting with technology, in a way that is scientifically sound.”
The reproducibility challenge
The researcher’s recent work address what is often referred to as the “reproducibility crisis” — a concern across sciences that many published findings cannot be trusted, because they cannot reliably be reproduced.
In a recent paper published at the flagship conference in the field, CHI, he and collaborators revisited a decade of HCI research, systematically attempting to reproduce statistical analyses using the original data and code provided.
The results reveal a structural problem: while some studies share data and methods, only a small fraction do so in a way that enables others to reproduce their findings. And only in half of the cases, the researchers were able to reproduce the same results, as the original papers.
“If you share your data and your analysis,” he says, “then others should be able to reach the same conclusions. If they can’t, then something doesn’t add up.”
This is not necessarily an accusation of error or misconduct. Rather, it points to inconsistencies in how research is documented and conducted — from unclear analytical procedures to selective reporting.
Only a minority of HCI studies, or studies of humans in computer science in general, he notes, currently meet the standards required for reproducibility. And even fewer adopt open science practices that are standard in fields like cancer research, psychology, and economics, such as pre-registration, where researchers publicly specify their methods and hypotheses before collecting data.
Validating knowledge in the age of AI
At a time when AI systems are developed and deployed at unprecedented speed, these issues take on new urgency.
Much of the research within artificial intelligence and related fields is validated primarily through technical benchmarks — measures of performance that do not acknowledge human variability, or how human users experience using such systems.
“There is a strong focus on what the technology can do,” Aske Mottelson explains. “But much less on how it actually impacts people.”
Empirical studies involving human participants remain rare in AI research. And when they do occur, they often rely on small sample sizes or methods and analytical approaches that would be considered insufficient in the behavioral sciences.
This creates a gap between technological capability and human understanding — a gap that HCI is uniquely positioned to address, but only if its methods are robust.
“If we want to build systems that affect people’s lives,” he says, “then we need evidence we can trust.”
Setting new standards
Mottelson’s work is both diagnostic and prescriptive. In addition to identifying methodological limitations, it proposes concrete ways of improving research practices — from clearer documentation to structured data sharing and reproducibility checklists.
Not surprisingly, such efforts can provoke debate.
“There are people who don’t like being told how to do their research,” he notes. “Especially in a field as diverse as HCI.”
But for Mottelson, the goal is not to enforce uniformity. It is to establish a minimum standard of transparency and reliability — particularly for studies that claim empirical insight into human behaviour.
“There will always be, and there such be, different approaches to conducting science,” he says. “But if we present something as hard evidence for instance for policymaking, then it has to be possible to verify it.”
Theis Duelund Jensen, Press Officer, phone +45 2555 0447, email thej@itu.dk