The economic complexity of the Roman Empire as inferred from Latin inscriptions by Matteo Mazzamurro, Petra Heřmánková, Michele Coscia, and Tom Brughmans, published in npj Complexity.

Economic complexity is a powerful tool to estimate the productive capabilities and future growth of modern economies. Little is known about how economic complexity evolves over long periods in history. In this paper, we use archeological evidence from the Roman Empire in the form of short texts preserved on a durable material (i.e., inscriptions) to estimate the economic complexity of the various provinces of the empire. By connecting the occupations listed in the text of inscriptions with the location in which the inscribed objects were found, we can estimate that the most complex areas during the first four centuries of the Roman Empire have a remarkable and statistically significant overlap with the most complex countries today. This is a surprising result running counter the expectations from historiography, and we lack an explanation for the reason of this supposed preservation of economic complexity through the ages – both factors pointing at the need of future work to integrate Greek inscriptions. However, this evidence provides a suggestion about how difficult the development of economic capabilities might be.

Last month Giovanni Astante (who worked on the topic during his research stay at NERDS) presented our work (w. Roberta Sinatra) on “Investigating the Robustness of a Danish Voting Advice Algorithm” (VAA) at the European Conference on Algorithmic Fairness.

Now the paper is finally out in the Proceedings of Machine Learning Research!
You can find a copy here –> https://proceedings.mlr.press/v350/astante26a.html

We look at a popular VAA (called ‘Kandidattesten’), developed by the large media house Politiken, and investigated the robustness of its matching algorithm.

In past Danish elections 62% of eligible voters used some form of voting assistance algorithms (‘Valgtest’ in danish) one month prior to the election and 45% reported they followed the recommendation of a VAA when voting. This means these voting tools can have a large influence on election results.

Through a freedom of information request we got access to the algorithm used to match users and political candidates. The system functions by users filling out a questionnaire which is compared to a database of answers from candidates. Users are then presented candidates with highest agreement.

These systems have been studied in the past (mostly how humans interact with them), but so far nobody has tested their robustness. We study how minor modifications to the algorithm’s weights can change the outcome of which political candidates are recommended. In addition, we study how dropping questions, i.e. calculating match similarity based on fewer questions (than the original 20), affects which candidates are recommended.

Our results demonstrate that even minor tweaks to the algorithm can produce entirely different outcomes, potentially affecting outcomes of political elections. A marginally changed matching algorithm can recommend entirely different political candidates, from different political parties.

Based on our findings we provide recommendations to VAA developers and policy makers, on how to monitor and evaluate the robustness of these systems.

Measuring narrative polarization in online discourse by Jan Elfes, Marco Bastos, and Luca Maria Aiello, published in PNAS Nexus.

We introduce narrative polarization as diverging representations of key actors through their positioning in popular narratives. For example, a representation of Palestinians as a people striving for rights and of Israel as a violent actor favors interpretations of violent actions by Palestinians as resistance. In contrast, a focus on the struggle of Jews to escape historical persecution and of Palestinian groups as a constant violent threat might render the same actions as terrorism. Such structural differences in how actors are positioned shape the information landscapes that audiences navigate and affect their opinions. We showcase this by measuring narrative polarization around the Israeli-Palestinian conflict on YouTube. We find that, while videos produce highly polarized narratives, comments converge on a shared narrative distribution, offering less polarized narrative representations.

Group size effects and collective misalignment in LLM multi-agent systems by Ariel Flint, Luca Maria Aiello, Romualdo Pastor-Satorras, and Andrea Baronchelli, published in PNAS.

As AI agents begin to operate in populations rather than one at a time, our new research suggests that the number of them changes what they collectively decide — amplifying a bias, inventing one from nothing, or flipping a group into the opposite of what each agent would choose alone. When artificial intelligence (AI) agents interact in groups, their number is not merely a technical detail. It is a decisive factor in what the group settles on: populations built from the same AI model, doing the same task, can reach opposite outcomes for no other reason than that one group is bigger.

In this work, we experimented with populations of LLM agents playing the “naming game”, a classic framework for studying how conventions emerge, in which randomly paired agents each pick a word from a shared pool and are rewarded when they happen to pick the same one. Agents see only their own recent interactions, never the wider population, and are never told they are in a group. Over many pairings, a population can converge spontaneously on a shared convention.

We found that interactions can pull a group away from what its members individually want in three ways. It can amplify an existing leaning until the group converges on it almost every time. It can induce a preference out of nothing, with populations of individually neutral agents reliably favouring one word over an equally viable alternative. And it can reverse a preference outright, so that a population settles on the word its own members disfavoured. Group size then determines how strongly these preferences bite, in ways that cannot be extrapolated. Larger populations became more predictable across every model and word pair tested, converging on one word until the outcome was effectively certain. But the size at which that tipping point arrived varied enormously: for some combinations as few as two agents, for others around ten thousand.

Overall, our results demonstrate that more is different for LLM populations: The number of interacting agents is a key driver of the dynamics, with implications for the design and governance of multi-agent AI systems.

Unmasking conversational bias in AI multiagent systems by Erica Coppolillo, Giuseppe Manco, and Luca Maria Aiello, published in PLOS One.

New paper on PLOS One by Luca Aiello Detecting biases of generative AI is critical, but it is often done considering models in isolation. In particular, biases emerging from interactions among conversational agents remain largely unexplored. In this paper we present a framework designed to quantify biases within multi-agent systems of conversational agents. We simulate echo chambers where agents are initialized with aligned perspectives on a polarizing topic and asked to develop the topic in multi-turn discussions. Surprisingly, we observe that, despite the echo-chamber setting, the agent stance shifts away from their initial position, often towards liberal positions. Crucially, the bias observed in these echo-chamber experiments remains undetected by traditional bias detection methods that probe models in isolation. This highlights a critical need for the development of a more sophisticated toolkit for bias detection and mitigation for AI multi-agent systems.

Do you work with human mobility data?
Then this publication could be important for you to read, and to reflect on!

A tale of ten cities: data bias in human mobility is pervasive and highly location specific by K. den Nijs, E. Omodei, & V. Sekara published in EPJ Data Science.

In this joint work w. Katinka den Nijs and Elisa Omodei we study ‘data production’ biases in human mobility data. Data production bias is a form of bias which deals not only with whether individuals are represented in datasets, but also how they are represented. Especially in terms of how much data they produce. Put differently, we don’t only care about if people are represented in data, we also care about how they are represented.

We study a GPS mobility dataset, collected from anonymized smartphones for ten major US cities, and find that data points are as, or more, unequally distributed than wealth. Imagine!
Their data production is more unequally distributed than wealth in society!

To understand the interplay between demographic factors and data production bias we build machine learning models to predict the number of data points we can expect to be produced by the composition of demographic groups living in census tracts.  We build one model per city and, generally, find strong effects of wealth, ethnicity, and education on data production.

While we find that bias is a ubiquitous phenomenon, occurring in all ten cities, we also find that each city suffers from its own manifestation of it, and that location-specific models are required to model bias for each city. Our work raises serious questions about general approaches to debiasing human mobility data and highlights the urgent need for further research on the topic.

In summary, be careful with biases in human mobility data (or any other digitally collected datasets). Data production biases can have significant downstream effects, when data is used to model everything from epidemics, behavior, etc.

TLDR; Representativeness does not mean we should only strive for individuals to merely be ‘present’ in data. As our results show, how people are represented, or misrepresented, via the quantity and quality of data the produce is of equal importance. Further, ‘debiasing’ data is no easy task, as it heavily depends on geographical context.

Comparing the Italian and Danish Music Industries With Network Analysis by L.M. Benetti, D. Fejerskov-Quist, & M. Coscia, published in Complexity

We build music band collaboration networks based on bands sharing artists playing for them. We do so for both Italy and Denmark. The paper shows that the two networks have similar structural properties, with clusters based on genres and temporal similarity. There are also differences, showing that the Danish and Italian music industries have different periods and genres in which they express a high degree of diversity. Finally, we can identify the bands that link together the Italian and Danish music scene.

Right-wing Votes Relate to Delays in Bicycle Network Development, by C. Sebastiao & M. Szell, published in Findings

Paris has implemented only 43% of its 2021-2026 bicycle network plan by its 2026 deadline, but the delays are not equally distributed. Correlating political voting and bicycle infrastructure data, we find that boroughs with a higher share of right-wing votes are also boroughs with the largest delays in bicycle infrastructure development. Although the number of voting units is only 17, this correlation is remarkably strong and significant. At the city scale, for each additional 1% of right-wing votes, we find 0.86% less protected bicycle lanes built.

Human, Algorithm, or Both? Gender Bias in Human-Augmented Recruiting, by Mesut Kaya & Toine Bogers, published in FAccT ’26: The 2026 ACM Conference on Fairness, Accountability, and Transparency

Recent years have seen rapid growth in the market for HR technology and AI-driven HR solutions in particular. This popularity has also resulted in increased attention to the negative aspects of using AI to support hiring practices, such as the risk of reinforcing existing biases against vulnerable groups based on gender or other sensitive attributes. Combining human experience with AI efficiency in making recruiting and selection decisions has the potential to help mitigate these biases, but despite a considerable amount of research on fairness in algorithmic hiring, actual empirical evaluations comparing the fairness of human, AI, and human-augmented decision-making remain scarce. In this study, we address this gap by presenting a quantitative analysis of gender bias across three scenarios of a real-world recruitment platform: (1) recruiters searching a CV database manually for relevant candidates, (2) AI-driven matching between candidates and jobs, and (3) a combination of human and AI-driven recruiting. We find that human recruiters produce lists of candidates that are fairer in terms of gender than the AI-only solution, with more deliberation by humans resulting in fairer outcomes. However, the combination of human and AI-driven is more than the sum of its parts and produces the fairest candidate lists: interacting with the slate of recommended candidates first before manually searching for additional candidates has a beneficial effect on the gender fairness of the set of candidates that are viewed, clicked, and contacted afterwards. Our work provides one of the first empirical comparisons of fairness across human, AI, and hybrid recruiting processes, offering evidence to inform the development of more equitable hiring practices and highlighting the importance of human oversight for mitigating bias in algorithmic hiring.