Social Effects in Science: Modelling Agents for a Better Scientific Practice

Science is a fundamental human activity and we trust its results because it has several error-correcting mechanisms. Its is subject to experimental tests that are replicated by independent parts. Given the huge amount of information available, scientists have to rely on the reports of others. This makes it possible for social effects to influence the scientific community. Here, an Opinion Dynamics agent model is proposed to describe this situation. The influence of Nature through experiments is described as an external field that acts on the experimental agents. We will see that the retirement of old scientists can be fundamental in the acceptance of a new theory. We will also investigate the interplay between social influence and observations. This will allow us to gain insight in the problem of when social effects can have negligible effects in the conclusions of a scientific community and when we should worry about them.

💡 Research Summary

The paper investigates how social dynamics within the scientific community can influence the acceptance or rejection of scientific theories. To this end, the authors construct an agent‑based opinion‑dynamics model that distinguishes two types of agents: “experimental agents” who directly experience the outcomes of experiments and therefore receive a signal from an external field representing the true laws of nature, and “theoretical agents” who do not have direct experimental access and must rely on the opinions of their peers. The external field has a strength parameter h; when an experimental agent’s current belief aligns with the field, its confidence in that belief is reinforced, and when it misaligns, confidence is weakened. All agents are embedded in a social network (either a random graph or a small‑world topology) and interact with neighboring agents with probability p. During each interaction, agents update their belief according to a mixture of conformity (moving toward the majority opinion) and contrarianism (maintaining a minority stance). The relative weight of these two strategies is fixed for each agent.

A key novelty of the model is the inclusion of a “generation turnover” mechanism. Each agent carries an age variable; when the age exceeds a predefined retirement threshold, the agent leaves the system and is replaced by a newcomer with a randomly assigned initial belief. The newcomer inherits the same network connections, so the overall topology remains unchanged while the composition of the population evolves over time. The retirement rate λ thus controls how quickly old scientists are removed and new ones are introduced.

Through extensive simulations the authors explore three main dimensions of the parameter space: (1) the magnitude of the external field h, (2) the intensity of social interaction p, and (3) the retirement rate λ. When h is large (i.e., experiments are highly reliable and reproducible), experimental agents quickly converge on the correct theory, and their influence spreads through the network, leading the whole community to adopt the new theory regardless of λ. Conversely, when h is weak or experimental data are noisy, social influence dominates. In this regime, if conformity dominates, the community tends to lock into the prevailing (often outdated) theory, and the introduction of a new theory is suppressed. The retirement rate becomes decisive: a low λ (slow turnover) creates a strong “inertia” effect because long‑standing agents keep reinforcing the old paradigm, making it difficult for even accurate experimental evidence to overturn it. A high λ (rapid turnover) injects many agents with random initial beliefs, increasing opinion volatility and allowing the correct theory to percolate more readily through the network.

The authors also examine the balance between conformity and contrarianism. High conformity leads to rapid consensus—either on the old or the new theory—while high contrarianism maintains a fragmented opinion landscape where no single theory dominates. This observation mirrors real scientific disputes where a minority of dissenting voices can sustain alternative hypotheses for extended periods.

In the discussion, the authors link their findings to historical episodes such as the shift from Newtonian mechanics to Einstein’s relativity, noting that the transition coincided with the retirement of senior physicists and the influx of younger researchers open to the new framework. They also relate the model to contemporary concerns about reproducibility crises, suggesting that when experimental validation is weak, social reinforcement can amplify false positives and hinder correction.

Policy implications derived from the model include: (i) managing career cycles to avoid excessive “parochial inertia” by encouraging timely retirement or sabbatical rotations; (ii) strengthening experimental standards and reproducibility to increase the effective external field strength; and (iii) fostering network structures that prevent echo‑chambers, for example by promoting interdisciplinary collaborations that increase exposure to diverse viewpoints.

The paper concludes that scientific progress is not driven solely by empirical verification but is co‑shaped by social interaction patterns and generational turnover. The agent‑based framework provides a quantitative tool to explore under which conditions social effects become negligible and when they pose a genuine threat to the reliability of scientific conclusions. Future work is suggested to calibrate the model with real bibliometric data (citation networks, co‑authorship graphs) and to test policy interventions in silico before implementation in actual research institutions.