In this commentary, I suggest a complementary view to the target paper’s idea that primate social metacognition evolved as an adaptation to living in large groups. I present metacognition as a necessary step in the development of complex allostatic systems and suggest that intrinsic and social metacognition are dissociable, which can be studied in the mammalian default mode network.

Dunbar’s emphasis on dyadic relationships in group formation overlooks the roles of interdependence and joint commitment in social cohesion. We challenge his premise by highlighting the importance of group-level processes, particularly where top-down group pressures like cooperative breeding and out-group threat can induce joint commitment as an alternate means to sustain group cohesion.

The paper of Dunbar (2025) on social stress is a strong demonstration that stress in itself can have a purely cognitive origin. The paper shows that the cognitive system can have profound impacts on the hypothalamus. As detailed in my commentary, this opens up new avenues of how to interpret psychiatric conditions, placebo, and other associations between perceptions and vegetative functions in the brain.

Human social networks are far larger than those of nonhuman primates. Maintaining cohesion in large networks requires a robust mechanism that can accommodate the dense webs of connections within communities. A parsimonious account of how humans achieve social cohesion is mental abstraction, which enables individuals to construct fuzzy network representations that facilitate information flow tracking and mitigate conflict.

Primate species deploy a suite of behavioural and cognitive adaptations to offset the costs of group-living. Dunbar uses species-level comparisons to posit a series of cumulative steps that describe large-scale phylogenetic patterns in the evolution of sociality. Here, we highlight the value of population-level variation within species for empirically testing the predicted socio-ecological correlations that underpin Dunbar’s hypothesis.

Dunbar’s framework highlights the challenge of maintaining large, stable social networks given cognitive constraints. Expanding on this, I propose that fractal social networks function as lossy compression algorithms, efficiently reducing the complexity of social storage and retrieval. Rather than tracking all relationships explicitly, individuals rely on hierarchical abstractions and transitive inference, shifting storage complexity from $O\left( {{N^2}} \right)$ to $O\left( {N\log N} \right)$. This insight suggests broader implications for cognitive evolution, institutional organization, and artificial intelligence.

Challenges of group-living include foundational problems of cooperation and coordination that extend beyond anthropoid primates and may potentially be managed through evolved group-mindedness rather than expanded neocortical size and enhanced capacities for executive functions.

Ectotherms, particularly fish, challenge traditional brain evolution theories by exhibiting advanced cognitive abilities despite their smaller brains. While the social brain hypothesis may apply within clades, sensory-motor systems likely explain the brain size differences between average-brained ectotherms and endotherms. Evolved complex sensory-motor systems suggest that brain evolution models should expand to include sensory and motor systems, beyond cognitive processes alone.

Grooming is one strong mechanism allowing primate groups to grow larger and more cohesive, but a reduction in reactive aggression responses can be expected to have contributed to this trend too. There is indeed a partial overlap between the neurobiology of grooming and the neurobiology of reactive aggression.

To explain human social sophistication, and proximal phylogenetic steps leading to it, Dunbar claims that mentalising expands to increasingly high levels of recursion. However, the evidential basis for this claim is weak, exposing both a limitation in Dunbar’s account and in the field’s current understanding of social sophistication.

Drawing on our previous work on human trust networks, we provide further evidence of how group structure can foster group cohesion. But this work also raises doubts about two central tenets of the target paper: (1) the role assigned to cognitive abilities in group cohesion and stabilization; and (2) the emphasis on group size as the critical variable.

Living in groups is more stressful and less straightforward than we usually assume. In order to live in large groups, mammals need to find solutions that allow these stresses to be defused. I show that, in primates, this has involved successively adding increasingly costly structural, behavioural, and cognitive solutions that are dependent on the evolution of large brains. Primate social evolution consists of a series of glass ceilings on group size that restrict the habitats that species could occupy.

The commentaries debated numerous points in the target article. Many questioned the existence of a crisis and the benefits of a paradigm shift, even though none countered the listed signs of a crisis. Paradigm shifts are an important way that science progresses. There remains hope for a unifying theory, and for a reawakening of an ambitious science of visual attention.

While I agree with Rosenholtz that attention as mechanism should often be “banned”—this conception is confused and often explanatorily useless—I suggest that the real crisis is the proliferation of different, too often underspecified, mechanisms as attention. Attention is not an explainer. It is what we are trying to explain. Confusion on this point leads to unnecessary theoretical disunity.

Rosenholtz’s paper is a solid addition to a long tradition of throwing out babies with attentional bathwater (Di Lollo, 2012) (Wolfe, 2012) (Hulleman & Olivers, 2017) (Wolfe, 2017). She is correct that the term “attention” has been used in profligate and often ill-defined ways. That said, I argue that any plausible model of visual search must include visual selective attention.

Rosenholtz argues that summary statistics explain attentional phenomena via peripheral vision. While we acknowledge their role, we challenge the claim that they serve as an alternative mechanism. Instead, we argue that summary statistics and selective attention are interdependent, shaping visual perception under limited capacity, as evidenced by perceptual biases in numerosity judgments and mean estimations of shape size, color, and position.

Why have some organisms evolved processes of attention as distinct from perception in general? Investigation of freely behaving organisms (not in laboratory tasks) suggests that attention as distinct from perception is critical for goal-directed organisms’ value-based decision making. As such, the target of attention is not punctate stimuli, but whole situations (scenes) that are relevant to that decision.

We agree that while there is a “crisis” in visual attention, the Rosenholtz’ article does not offer bold enough solutions. We argue that the real crisis extends beyond attention, reflecting a broader need for theoretical integration. Addressing this requires abandoning artificial subdivisions and adopting a more ecologically valid, contextually grounded approach to cognition.

Here, we argue that Rosenholtz’ call for a paradigm shift in attention theory is unwarranted based on psychological evidence as well as philosophical theory and would disrupt scientific progress by preventing incremental science. To move forward, we suggest a different philosophical view on attention research that preserves the cumulative nature of scientific progress rather than waxing and waning theoretical paradigms.

Paradigm shifts as advocated by Kuhn (1962), should be rarely occurring and based upon true crises. The study of attention, however, is not in crisis requiring a paradigm shift but instead has a firm empirical foundation that can accommodate the findings in visual search cited in the target article.