The vertebrate motor system employs dimensionalityreducing strategies to limit the complexity of movement coordination, for efficient motor control. But when environments are dense with hidden action-outcome contingencies, movement complexity can promote behavioral innovation. Humans, perhaps uniquely, may infer the presence of hidden environmental dynamics from social cues, by drawing upon computational mechanisms shared with Theory of Mind. This proposed "Theory of Environment" supports behavioral innovation by expanding the dimensionality of motor exploration.
The flexibility and creativity of human behavior remain an enigma. Theories of cultural evolution explain how the emergence of conformism and imitation enabled behavioral innovations to persist and cumulate across generations, resulting in the ecological success of our species (Boyd and Richerson 1985). Behavioral innovation itself, however, remains poorly understood, often relying on assumptions of random variation that are formally analogous to genetic mutation. Here we propose a novel socio-cognitive mechanism, grounded in Theory of Mind computation (Barnby et al. 2024), that helps bridge this explanatory gap.
The human brain controls approximately 600 muscles and 350 joints to generate desirable outcomes in a 3-dimensional space, yielding a highly redundant system in which a given action objective can be realized by a vast number of possible motor configurations. To reduce this sprawling complexity (Bernstein 1967), the vertebrate motor system organizes muscular activation into coordinated “muscle synergies” (Overduin et al. 2008) that impose strategic lowdimensional constraints onto high-dimensional biomechanics. By constraining the variability of movement coordination, muscle synergies facilitate efficient whole-body control, but necessarily limit the exploration of movementcoordination structures and thus possible behaviors. Openended behavioral exploration is generally a costly investment, as evolution optimizes for multiplicative (i.e., geometric mean) fitness, where a single zero-fitness episode wipes out all prior gains. Assumptions of additive utility in reinforcement learning hence underestimate this vulnerability to exploration risk. The restricted behavioral repertoire of nonhuman primates (Tennie, Call, and Tomasello 2009) should be construed not as a functional deficit, but as reflecting a general solution to the problem of motor complexity.
Recently, leading research groups in human evolutionary biology (Morgan and Feldman 2024) and computational cognitive science (Chu, Tenenbaum, and Schulz 2024) have independently argued that the species-unique feature of human behavior is its open-ended variability. Such claims suggest that humans may have innovated the means to “unbind” acquired constraints on movement degrees-of-freedomthus becoming able to not only reduce but also expand motor exploration complexity. Recent approaches in the movement sciences illustrate how such increases in the dimensional complexity of motor coordination can facilitate skill acquisition (Dhawale, Smith, and Ölveczky 2017).
Real ecological environments typically contain an unbounded number of hidden action-outcome contingencies (i.e., environmental dynamics) that can be potentially unlocked by skill acquisition -constituting an open-ended search space. The density of these latent environmental goals (“teleological depth”) thus determines the scope of prospective future gains in the controllability of environmental outcomes (Ligneul et al. 2022;Mancinelli, Roiser, and Dayan 2021). Such untapped prospective goal-states can offset the investment cost of behavioral exploration (Molinaro et al. 2024), incentivizing learners to unbind their motor constraints, rather than remain locked into a lowdimensional repertoire optimized for known goals. But such calibration presumably requires a means to infer the teleological depth of a given environment. How might this work?
A four-fold typology of social goal inference Echoing Vygotsky (1980), we argue that teleological depth is cued by the socio-cultural environment. Without such cued information, the density of latent goals in an environment could only be probed through actual open-ended behavioral exploration -a prohibitively risky investment, as discussed. We refer to this social inference of teleological depth as theory of environment (ToE), and situate it in a 2×2 typology with other better studied mechanisms of social goal inference (Figure 1):
Goal attribution: From the first year of life, human infants expect others’ actions to be goal-directed. Infants are prolific in their attribution of goals not only to observed behaviors, but also to artifactual and natural objects, for ex-arXiv:2601.01599v1 [q-bio.NC] 4 Jan 2026 ample when interpreting the agentic purpose of wrenches or clouds (Kelemen 1999).
Theory of Mind (ToM): When observing an agent who acts upon a false belief, simple goal attribution is thwarted, instead requiring “meta-representation” of hidden mental states and counterfactual goals, i.e., theory of mind (ToM). Full-fledged ToM appears later in development than goal attribution (Gergely and Csibra 2003), and is observed reliably only in humans. Some non-human primates use a simpler, “factive” ToM that circumvents the computational cost of counterfactual simulation (Phillips et al. 2021). Due in part to this cost of counterfactual use, hypothesis-generation in ToM is constrained to the well-defined (“in-distribution”) space of known goals. This
This content is AI-processed based on open access ArXiv data.