Impact of technical reasoning and theory of mind on cumulative technological culture: insights from a model of micro-societies

The paper investigates which cognitive skills are necessary for the emergence and maintenance of cumulative technological culture (CTC). Prior work often emphasizes high-fidelity imitation supported by social-cognitive skills such as theory of mind (ToM). An alternative, the technical-reasoning (TR) hypothesis, posits that reasoning about physical object properties can drive both transmission and innovation in technology. The authors aim to integrate these views in a computational model inspired by micro-society paradigms, examining three social-learning forms—reverse engineering (RE), observation (OBS), and communication (COM)—and two cognitive skills (TR and ToM). They first model refinement-only in a closed solution space to mirror typical micro-society studies, then add innovation to create an open-ended space, analyze single-chain dynamics, and finally compare models with and without ToM to assess its necessity for CTC. The overarching hypothesis is that TR can underwrite CTC, while ToM mainly accelerates growth and stabilizes transmission but is not strictly necessary for CTC to emerge.

The authors synthesize a large body of work on CTC, noting that many models address macro-level factors such as demography, copying biases, population structure, language, number of traits, migration, and transmission fidelity. However, these leave open the role of micro-level cognitive skills (e.g., ToM, meta-cognition, TR, working memory). Classical perspectives link CTC to ToM-enabled high-fidelity imitation, intentional teaching, and shared intentionality. Yet micro-society experiments have shown cumulative improvements even under reverse engineering, challenging the necessity of explicit teaching or ToM. Recent empirical results indicate that technical reasoning best predicts cumulative performance across RE, OBS, and COM, and improvements in a system co-occur with increased causal understanding. This supports the TR hypothesis: non-social cognitive skills plus basic social learning can generate CTC; ToM serves as a catalyst, particularly in opaque contexts.

The authors build an agent-based computational model of a micro-society transmitting a single technology T over generations. Technology is decomposed into n semi-independent traits, each with a quality and an upper limit; overall technology quality is the sum of trait qualities, and limit(T) is the sum of trait limits. The environment quality env is a noisy function of technology quality (env = quality(T) + ε, ε uniform in [−quality(T), quality(T)]). Individuals possess trait-specific knowledge levels (cogi) and two cognitive skills: technical reasoning (ITR) and theory of mind (ITOM). ITR is drawn from a Gaussian bounded by [0, env], reflecting acquisition from the technological environment. ITOM is drawn from a normal distribution centered at 0.5 (bounded [0,1]). Initial cogi are sampled without replacement from a pool totaling ITR so that Σ cogi = ITR. Learning occurs via three social-learning forms: - RE: access only to technology; update cogi by an amount proportional to observable differences and modulated by receiver’s TR. No ToM term. - OBS: receiver watches the sender modify the technology without communication; OBS implies RE, so cogi increase by RE plus δOBS, where δOBS = quality(Tt) − quality(Tt−1). - COM: direct sender–receiver communication; receiver benefits from RE and OBS plus a communication term δCOM scaled by receiver TR and sender ToM: cogi ← cogi + (RE) + (δOBS) + (δCOM × ITR × (STOM + 1)), where δCOM = sender.cog − receiver.cog and STOM is the sender’s ITOM. STOM acts as a catalyst; with STOM = 0, COM reduces to RE/OBS dynamics dependent on receiver TR. Modification (refinement) step: After learning, the individual attempts to modify each trait. The probability of improvement Pimprove increases with ITR; Pdeteriorate = 1 − Pimprove, with a constraint that individuals with above-average ITR (ITR > n) cannot deteriorate. Trait updates are multiplicative: if improving and cogi ≥ current trait quality, multiply by βimprove (>1); if deteriorating and cogi < trait quality, multiply by βdeteriorate (<1); otherwise, the trait remains unchanged. Innovation step: With probability Pinvention = 0.01 per generation, an innovation opportunity arises. Innovation occurs if either ITR or current technology quality is in the top 20% (≥0.8 of maximum). Innovation adds a new trait (groundbreaking innovation) and increases the limits of all existing traits by a fixed strength θ (innovative combination). This opens an unbounded, open-ended solution space. Parameter defaults (unless stated): limit(tech) = 2, n = 2, θ = 5, βimprove = 1.2, βdeteriorate = 0.8, Pinvention = 0.01. Simulations typically run 1000 generations and are averaged over 1000 runs; additional analyses fix random seeds to examine single-chain dynamics.

Refinement-only (closed solution space): - All learning forms exhibit initial growth then plateau, but at different levels and with different stability. COM reaches the optimized-technology level fastest and maintains it stably; RE and OBS reach plateaus later and show fluctuations around the limit. - Time to reach plateau/optimized level: COM markedly earlier than OBS and RE (e.g., COM: M = 8.68, SD = 25.01; OBS: M = 35.71, SD = 22.68; RE: M = 42.55, SD = 2.92). Increasing limit(tech) accentuates COM’s advantage. Refinement + innovation (open-ended): - All conditions show initially exponential increases in quality; COM outperforms OBS and RE at the last generation: RE M = 213.11 (SD = 141.13), OBS M = 266.39 (SD = 166.06), COM M = 591.87 (SD = 322.39). - Asymmetry in gains/losses: when improvement steps are weak relative to deterioration (βimprove = 1.1, βdeteriorate = 0.8), RE and OBS quality drop markedly (RE M = 24.67; OBS M = 45.41) but still exhibit CTC; COM remains robust (M ≈ 601.87). When βimprove increases (βimprove = 1.3), RE and OBS increase (RE M = 321.67; OBS M = 368.43), COM modestly increases (M = 635.56) as it already often reaches limits. - Innovation-combination strength θ: increasing θ lifts limit(T) and benefits all forms; OBS approaches COM as θ grows (e.g., θ=5: OBS M = 266.39 vs COM M = 591.87; θ=50: OBS M = 4602.15 vs COM M = 6161.22), while the gap between RE and OBS widens due to higher dependence on ITR for innovation thresholds and a snowball effect linking better technology to higher ITR and further gains. Single-chain dynamics: - COM displays punctuated patterns: long stases at limits punctuated by rapid bursts following innovations; RE and OBS reach limits later (e.g., RE M = 63.2; OBS M = 52.4; COM M = 18.4 for reaching optimized level) and lack sharp bursts, with more fluctuations and sometimes failure to reach optimized levels. Averaging across chains masks these heterogeneous trajectories for RE/OBS. Role of ToM: - Comparing COM with vs without ToM shows minimal differences in average technology quality over 1000 generations. ToM’s effect is concentrated early, marginally accelerating initial ascent and reducing variance in transmission; once near-optimal, progress depends primarily on TR and innovation.

The model shows that technical reasoning (TR) can support both the acquisition and modification of technology, enabling cumulative technological culture to emerge even without sophisticated social-cognitive mechanisms. Communication (COM), which often engages teaching and ToM, accelerates reaching high-quality solutions and stabilizes transmission, reducing cultural losses; however, ToM is not strictly necessary for CTC’s emergence. In open-ended contexts with innovation, TR drives a snowball dynamic whereby improved technology and increased understanding mutually reinforce each other, producing exponential growth. Single-chain analyses reveal punctuated equilibria under COM—bursts following innovations separated by stasis at local optima—mirroring archaeological and theoretical accounts of cultural evolution. These findings reconcile macro-level demographic and structural determinants with micro-level cognition by positioning TR as a core engine of cumulative change and ToM as an early-stage catalyst and variance reducer, particularly in opaque tasks or communication-rich settings.

This work contributes a micro-society computational framework integrating technical reasoning and theory of mind to study cumulative technological culture. It replicates micro-society patterns (COM > OBS ≈ RE) and demonstrates that TR alone can generate and maintain CTC, while ToM speeds early growth and stabilizes transmission but is not essential. With innovation, the model yields open-ended, exponential improvements driven by a TR-mediated snowball effect. Future research should incorporate dependencies among traits, multiple technologies and interacting micro-societies, and additional cognitive factors (e.g., creativity, memory, metacognition, altruism), and delve deeper into the cognitive mechanisms of innovation itself. Extending beyond a single technology and refining observational learning implementations would offer a more comprehensive account of real-world cultural evolution.

Key limitations include: (1) Simplified social-learning forms and their implementation (e.g., teaching can occur in RE/OBS in reality; literature definitions of OBS vary). (2) Focus solely on TR and ToM, omitting other relevant cognitive abilities (e.g., metacognition, memory, creativity, altruism). (3) Assumption of semi-independent traits; no explicit inter-trait dependencies, which would increase realism but also complexity. (4) A single-technology environment; real settings involve multiple technologies and interacting groups. (5) Innovation modeled as an event with fixed probability and thresholds, without detailed cognitive mechanisms. These simplifications were chosen to align with micro-society paradigms and isolate TR/ToM effects, but they constrain generalizability.