Human Biases Limit Cumulative Innovation

Abstract

Is technological advancement constrained by biases in human cognition? People in all societies build on discoveries inherited from previous generations, leading to cumulative innovation. However, biases in human learning and memory may influence the process of knowledge transmission, potentially limiting this process. Here we show that cumulative innovation in a continuous optimization problem is systematically constrained by human biases. In a large (n = 1,250) behavioral study using a transmission chain design, participants searched for virtual technologies in one of four environments after inheriting a solution from previous generations. Participants converged on worse solutions in environments misaligned with their biases. These results substantiate a mathematical model of cumulative innovation in Bayesian agents, highlighting formal relationships between cultural evolution and distributed stochastic optimization. Our findings provide experimental evidence that human biases can limit the advancement of knowledge in a controlled laboratory setting, reinforcing concerns about bias in creative, scientific, and educational contexts.

Cultural selection and biased transformation: two dynamics of cultural evolution

“…Similarly, Model 2 illustrates that cross-cultural stability in cultural evolution can be generated by either biased transformation or cultural selection (plus migration), again reflecting previous model results [38,[50][51][52]. Like for directional change, the presence of cross-cultural stability alone cannot be taken as evidence for one or the other.…”

“…This seems like a reasonable assumption given empirical case studies to date, each of which find or assume a single attractor, e.g. blood-letting [34], direct eye gaze [35], colour terms consistent with the World Colour Survey [36], linear relationships between variables [32] or symmetrical arrowhead designs [38]. However, where there are different ecological pressures in different areas, then biased transformation may equally generate cross-cultural divergence (for an experimental demonstration of this, see [83]).…”

“…The R code for both models is available at https://github.com/amesoudi/culturalselection_biasedtransformation, and readers can play with them without needing to run the code at https://amesoudi.shinyapps.io/CSBT_model1/ and https://amesoudi.shinyapps.io/CSBT_model2/. These models are intended to complement previous models that examine these cultural dynamics [3,38,42,47,[50][51][52]. The insights are not particularly novel compared to these previous studies, but (i) they provide a replication and confirmation of previous models' results using different assumptions and implementations, thus contributing to a broad family of models addressing the same issue (which is often better than seeking a single 'perfect' model: [49]); (ii) their insights are integrated below with more recent empirical studies and placed within the context of ongoing debates concerning the nature of cultural evolution; and (iii) full code and interactive online versions of the models are provided alongside this paper, to allow others to directly explore the results and extend the models using different assumptions.…”

A Quantum Walk Model For Emotion Transmission In Serial Reproduction Of Narratives

“…Emotion transmission refers to how the intensity of emotions conveyed in narratives develops during serial reproductions. Previous research (Thompson & Griffiths, 2021;Stubbersfield, Tehrani, & Flynn, 2015;Breithaupt et al, 2022) found that Figure 2: The scatter plots of models' predictions and the data. For all sequences whose heads are the selected original narratives displayed in the titles, the ratings x n+1 from the data, the predicted rating of xn+1 given data x n by the best version of the Bayesian model (in terms of lowest mean BIC for the emotion), and that by the best version of the quantum model, are all plotted against x n from the data.…”

Negative observational learning might play a limited role in the cultural evolution of technology

“…The participants can access only the fitness outcomes of their artifacts but not directly the shape of the underlying landscape itself. This framework was originally introduced to explain the cultural variability of archaeological artifacts in the Great Basin (Nevada and California in the US) and was subsequently applied to the investigation of the social learning strategies of individuals 17 , their cultural differences 42 , or the role of inductive bias in cultural evolution 43 . Using this framework, a series of experiments showed that the participants were highly dependent on social information and tended to copy the most successful individual 17 , 39 , 41 , 42 .…”