TY - JOUR
T1 - Human inference reflects a normative balance of complexity and accuracy
AU - Tavoni, Gaia
AU - Doi, Takahiro
AU - Pizzica, Chris
AU - Balasubramanian, Vijay
AU - Gold, Joshua I.
N1 - Funding Information:
We thank A. Filipowicz for sharing the codes to run the psychophysical experiment for the Bernoulli task. We also thank K. Krishnamurthy and E. Piasini for interesting discussions, and A. Cavagna and A. Ingrosso for pointing out a possible connection between one of our results and spin-glass systems. G.T. was supported by the Swartz Foundation (award no. 575556) and the Computational Neuroscience Initiative of the University of Pennsylvania, and is currently supported by Washington University in St. Louis. V.B. and J.I.G. are supported in part by NIH BRAIN Initiative grant no. R01EB026945. J.I.G. is also supported by grant nos R01 MH115557 and NSF-NCS 1533623. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Publisher Copyright:
© 2022, The Author(s), under exclusive licence to Springer Nature Limited.
PY - 2022/8
Y1 - 2022/8
N2 - We must often infer latent properties of the world from noisy and changing observations. Complex, probabilistic approaches to this challenge such as Bayesian inference are accurate but cognitively demanding, relying on extensive working memory and adaptive processing. Simple heuristics are easy to implement but may be less accurate. What is the appropriate balance between complexity and accuracy? Here we model a hierarchy of strategies of variable complexity and find a power law of diminishing returns: increasing complexity gives progressively smaller gains in accuracy. The rate of diminishing returns depends systematically on the statistical uncertainty in the world, such that complex strategies do not provide substantial benefits over simple ones when uncertainty is either too high or too low. In between, there is a complexity dividend. In two psychophysical experiments, we confirm specific model predictions about how working memory and adaptivity should be modulated by uncertainty.
AB - We must often infer latent properties of the world from noisy and changing observations. Complex, probabilistic approaches to this challenge such as Bayesian inference are accurate but cognitively demanding, relying on extensive working memory and adaptive processing. Simple heuristics are easy to implement but may be less accurate. What is the appropriate balance between complexity and accuracy? Here we model a hierarchy of strategies of variable complexity and find a power law of diminishing returns: increasing complexity gives progressively smaller gains in accuracy. The rate of diminishing returns depends systematically on the statistical uncertainty in the world, such that complex strategies do not provide substantial benefits over simple ones when uncertainty is either too high or too low. In between, there is a complexity dividend. In two psychophysical experiments, we confirm specific model predictions about how working memory and adaptivity should be modulated by uncertainty.
UR - http://www.scopus.com/inward/record.url?scp=85131039062&partnerID=8YFLogxK
U2 - 10.1038/s41562-022-01357-z
DO - 10.1038/s41562-022-01357-z
M3 - Article
C2 - 35637296
AN - SCOPUS:85131039062
SN - 2397-3374
VL - 6
SP - 1153
EP - 1168
JO - Nature Human Behaviour
JF - Nature Human Behaviour
IS - 8
ER -