Efficient sampling and noisy decisions

Abstract

Human decisions are based on finite information, which makes them inherently imprecise. But what determines the degree of such imprecision? Here, we develop an efficient coding framework for higher-level cognitive processes in which information is represented by a finite number of discrete samples. We characterize the sampling process that maximizes perceptual accuracy or fitness under the often-adopted assumption that full adaptation to an environmental distribution is possible, and show how the optimal process differs when detailed information about the current contextual distribution is costly. We tested this theory on a numerosity discrimination task, and found that humans efficiently adapt to contextual distributions, but in the way predicted by the model in which people must economize on environmental information. Thus, understanding decision behavior requires that we account for biological restrictions on information coding, challenging the often-adopted assumption of precise prior knowledge in higher-level decision systems.

Neural codes in early sensory areas maximize fitness

“…One may think that this criterion makes sense for early sensory systems as this is precisely the role of a sensor: a good measurement instrument must reliably measure the environmental variable that it was built for. However, the information maximization criterion does not necessarily take into consideration the behavioral needs of the organism (15)(16)(17)(18)(19). In line with this idea, there is evidence showing that early sensory systems represent not only information about physical sensory inputs, but also non-sensory information according to the requirements of a specific task and the behavioral relevance of the stimuli (20)(21)(22)(23)(24).…”

Variance misperception under skewed empirical noise statistics explains overconfidence in the visual periphery

“…Incorrect priors in general, regardless of simplification, can lead to illusions in many areas of perception (Flanagan et al, 2008; Geisler & Kersten, 2002; Teufel et al, 2013; Valton et al, 2019; Weiss et al, 2002), and even if skewed priors are initially learned appropriately it seems the system has difficult using them optimally in cue integration (Acerbi et al, n.d.). Therefore, perhaps due to biological constraints restricting information coding (Heng et al, 2020), we’ve shown here how it is possible that in metacognition, too, efficient, simplified coding of noise expectations in the periphery may lead to overconfidence relative to performance capacity.…”

Normative principles for decision-making in natural environments

“…If valuation is conceived as a sampling process, and preferences are dynamically constructed by retrieving memories of past experiences to a current prospect, then utility and probability functions will be naturally tailored to match previously encountered distributions of value and probability. These distributions may favour low values and extreme probabilities, thereby accounting for idiosyncrasies in value (Bhui & Gershman 2018;Heng et al 2020;Stewart et al 2006). A different view suggests that biases at the extremities of the probability weighting function -whereby high and low probabilities are biased away from certainty -is a rational strategy for a noisy decision-maker, because it corrects for an estimation bias that occurs when variability in the inference process is irreducible (Steiner & Stewart 2016).…”