Summary: The human brain encodes a detailed, mathematical and probabilistic representation of the aspects of our surroundings that it considers important.
Source: UPF Barcelona
How do people perceive their environment and make decisions? To interact successfully with the world, humans need more than isolated sensory evidence. Sensory information is often ambiguous and must be interpreted within a context to reduce uncertainty. At the same time, that context can be ambiguous—for example, whether a place is safe or dangerous may not be immediately clear.
A study published on 28 November in Nature Communications by Philipp Schustek, Alexandre Hyafil and Rubén Moreno-Bote, researchers at the Center for Brain and Cognition (CBC) in the Department of Information and Communication Technologies (DTIC) at UPF, provides evidence that the brain maintains a refined, hierarchical representation of uncertainty that includes contextual variables. In other words, the brain appears to form a nearly mathematical, probabilistic model of the elements in our environment that matter for perception and decision-making.
“Probability concepts are intuitive but hard to quantify precisely,” says Rubén Moreno-Bote, coordinator of the Research Group on Theoretical and Cognitive Neuroscience at the CBC. “For example, many statistics students struggle with formal probability problems. In our experiments, we show that people can intuitively solve a complex probabilistic reasoning task when it is presented in a simple, natural context.”
Cognitive tasks that require hierarchical integration
To illustrate the problem, consider a busy city airport on the day of a football final. You watch a small group of passengers disembark from a plane: four wear red and two wear blue. From that limited sample you might infer that more fans of the red team are in the city than fans of the blue team. That inference, however, could be improved by combining the observation with contextual information. If, for instance, there are generally more blue-team supporters worldwide, you would update your judgment about whether this small sample reflects broader patterns. Conversely, you might decide the sample deviates from the general context and adjust your expectations about future observations.
The researchers designed experiments using this “airplane” scenario to probe hierarchical integration. “Participants were told they were at an airport where some planes tended to carry more of one type of passenger than another—for example, more supporters of Team A than Team B,” Moreno-Bote explains. “After seeing a handful of passengers walking off several planes, participants were asked to predict, with mathematical precision, how likely the next arriving plane would carry more of one passenger type than the other.”
These tasks intentionally create hierarchical dependencies between hidden variables: observers must infer a higher-level contextual state from past observations (a bottom-up process) and then use that inferred context to interpret current sensory evidence (a top-down process). In effect, the brain is required to pass probabilistic “messages” between levels of representation to arrive at an updated belief about the current situation.
The study’s results indicate that participants built probabilistic representations of the context based on the samples they observed. Crucially, people also inferred the reliability (certainty) of that contextual information and combined it with the uncertainty of current observations when forming confidence judgments. That is, participants did not simply tally counts or rely on crude heuristics; instead, their behavior closely matched computations consistent with probabilistic message passing across hierarchical states.
These findings clarify how humans form internal models of their surroundings and how they represent and integrate uncertainty across multiple levels and timescales. The work contributes to our understanding of perception, decision-making, and confidence by showing that human observers are sensitive to both the content and the reliability of contextual cues when making probabilistic inferences.
Source:
UPF Barcelona
Media Contacts:
Press Office – UPF Barcelona
Image Source:
The image is credited to UPF.
Original Research: Open access
“Human confidence judgments reflect reliability-based hierarchical integration of contextual information”. Philipp Schustek, Alexandre Hyafil & Rubén Moreno-Bote. Nature Communications. DOI: 10.1038/s41467-019-13472-z.
Abstract
Human confidence judgments reflect reliability-based hierarchical integration of contextual information
Immediate sensory observations must be supplemented with contextual information to resolve ambiguities, but context is often ambiguous and must itself be inferred to guide behavior. The authors introduce a hierarchical “airplane” task in which participants infer a higher-level contextual variable to inform probabilistic inferences about a dependent lower-level variable. By manipulating the reliability of past sensory evidence (through sample size), the study shows that humans estimate context reliability and combine it with current sensory uncertainty to form confidence reports. Behavior aligns closely with probabilistic inference implemented via message passing between latent variables across hierarchical representations. Common inferential errors such as ignoring sample size were not observed, and participants did not appear to rely on simple heuristics. These results demonstrate uncertainty-sensitive integration across hierarchical levels and temporal scales.