Daily Summary for 2026-10-23

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: It's the twenty-third of October, twenty twenty-six, and this is the day's research.

Warren: One new paper came out today.

Ruth: I'm Ruth.

Warren: And I'm Warren, here with Iris.

Ruth: We'll take the day in one pass, then pull out the paper we're staying with.

Warren: Start us off with Learning in a Small/Big World.

The day's research: Ruth: Today we focus on how cognitive limitations interact with environment structure during uncertain decisions.

Warren: That interaction profoundly affects learning and adaptation processes.

Iris: We examined how finite automata can model belief formation to see optimal learning in small versus large worlds.

Ruth: The paper Learning in a Small/Big World contrasts low complexity environments with high complexity ones relative to cognitive ability.

Warren: In very small worlds, the optimal long-run average payoff and action distribution are well approximated by the Bayesian benchmark.

Iris: This suggests standard probabilistic reasoning functions best when things are simple.

Ruth: Optimal ignorance means ignoring states when the world is large.

Warren: That suggests focusing limited cognitive power on a smaller subset of states.

Iris: This points toward non-Bayesian learning, like using heuristics or skipping correlations.

Ruth: The link between environmental complexity, cognitive ability, and these behaviors is established in Learning in a Small/Big World.

Warren: What does that paper specifically show about the relationship between those factors?

Iris: It uses finite automata to study how people learn best in simple versus complex environments relative to their minds.

Ruth: So, ignoring possibilities seems smarter than trying to know everything when the world is big.

Warren: That idea of smart ignorance is key when complexity limits our ability to process everything.

Iris: Learning in a Small/Big World shows how finite automata model this learning preference under uncertainty.

Ruth: It illustrates that in big worlds, ignoring certain states can be optimal for decision makers.

Warren: So, the paper suggests non-Bayesian patterns emerge when cognitive resources are constrained by complexity.

Iris: Indeed, it establishes that environmental complexity directly affects how people learn and make decisions under uncertainty.

Ruth: It seems the finite automata model provides a framework for this optimal ignorance concept.

Warren: That framework helps explain why neglecting correlations becomes a viable learning strategy sometimes.

Iris: Learning in a Small/Big World explores how limited ability interacts with environmental size to shape learning outcomes.

Ruth: So, the research confirms that complexity dictates whether we should seek complete knowledge or selective focus.

Warren: It shows that the potential for non-Bayesian behavior is tied directly to those cognitive limitations.

Iris: It demonstrates that in large environments, focusing resources selectively is a valid adaptive strategy.

Ruth: That means ignoring states isn't just error; it can be an optimal response to overwhelming information.

Warren: This supports the idea that heuristics become useful when exhaustive knowledge acquisition is impossible given the scale.

Iris: The study confirms that big worlds favor strategies where cognitive resources are not stretched thin across all possibilities.

Ruth: Learning in a Small/Big World presents evidence for this selective focus under uncertainty.

Warren: It shows how finite automata simulate this learning process across environments of varying complexity.

Iris: It highlights that ignoring possibilities is a smart way to navigate environments too large for full cognitive mapping.

Lucky paper: 2009.11917: Ruth: So we're diving into the specifics of Learning in a Small/Big World now.

Warren: It models a decision maker trying to learn the true state of a world from a finite state space where N measures complexity and M memory states measure cognitive ability.

Iris: The setup defines small worlds as cases where N over M is small, but big worlds are defined as when that ratio increases.

Ruth: In those small worlds, the asymptotic behavior of the decision maker is close to that of a Bayesian individual.

Warren: Specifically, the paper states that if N over M goes to zero, the decision maker with bounded memory almost always makes the same guess as a Bayesian individual.

Iris: That means in simple environments, increasing memory states M makes the behavior closer to a true Bayesian one.

Ruth: But when N over M increases in big worlds, the decision maker starts making more mistakes and his behavior deviates more from the Bayesian model.

Warren: The paper says that in big worlds, it could be optimal for the decision maker to ignore some states entirely.

Iris: That concept of ignorance in big worlds is linked to several documented learning biases like heuristics and correlation neglect.

Ruth: It suggests that this "ignorant" behavior encompasses things like persistent over-confidence and inattentive learning.

Warren: The paper argues that the optimal learning strategy could resemble a large set of these ignorant behaviors, like using heuristics from Tversky and Kahneman one thousand nine hundred seventy-three.

Iris: And it also supports assumptions in models of bounded rationality, including misguided learning from Heidhues and K˝oszegi and Strack two thousand eighteen.

Ruth: The paper also touches on disagreement among individuals in big worlds, which is a really interesting part.

Warren: In big worlds, individuals with different prior beliefs or cognitive abilities might adopt different optimal learning mechanisms.

Iris: The paper suggests that these individuals could disagree with each other with probability one even after receiving the same infinite sequence of public information.

Ruth: That means the driving force of disagreement is actually their difference in cognitive ability M, not just their initial priors.

Warren: That result from Learning in a Small/Big World illustrates the link between complexity and non-Bayesian learning behaviors.

Iris: It explains a wide range of behavioral anomalies using the same mechanism: efficient allocation of limited cognitive resources given complexity.

Ruth: The paper also points to future research directions, specifically how individuals construct their perceived state space.

Warren: The paper suggests that ignorance might mean individuals only include the most important or a priori probable states in their state space construction.

Iris: And the reminder effect suggests that this entire construct of the state space also depends on what information the individual actually receives.