2020 Theses Doctoral

Making the Implicit Explicit: The Effects of Summarizing Knowledge on Behavior in Repeated Decisions from Experience

Mao, Yaoli

In a dynamically changing world with unprecedented uncertainty, complexity and turbulence, continual learning and adapting is vital for one’s living and well-being. According to dual-systems accounts of cognition, learning has two major forms, implicit learning (System 1) that is fast and frugal but sometimes error-prone, and explicit learning (System 2) that is reliable but slow and effortful. These two systems are separate but must interact with each other. We gain implicit knowledge from experiencing trials and making errors (e. g., making financial investments repeatedly), receive vicarious knowledge transmitted to us in summarized forms (e.g. a quarterly report of investment options and past returns), and derive our own explicit knowledge (e.g. investment strategies) from experience to inform our future practices or to use in advising others.

The present project explores the interaction between these forms of learning in the context of repeated decisions. Is it merely implicit behavioral tendencies that are learned from experience? If so, would articulating or summarizing what is implicitly learned change subsequent choice behaviors? To address these questions, three experimental studies are conducted with online participants to investigate whether asking individuals to explicitly summarize what they have learned in a Decision from Experience (DfE) paradigm will create an explicit-implicit learning interaction that will affect their subsequent choice patterns. Decisions from explicit descriptions (DfD) refers to situations where quantitative information regarding the outcome values and probabilities of decision options is provided to the decision maker. Behavior in such situations has been found to exhibit irrational choice patterns characterized by cumulative prospect theory (CPT), overweighting the rare events while underweighting the more likely events (Tversky & Kahneman, 1992). In comparison, DfE is characterized by a different pattern of initial irrationality (underweighting the rare events while overweighting the more likely events) but moving gradually over time towards rationality as defined by Expected Value (EV)-maximization (Chen & Corter, 2014; Hertwig et al., 2004). The different choice biases between DfE and DfD is known as the Description-Experience Gap (“D-E gap”, Hertwig & Erev, 2009). The present project investigates if explicit summarization of knowledge gained from experience can affect subsequent choice patterns in DfE. Two main hypotheses are examined.

Firstly, explicit summarization might accelerate a shift to EV-maximization because summarization might promote the externalization of the implicitly learned behavior tendency in the pure DfE paradigm. A second possibility is that explicit summarization might lead to a choice pattern consistent with that in DfD characterized by a CPT-like pattern, because the summarized information of option payoffs resembles that in the DfD paradigm. In the described studies, three summarization conditions are compared including: summarizing knowledge and estimating payoff probabilities for themselves (Self condition), summarizing for another hypothetical player (Other condition), and not summarizing such information (Control condition).

The results across the three studies found a consistent summarization effect, particularly for low probability gain (Gain-Low) and high probability loss (Loss-High) problems. Those who summarized to another person (Other condition) made decisions more consistent with CPT predictions, choosing significantly more choices associated with higher CPT values. In contrast, participants in the pure DfE (Control) condition exhibited a similar DfE choice pattern, which is in the opposite direction compared to those in the Other condition. Participants in the Other condition gave more accurate probability estimates (closer to the true objective probabilities) for the risky outcomes for low-probability gains and high-probability losses. In contrast, participants in the Self condition tend to show underestimation for both high- and low-probability gains but overestimation for both high- and low-probability losses. Also, a majority of participants in the Other condition recommended to choose the EV-maximization choices in their summarizations, yet showed CPT-approximating choices in their own subsequent choices. In general, the overall findings suggest that “a probabilistic mindset” induced by the social messages in the Other condition seems to attenuate the D-E gap. Implications for learning and decision making are also discussed in the end.