Exemple 1 pissarra + veu alumnat
15. Taula: Altres opcions valorades gestio recursos
How do choice option representations handle the trade-off in requirements imposed by task goals and stimulus properties? At this point a general outline of the dissertation can be presented that aims to answer this question. The general strategy of the work is divided into a top-down approach (chapters 2 and 3), working with computational models such as EUT and BMC, and a bottom-up approach (chapters 4 and 5), working with algorithmic models such as
fast and frugal heuristics and the analysis of similarity measures through RSA. Prospect theory
(a descriptive model) is also employed in chapter 2 to obtain certain parameters of interest for the study (loss aversion and risk aversion parameters). Each approach starts with presenting a behavioural study (chapters 2 and 4) and then continues with an fMRI study (chapters 3 and 5). Adding an implementational perspective to each approach through fMRI provides added value and extra analytical degrees of freedom in addressing the trade-off hypothesis for choice option representations.
1.11.1 The top-down approach
To test the hypothesis that the representation of choice options is dependent on both informational input and task goals, the thesis starts by presenting a set of studies that tests how people integrate beliefs and preferences in the social domain. The social domain was chosen as a testbed because the integration of conflicting information from different sources (self versus other) naturally arises in everyday decision-making. These studies are presented in chapters 2 and 3. They rely on the assumption that the participants in the studies are trying to maximise their expected utility with respect to potential rewards in the task. Importantly, the tasks differ with respect to the choice options that are offered and they also differ in the
assumptions regarding the computation of uncertainty. The study in chapter 3 explicitly strives to model the integration of preferences within a Bayesian update framework, which implies that people represent preferences as distributions governed by parameters with their own uncertainty. In contrast, the study in chapter 2 only assumes point estimates of certain quantities are necessary for successful task completion.
1.11.1.1 The social domain
The study in chapter 2 starts with addressing a question that is pertinent to all decision- making; does the agent want to make a choice or would the agent prefer someone else to make a choice for them? This research question queries the role of agency in opposition to the faculty of delegating a choice. The choice options are actually the (potential) decision makers themselves – the delegator and the potential delegate. The study additionally tests whether people will be willing to incur a cost as a consequence of choosing themselves.
The study in chapter 3 is an fMRI study that directly investigates the integration of new information within the context of a value-based judgment. Specifically, the study tests how social information gets integrated into value judgments while considering the reliability of that information. The social information is presented as other people’s preferences for a set of retail products. Participants are tasked to integrate this information with their own preferences. Thus, the integration of preferences of the participants are analysed in relation to the retail products, informing on the representation of these choice options. Furthermore, the study seeks to understand if such an integration of the information’s reliability is consistent with Bayesian integration of uncertainty. The neural representation of this uncertainty is also discussed.
1.11.2 The bottom-up approach
The studies in chapters 2 and 3 make assumptions on the computational goals that participants will seek to achieve while letting participants freely choose the decision strategy to accomplish this. The assumed goals are that participants are trying to maximise expected utility and that they use uncertainty in this estimate to inform that maximisation process. The exact algorithm (i.e., decision strategy) that participants use is not known. This uncontrolled aspect restricts the extent of permissible conclusions regarding choice option representations, given the intricate relation between representation and algorithm. This caveat naturally leads the research agenda to setting up a study that explicitly controls for the decision strategy being
used in the experiment (chapter 4). This greater experimental control enables testing the interaction of decision strategies with other contextual variables of interest that inform on the nature of choice option representations such as stimuli format (colour-coding, numericity, and serial order) and timing constraints. The approach in chapter 5 is deemed bottom-up because of its focus on comparing similarity measures between stimuli sets (from two separate, previously published, datasets) and between individual stimuli.
1.11.2.1 Controlling for decision strategy
Chapter 4 presents a study that investigates the effect that stimuli formats have on a pair of decision strategies; specifically, two decision heuristics known as Tallying and Take- the-Best (see chapter 4 for their definitions). These simple heuristics are easily explained to participants; hence they are explicitly instructed to use them during the task. Thus, compliance with each heuristic is the dependent variable of interest. The results in this chapter show that different algorithms require different cognitive processes such that a stimulus format that is good for one algorithm is not adequate for the other.
1.11.2.2 Fundamental representational issues in the human brain
The study in chapter 5 seeks to deconstruct the very nature of the neural representations of choice options down to their core essence; that of similarity relations. It opens a new line of inquiry by questioning the appropriateness of various similarity measures (not necessarily distance metrics) for analysing fMRI data. Formal competition amongst measures is required to evaluate the brain’s representational capacities. The findings here show that similarity is computed consistently across brain areas but is modulated by stimulus format and/or task.