The eyes of the moral mindaffect based-gain control and contextual modulation of moral decisions

Resumo

A central aim of current research in moral decision making is to propose a mechanism to explain how a moral choice is made. Most current studies are based on the traditional postulates of the Dual Process Model (DPM) (Greene, Sommerville, Nystrom, Darley, & Cohen, 2001; Haidt, 2001; Cushman, Young, & Hauser, 2006; Greene, 2007; Bartels, 2008). The automatic emotional process is posited to underlie the aversion to doing harm in up close personal dilemmas, refusing the behavior as morally acceptable, while the controlled cognitive process operates to promote welfare, maximizing choices that opt for greater number of lives saved. However, the simple division between emotional and rational components might not provide a complete picture of the moral mind. This point is appreciated by both critics and supporters of DPM arguing that affective components must be integrated in the process (Moll, Oliveira-Souza, & Zahn, 2008; Cushman, Young, & Greene, 2010; Kvaran, & Sanfey, 2010). We propose that moral decisions are shaped by the same gain control mechanism that operates in sensory pathways representing values in a relative rather than absolute manner (Carandini, & Hegger, 2012). Context should thus strongly influence how people make affective perceptions about actions and actors in a very explicit way that is independent of purely deontological (normative ethical position) or absolute utilitarian considerations (best moral action is the one that maximizes utility). To test this hypothesis we have designed 15 new, ecological moral dilemmas in which both the actors and actions were kept constant while context was sequentially altered as more information was added to the moral scenes. We have used these dilemmas in four different experiments where participants were asked about three aspects of the choice. Firstly, we were interested in the moral acceptability (Acc) of the dilemmas. Secondly, we also asked participants to evaluate the Stereotype Affect (SA), a basic reaction about the likeability of the active character on the dilemma. And finally, we have recorded the Perceived Affect (PA). This variable is the same as the previous one, but in this case, participants should judge the protagonist when she/he was involved in a given situational context. Our results show that subjects judge the protagonist of the moral action encoding a subjective value (PA), which is directly dependent on the contextual value of the scenes embedding the actions. This subjective value correlates linearly with moral acceptability. This context-dependent modulation of moral decision making is precisely described by divisive normalization, an adaptive form of gain control that may be a general mechanism for sensory and cognitive computations providing an explanation for several otherwise puzzling phenomena about decision making in general (Carandini, & Hegger, 2012). To test the dynamics of moral judgments, we have conducted an experiment where 35 subjects made moral decisions while their eye scan paths and computer mouse trajectories were continuously monitored to measure Reaction Times (RTs), response type and uncertainty creating a Doubt Index (DI). We have found that cognitive load correlates with behavioral responses. Low cognitive load mainly occurs in trials that corroborate the acceptance of a previously evaluated moral scene, while high cognitive load mainly characterize trials where the responses were deontological or the first contextual (the first accepted). Regarding to the RT’s recordings, we have not found statistically significant differences between contextual and deontological responses. Both results, the cognitive load pattern and the RT, are not consistent with the proposal of the DPM, where the deontological trials resulted of the fast automatic process, while the utilitarian ones from the slower and more rational and controlled process. In addition, we have devised the Doubt Index (DI) with two different aims. First, just considering a typical behavioral DI pattern, obtained from the eye scan paths recorded during the last two trials of each dilemma, we were able to predict the participant´s responses above chance. Second, using signal detection theory to compare objective and subjective awareness of cognitive load, we were able to show that 63.2% of the participants were aware of their own deliberative process. Finally, we have applied a Rasch Model to categorize the dilemmas depending on their difficulty and the participants’ ability (Baron, Gürçay, Moore, & Starcke, 2012). We are able to predict the participant’s output with these two criteria. In summary, a more precise characterization than the classical DPM within the moral domain is based on the value representation assigned to the moral scene by divisive normalization. Its explanatory power becomes especially strong when a single computation is enough to explain all the dilemmas presented. Current research in computational neuroscience can improve this new approach to highlight one of the most interesting domains for cognitive researchers: morality.