Neural basis of decision making and fostering decision superiority

A ubiquitous phenomenon in cognitive psychology, which reflects strategic decision making, is the speed accuracy trade off (SAT); the faster individuals make a decision, the more likely they are to make an error. Previous imaging work has implicated frontal (including the pre-supplementary motor area) and striatal regions, and connectivity between the two, in the SAT, but the nature of this association remains unclear and, to date, previous studies on this topic have employed small sample sizes. Here, we report a large-scale pre-registered, study combining transcranial direct current stimulation (tDCS), ultra-high field (7T) imaging, and computational modelling investigating white matter connectivity and its association with the influence of brain stimulation of decision-making strategy. Across two stimulation sessions, participants completed a task with an instruction-based manipulation of decision strategy – prioritise speed or accuracy, or balance the two – whilst discriminating motion from a patch of moving dots. Offline cathodal stimulation (or sham) was applied at 0.7mA to the superior medial frontal cortex or the left prefrontal cortex. Utilising an individual differences approach, imaging of white matter tracks provided evidence for a role of structural connectivity between frontal and striatal regions in both the SAT and the efficacy of tDCS to modulate performance. Further, group level effects of tDCS on the SAT were found following stimulation to both regions, replicating previous findings. This work provides causal evidence for the involvement of both the left prefrontal and superior medial frontal cortex in the SAT and highlight a role of fronto-striatal connectivity in decision making strategy more broadly. 

Risk preferences – the willingness to accept greater uncertainty to achieve larger potential rewards – determine many aspects of our lives and are often interpreted as an individual trait that reflects a general ’taste’ for risk. From a neural perspective, this subjective attitude towards risk is often proposed to be determined by properties of neural subjective value calculations. However, this perspective cannot explain why risk preferences can change considerably across contexts and even across repetitions of the identical decisions. Here we provide modelling, fMRI, and TMS evidence that contextual shifts and moment-to-moment fluctuations in risk preferences can emerge mechanistically from Bayesian inference on noisy magnitude representations in parietal cortex (rather than in subjective-value-coding areas like ventro-medial PFC). Our participants underwent fMRI while choosing between safe and risky options that were either held in working memory or present on the screen. In a second experiment, we applied continuous theta-burst magnetic stimulation over individually-defined parietal magnitude areas or a vertex control site, before fMRI of the same behavioural paradigm. Risky options that were held in working memory were less likely to be chosen (risk aversion) when they had large payoffs but more likely to be chosen (risk-seeking) when they had small payoffs. These counterintuitive effects are mechanistically explained by a computational model of the Bayesian inference underlying the perception of the payoff magnitudes: Options kept in working memory are noisier and therefore more prone to central tendency biases, leading small (or large) payoffs to be overestimated (or underestimated) more. Congruent with the behavioural modelling, fMRI population-receptive field modelling showed that on trials where intraparietal payoff representations were noisier, choices were also less consistent and less risk-neutral, in line with participants resorting more to their prior belief about potential payoffs. Finally, the same model could also account for behavioural and neural effects of continuous theta-burst magnetic stimulation over the individually-defined parietal magnitude representations: This led to lower choice consistency and a larger tendency towards risk-seeking choices, particularly when safe options were presented before risky options. These behavioural changes were accompanied by reduced fidelity of neural magnitude representations, as reflected by decreased nPRF amplitudes, noisier neural signals, and reduced decoding accuracy for payoff magnitudes. Individual estimates of the increase in noise of our computational decision-making model correlated with the reduction in nPRF amplitude after parietal cTBS across subjects. Our results highlight that individual risk preferences and their puzzling changes across contexts and choice repetitions are mechanistically rooted in perceptual inference on noisy parietal magnitude representations, with profound implications for economic, psychological, and neuroscience theories of risky behaviour. 

We have shown that the lateral hypothalamus differently regulates learning about cues depending on the relative temporal distance of the cues to rewarding outcomes (Sharpe et al., 2017, Current Biology; Sharpe et al., 2021, Nature Neuroscience). This opens up a new avenue for neuroscience research, which is to understand the dynamics of how the brain prioritises learning and behaviour towards information most relevant to desirable outcomes. To formally investigate this using computational modelling and optogenetics in rats, we adapted the “Daw two-step task”, which quantifies the ability of human subjects to use complex task structure to predict rewards (e.g., Daw et al., 2011, Neuron). This task enables us to measure the weights of distal and proximal cues on future decision making. In our task, rats first receive one of two distal cues followed by presentation of two levers. Rats press one of the levers and then receive one of two proximal cues. The distal cues inform the probabilistic state transitions from the lever choice to the proximal cues. In turn, the proximal cues inform the fluctuating reward probabilities. We found that rats are able to guide their choices by using the transitional structure of the task, including significant weightings on decision making by both the distal and proximal cues. We are now combining our task with optogenetic approaches to parse the contribution of lateral hypothalamus and orbitofrontal cortex in the balance of learning and behaviour between distal and proximal cues of rewards. 

Real-life choices often require striking a balance between the value of choice outcomes and their likelihood. When taking out flood insurance, for example, it is important to consider both the cost of any damage and the probability of flooding. Rational choice theory assumes that risky choices rely on an optimal integration of choice values with their probabilities. The literature, however, is replete with examples of irrational, biased choices, and these observations have motivated piecemeal modifications of the theory. Recently proposed cognitive imprecision theory can account for most reported biases in a principled fashion by assuming that noisy subjective representations of choice value and probability are integrated optimally. Here, we developed a novel perceptual game to test a key prediction of this theory, namely, that the temporal dynamics of risky choices and their neural correlates should co-vary with cognitive noise. Forty healthy, adult humans first estimated the average orientation of a briefly presented circular array of twelve differently oriented gratings and then chose to either play the game or not (risky and safe choices). If they chose to play, they were awarded points proportional to the average orientation in that trial. Noise was manipulated by randomly switching between high and low variability in the orientations of displayed gratings. Computational modelling of behaviour showed that cognitive noise impacted estimates of loss and risk aversion. Similarly, multivariate feature-specific analyses of functional brain activity showed an effect of noise on the precision of neural value representations, lending neurobiological support to the cognitive imprecision theory.