Task-evoked pupil responses reflect internal belief states, Scientific Reports (2018)
By Olympia Colizoli, Jan Willem de Gee, Anne E. Urai & Tobias Donner
Pupil diameter is a reliable marker of central arousal states and neuromodulatory responses. The pupil dilates whenever people engage in decisions. Previous work from our group (Urai et al., 2017) showed that these pupil responses evoked by decisions are specifically driven by the uncertainty about one’s decision: the probability that a choice just made is wrong, given the evidence on which the choice was based.
Interestingly, a similar encoding of decision uncertainty was recently observed in the activity of dopamine neurons in the midbrain – only with opposite sign, reflecting the complement of decision uncertainty: confidence (Lak et al., 2017). Those authors rewarded monkeys for correct decisions and found reflections of graded decision confidence not only before, but also after the rewarding feedback. The response after feedback is known as the ‘reward prediction error’: the difference between the expected and received reward. Reward prediction errors are important signals for learning, as such signals provide the information that can be used for improving one’s model of the outside world in order to make better decisions in the future.
We reasoned that we might find similar signals as those observed in dopamine neurons also in pupil responses. Indeed, responses in dopaminergic regions of the midbrain are uniquely correlated with pupil responses during perceptual decisions (de Gee et al, 2017). This correlation holds even after controlling for the known correlations between responses of the noradrenergic locus coeruleus and superior colliculus on the one hand and pupil dilation on the other hand. Several other observations also speak for a link between central dopamine release and pupil dilation (e.g., Manohar & Husain, 2015).
In the current paper, we measured pupil responses of human participants in a similar task to the one used by Lak and colleagues. Participants judged the overall motion direction of cloud of dots (up or down). There were two difficulty levels and participants received a monetary reward after each correct choice. There were long and variable delays between choices and feedback, allowing us to quantify the independent and sustained components of all responses. We related the measured pupil responses to predictions of a statistical model (Figure 1a), in which reward prediction errors are based on a noisy internal representation of the outside world – this representation is transformed into the observer’s graded ‘belief’ in a correct decision (Figure 1b,c). This approach enabled us to show that evoked pupil responses reflect a graded level of decision uncertainty, both before and after feedback (Figure 1d, compare with Figure 1b,c).
Our results imply that pupil responses, which can be measured non-invasively, provide access to an agent’s internal belief states and (belief-state modulated) reward prediction errors. Choice reaction times also scale with the uncertainty about the correctness the choice (e.g., Sanders et al, 2016; Urai et al, 2017). However, no overt behavioral marker has so far been linked to reward prediction error signals. The insight that pupil dilation constitutes such a marker may be of great use for future studies of learning, decision-making, and meta-cognition.
Our findings also provide further support for the hypothetical link between pupil and central dopamine responses (Figure 2), a link that would also have important implications for future studies into the physiology and pathophysiology of the dopaminergic system. This hypothesis should now be evaluated in invasive physiological recordings in animals.
Colizoli, O., de Gee, J. W., Urai, A. E. & Donner, T. H. Task-evoked pupil responses reflect internal belief states. Scientific Reports 8, 13702 (2018).
de Gee, J. W. et al. Dynamic modulation of decision biases by brainstem arousal systems. eLife 6, (2017).
Lak, A., Nomoto, K., Keramati, M., Sakagami, M. & Kepecs, A. Midbrain Dopamine Neurons Signal Belief in Choice Accuracy during a Perceptual Decision. Curr. Biol. 27, 821–832 (2017).
Manohar, S. G. & Husain, M. Reduced pupillary reward sensitivity in Parkinson’s disease. NPJ Park. Dis. 1, 15026 (2015).
Sanders, J. I., Hangya, B. & Kepecs, A. Signatures of a Statistical Computation in the Human Sense of Confidence. Neuron 90, 499–506 (2016).
Urai, A. E., Braun, A. & Donner, T. H. Pupil-linked arousal is driven by decision uncertainty and alters serial choice bias. Nat. Commun. 8, 14637 (2017).