Periodic Reporting for period 1 - NEUROCIRCLE (Probing the neural processes underpinning perceptual decisions on a continuum)
Reporting period: 2019-07-01 to 2021-06-30
The overarching goal of this project was to better understand the kinds of simple perceptual decisions that people make throughout the day. Sensorimotor decisions, in which we perceive a stimulus and respond with an appropriate action underlie many of our everyday interactions with the world. As fundamental building blocks of cognition, there can be serious consequences when these decision processes are impaired due to neurological disorders or ageing. There is a massive interdisciplinary research effort aimed at understanding the cognitive mechanisms underpinning perceptual decision making, but the vast majority of this research employs decision tasks with just two alternative outcomes. For example, a classic task in psychology and neuroscience involves looking at a cloud of moving dots and identifying whether they are moving more to the left or to the right. This work has facilitated the development of detailed computational models of cognitive processes that can explain people’s responses and response times. These models are used to estimate parameters corresponding to meaningful psychological constructs, such as the speed of information processing and response caution, which can then be compared across interesting task conditions or individuals. However, many of our everyday decisions cannot be boiled down to just two alternatives. When playing sports, cooking, or walking down the street, we face a constant stream of perceptual decisions for which the potential outcomes lie on a continuum. In this project we took advantage of a new cognitive model for perceptual decisions with continuous outcomes, to better understand these types of decisions.
Decision neuroscience research shows that perceptual decision processing takes place simultaneously at multiple levels in the brain, suggesting a continuous flow of information from sensory to motor systems. With non-invasive brain recordings from electroencephalography (EEG), we can identify distinct signals representing sensory evidence encoding, the accumulation of evidence forming a decision, and motor signals related to preparing the response. Computational models that are designed solely to explain decision-making behaviour are not equipped to distinguish between processing at these different levels. Thus, the aim of this project was to use neural decision signals at multiple levels of processing in the human brain to more clearly parse the psychological effects identified by cognitive models. Furthermore, we wanted to use these decision signals to design more complex models that better reflect the hierarchical structure of decision processing. We found that incorporating neural signals into cognitive models of decision making allowed us to identify important decision process components that are missed by more traditional behavioural modelling.
Decision neuroscience research shows that perceptual decision processing takes place simultaneously at multiple levels in the brain, suggesting a continuous flow of information from sensory to motor systems. With non-invasive brain recordings from electroencephalography (EEG), we can identify distinct signals representing sensory evidence encoding, the accumulation of evidence forming a decision, and motor signals related to preparing the response. Computational models that are designed solely to explain decision-making behaviour are not equipped to distinguish between processing at these different levels. Thus, the aim of this project was to use neural decision signals at multiple levels of processing in the human brain to more clearly parse the psychological effects identified by cognitive models. Furthermore, we wanted to use these decision signals to design more complex models that better reflect the hierarchical structure of decision processing. We found that incorporating neural signals into cognitive models of decision making allowed us to identify important decision process components that are missed by more traditional behavioural modelling.
We developed a continuous-outcome version of the classic dot motion decision task in which participants had to identify the direction of motion from a complete 360 degree range, and report their response using either a joystick or eye movement. We isolated a novel sensory-level EEG decision signal representing the strength of the dot motion in visual areas of the brain, and a signal associated with the accumulation of sensory evidence, both of which were stronger for more accurate decisions. We found that individuals had strong biases towards particular directions of motion, but these varied by individual and were idiosyncratic. These directional biases were associated with the evidence-accumulation EEG signal but not the sensory-level signal, suggesting that they occur as part of the cognitive decision process and are not just the result of movement biases. The strength of sensory encoding, on the other hand, probably reflects trial to trial variations in arousal but does not cause the biases. These results were presented at the 2021 Society for Neuroscience and Neuromatch 4.0 conferences (see video on the project webpage), and will soon be submitted to a journal. Another interesting finding from these experiments was that some participants, on a small but appreciable number of trials, perceived the motion in the exact opposite direction to the true direction of motion. We developed a new decision model that could explain this unexpected behaviour (currently under review in a journal), but the cause of this phenomenon is not yet understood and will be the subject of future research.
Finally, we used EEG motor preparation signals to guide the construction, and constrain key parameters of, a multilevel model of biased decision making. Perceptual decisions are biased toward higher-value options when overall gains can be improved. When decisions are made under time pressure, it is possible to observe the neurophysiological decision process in human EEG dynamically evolving through distinct phases of growing anticipation, detection and discrimination. By parsing motor preparation signals we uncovered a multiphasic pattern of biases evolving over the course of the decision. Before the stimulus appeared, people began preparing for higher-value actions earlier conferring a “starting point” advantage, but then quickly countered with increased preparation of lower-value actions. We used these anticipatory motor preparation signals to constrain motor-level parameters of a decision model which was then able to explain both behaviour and motor preparation dynamics. This work showed that the interplay of distinct biasing mechanisms in time-constrained perceptual decisions is much more complex than can be captured by standard models based on behaviour alone. This paper has been published as a preprint and a science communication video describing the results is available on the project webpage.
Finally, we used EEG motor preparation signals to guide the construction, and constrain key parameters of, a multilevel model of biased decision making. Perceptual decisions are biased toward higher-value options when overall gains can be improved. When decisions are made under time pressure, it is possible to observe the neurophysiological decision process in human EEG dynamically evolving through distinct phases of growing anticipation, detection and discrimination. By parsing motor preparation signals we uncovered a multiphasic pattern of biases evolving over the course of the decision. Before the stimulus appeared, people began preparing for higher-value actions earlier conferring a “starting point” advantage, but then quickly countered with increased preparation of lower-value actions. We used these anticipatory motor preparation signals to constrain motor-level parameters of a decision model which was then able to explain both behaviour and motor preparation dynamics. This work showed that the interplay of distinct biasing mechanisms in time-constrained perceptual decisions is much more complex than can be captured by standard models based on behaviour alone. This paper has been published as a preprint and a science communication video describing the results is available on the project webpage.
While interest in continuous-outcome tasks is increasing in psychology and neuroscience, very little current research accounts for response times as well as precision. Decision-making researchers have long known that being able to explain the joint distributions of response times and accuracy is a key test of a cognitive model. This work has produced convergent neural and behavioural accounts of the cognitive processes involved in decisions with continuous outcomes. In particular, the complex pattern of decision biases that occur in continuous-outcome tasks are a prominent feature of everyday perception that is overlooked by most research restricted to two-alternative decisions.
We now have a well-validated experimental paradigm for simultaneously recording behaviour and EEG in continuous-outcome decision tasks that can be taken forward to probe more complex continuous-outcome decisions in future. This includes a sensory-level EEG signal for the random-dot-motion task, whose relationship with decision making behaviour we have newly characterised.
This work has also produced a new neurally-constrained modelling framework that directly incorporates neural signals and can parse the decision process as it evolves over multiple phases of processing (anticipation, stimulus detection, emerging sensory evidence) at multiple levels in the brain. Such a neurally informed approach provides a more detailed account of decision processes than traditional behavioural modelling, and can be used in future research to gain a more mechanistic understanding of cognitive deficits due to clinical conditions and ageing.
We now have a well-validated experimental paradigm for simultaneously recording behaviour and EEG in continuous-outcome decision tasks that can be taken forward to probe more complex continuous-outcome decisions in future. This includes a sensory-level EEG signal for the random-dot-motion task, whose relationship with decision making behaviour we have newly characterised.
This work has also produced a new neurally-constrained modelling framework that directly incorporates neural signals and can parse the decision process as it evolves over multiple phases of processing (anticipation, stimulus detection, emerging sensory evidence) at multiple levels in the brain. Such a neurally informed approach provides a more detailed account of decision processes than traditional behavioural modelling, and can be used in future research to gain a more mechanistic understanding of cognitive deficits due to clinical conditions and ageing.