Session 7: Friday 12 February, 9am-10am
Prof. Ben Newell
Decision aids are increasingly integrated into everyday choices. For example, Netflix might keep the binge rolling by automatically recommending another show. How the recommendation is acquired, automatically or actively sought out, can lead to different ways of using that information to make a choice. We present two experiments involving a decision aid in a dot motion task. Participants were told an algorithm would provide recommendations that were correct 70% of the time. This accuracy bisected performance for easier difficulty trials (~ 95%) and harder difficulty trials (~ 55%). In Experiment 1, participants could choose to seek out a recommendation. We manipulated the accuracy of the algorithm (70% vs. 80%) and found higher algorithm accuracy led to greater recommendation seeking for the easier trials when it was seemingly unnecessary. In Experiment 2, the recommendation automatically loaded after a period of time (1.8 seconds vs. 2.8 seconds). Reducing the time cost led individuals to examine but disagree with the recommendation more often. RT data identifies at least two distinct groups; one subset quickly agrees with the 70% recommendation akin to a better-than-chance guess, while another group effortfully tries to integrate the recommendation with the stimulus.
Dr. Kohinoor Darda
Dr. Richard Ramsey
Many well-established cognitive paradigms such as the Stroop task and the Posner cueing task show robust effects at a group-level but low test-retest reliability at an individual within-participants level. This “reliability paradox” (Hedge et al. 2017) has limited researchers’ ability to study individual differences and draw inferences about cognitive processes from such behavioural measures. One solution to this problem is to apply a generative modelling approach which takes into account both individual and group-level uncertainty (Haines et al., 2020). Unlike traditional summary statistics approaches, which use only point-estimates, multi-level models allow for a more precise interpretation of statistical effects by including the full distribution of participant responses at an inter-trial and inter-item level. In the present study, we applied this multi-level modelling approach to existing data collected from three experiments (> 600 participants in total) investigating individual differences in an automatic imitation task (Darda et al., 2020). Data analysis is still in-progress; however, we aim to determine whether a multi-level modelling approach, as opposed to a summary statistics approach, will yield any meaningful differences in the results. We are particularly interested in examining changes in individual sex and personality differences as a function of the spatial and imitative compatibility effect.
Dr. Ami Eidels
In human-computer interaction, adaptive user interfaces are designed to adjust themselves to suit the needs of users. When task demands negatively affect user performance or workload, adaptive systems can change their characteristics to moderate their impact on users. However, an adaptive system is only as good as its estimate of the user’s current state. Measures of cognitive workload have been used within adaptive user interfaces, but few objective workload measures exist that can be embedded within an interface. The Detection Response Task (DRT) is one such measure. In two experiments, a mouse-tracking task was developed with ordinal difficulty levels, and adaptive systems of adjusting difficulty based on DRT performance and tracking task performance were evaluated. The difficulty of the tracking task was also evaluated using an algorithm that simulated random responses, which identified a confound in Experiment 1’s difficulty settings. When data from the DRT was used to adapt tracking task difficulty, participants performed better than when difficulty changed randomly, though this system was no more successful than a system driven by tracking task performance data. However, in certain scenarios, main task performance may not be available, or could be too costly to collect. In such cases the DRT could be a promising measure as part of more complex adaptive systems due to its simplicity, objectivity, and applicability in real-world settings.
Dr. Avinash Singh
Dr. Tien-Thong Do
Prof. Chin-Teng Lin
Recently, mental workload adaptive systems have received great attention with the employment of physiological measures of mental workload to trigger the adaptive automation. However, currently, physiological correlates of the mental workload are only used to decide “when” to adapt and not “what” to adapt, keeping the strategies employed by the adaptive automation system still primitive. In this study, we study the workload experienced by air traffic controllers (ATC). Even though several factors influence the complexity of ATC tasks, keeping track of the aircraft and preventing collision is the most crucial. We have designed tracking and collision prediction tasks to elucidate the differences in the physiological response to the workload variations in these basic ATC tasks to untangle the impact of workload variations experienced by operators working in a complex ATC environment. Electroencephalogram (EEG) data was recorded from 24 participants performing tracking and collision prediction tasks with three levels of difficulty. Our findings indicate markedly distinct neurometrics of workload variations in the tracking and collision prediction tasks. Our study provides compelling evidence in the viability of developing intelligent closed-loop mental workload adaptive systems that ensure efficiency and safety in ATC and beyond.
Dr. Dora Matzke
Prof. Denny Borsboom
I describe a discrete choice approach to microscopic agent modelling pedestrian behaviour. Crossed nested logit models are used by agents to make utility maximising choices in a dynamic and individual-based discretization of speed and direction. Each agent has a first-order theory-of-mind, basing their choices on predictions about the behaviour of other agent in the next time step. In contrast to most existing pedestrian models I introduce individual differences in agent parameters and present simulations of how groups of pedestrians interact in a range of scenarios requiring them to achieve a set of movement goals. If time allows I will present preliminary results on estimating agent parameters.
Faces are considered a special class of holistically-processed object. The composite face task is a widely-used paradigm for inferring holistic processing. In this task, recognition of one half of a composite face is shown to be hampered by interference from the other half of the face when faces are aligned but not when misaligned. Although this effect has been documented numerous times, when used in different paradigms, composite faces do not always exhibit effects consistent with holism. The present study explored the cause of these discrepant findings by combining a composite face task with a signal-to-respond paradigm. The amount of time to make a face recognition decision was manipulated by introducing a response signal, and the resulting changes in accuracy were mapped over the time course of processing, which was then used to fit a speed-accuracy trade-off model. We found that holistic processing emerges late in the time course after approximately 400 ms processing time for easy to discriminate faces and after approximately 1000 ms for difficult to discriminate faces.
Dr. Emily Freeman
Over the last few decades, the social movement of involved fatherhood has stimulated a research focus on fathers, leading to an increase in the body of evidence into the paternal contributions to child development. Past research has suggested that rough-and-tumble play (RTP), which involves wrestling, chasing and tumbling, is the preferred play type of western fathers. While this play has been perceived as being dangerous or too aggressive, the limited research available has shown a relationship between high quality rough-and-tumble play interactions, and both lower childhood aggression and improved child emotional regulation. However, a cognitive approach has not been explored. Thus, the aim of this study was to examine father-child RTP interactions with children aged 4-7 years and explore the impacts on child development and cognition. Analyses revealed that father-child RTP play quality was positively related to working memory outcomes in children. Furthermore, the amount of rough-and-tumble play father and child did together on a regular basis was also related to working memory outcomes. While father-child play interactions remain an understudied area of research, this study outlines the importance of examining the paternal play role in children’s cognitive development.