United States

The best-performing artificial intelligence systems for object recognition are deep neural networks (DNNs). Engineers and neuroscientists have claimed that DNNs show human-level performance in object recognition. However, Barbu et al. (2019) reported accuracies of around 30% for state-of-the-art object recognition systems when testing on their better-controlled image set – ObjectNet. How do humans perform on ObjectNet? In our experiment, we tested 25 undergraduates on their ability to classify the ten categories of objects in ObjectNet that the Deep Convolutional Neural Networks (DCNNs) found easiest, moderate and hardest. Although humans and DCNNs had similar levels of overall accuracy, there were some everyday, basic-level categories for which machine performance was much lower than humans. The pattern of errors generated by the DCNNs were about as similar to human error patterns as the individual human error patterns were to each other. Implications of these results and plans for future work are discussed.

CogSci 2023

A Comparison of Human and Machine Performance in Object Recognition Using The ObjectNet Image Set

poster

The 45th Annual Meeting of the Cognitive Science Society will take place in Sydney, Australia as a modified hybrid conference. Presentations from the main-session room will be live-streamed every day for all events. This will include all invited talks and award sessions. Recordings of all CogSci 2023 sessions will be made available for on-demand viewing after the conference.<br>
The in-person component will be held at the International Convention Centre Sydney at Darling Harbour. The Conference will take place Wednesday July 26th – Saturday July 29th, 2023 (Wednesday, July 26th is the Workshops Day).

Join us!

<br>[**The Program at a Glance can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2023/07/COgSci-PAAG-V2.pdf)
<br>[**The Detailed Program can be found here**](https://online.fliphtml5.com/ebtyf/mswz/)
<br><br>

#### **CogSci 2023 Program Book**
<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/tzgk/"  seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>
<br>

The 45th Annual Meeting of the Cognitive Science Society presents the latest research across cognitive science and highlights the theme of Cognition in Context.

Neural network models of language have long been used as a tool for developing hypotheses about conceptual representation in the mind and brain. For many years, such use involved extracting vector-space representations of words and using distances among these to predict or understand human behavior in various semantic tasks. In contemporary language AIs, however, it is possible to interrogate the latent structure of conceptual representations using methods nearly identical to those commonly used with human participants. The current work uses two common techniques borrowed from cognitive
psychology to estimate and compare lexical-semantic structure in both humans and a well-known AI, the DaVinci variant of GPT-3. In humans, we show that conceptual structure is robust to differences in culture, language, and method of estimation. Structures estimated from AI behavior, while individually fairly consistent with those estimated from human behavior, depend much more upon the particular task used to generate behavior responses–responses generated by the very same model in the two tasks yield estimates of conceptual structure that cohere less with one another than do human structure estimates. The results suggest one important way that knowledge inhering in contemporary AIs can differ from human cognition.

Behavioral estimates of conceptual structure are robust across tasks in humans but not large language models

We investigated the potential and limitations of generative artificial intelligence (AI) in reflecting the authors' cognitive processes through creative expression. The focus is on the AI-generated artwork's ability to understand human intent (alignment) and visually represent emotions based on criteria such as creativity, aesthetic, novelty, amusement, and depth. Results show a preference for images based on the descriptions of the authors' emotions over the main events. We also found that images that overrepresent specific elements or stereotypes negatively impact AI alignment. Our findings suggest that AI could facilitate creativity and the self-expression of emotions. Our research framework with generative AIs can help design AI-based interventions in related fields (e.g., mental health education, therapy, and counseling).

A Portrait of Emotion: Empowering Self-Expression through AI-Generated Art

Categorical learning plays a foundational role in language development. By reviewing comparative studies on categorical learning in humans and nonhuman animals, we show that categorical learning displays evolutionary continuity across invertebrates and vertebrates. Great apes and parrots can be trained to produce categories of (proto-)language-like symbols in different modalities. From the neurological perspective, we show that as a conserved brain structure, the basal ganglia are involved in categorical learning across species, and language processing in humans. This raises the possibility that categorical learning is one of the crucial cognitive foundations for language evolution.

Categorical Learning and the Cognitive Foundations of Language Evolution and Development

Alcohol binge drinking is ubiquitous among young adults. While the neurotoxic effects of general alcohol consumption are well known, binge drinking has unique effects on the brain through the intoxication and withdrawal cycle. In particular, the research regarding binge drinking and attention is mixed; this study investigates this relationship in young adults. University students (N=105) were recruited. Socio-demographic and alcohol use information was collected, then participants completed cognitive tasks based on the Attention Network Theory: alerting, orienting, and executive control. Binge drinking, sex, and age of first binge did not predict attention impairment, nor did sex moderate the relationship. The tasks measuring attention did not relate in the expected manner. While there were no differences in attention performance between those who binge drink and controls in this study, the relationship between binge drinking and attention impairments in young adults may be more nuanced and future research directions are suggested.

The Effects of Binge Drinking on Attention in Young Adults.

Traditionally considered random noise, variability in behaviour in fact contains meaningful structure. Healthy functioning can be characterised by behavioural variability that is statistically self-similar (i.e., exhibits 1/f scaling), while disease or injury can result in deviations from this pattern. To date this work has focused predominantly on physical conditions. Little is known about the link between mental health and 1/f scaling. To address this gap, in the present study participants (n = 61) completed a simple walking task during which their gait variability was captured. Cognitive load, a factor known to impact 1/f scaling, was manipulated and symptoms of social anxiety assessed. As expected, being distracted by a cognitive task led to deviations in gait structure away from 1/f scaling. Notably, this effect was amplified by symptoms of social anxiety. This provides initial evidence that difference in mental health status impacts the structure of behaviour.

Exploring the impact of social anxiety on behavioural variability

The current study investigates whether a running man's pictogram on emergency signage is as effective as an arrow in conveying direction, especially in emergencies. We conducted two experiments to address the following questions: first, comparing the ease of reading the two signages, and second, presenting an incomplete image of a running man (RM) signage design pictogram to investigate important components of running man that contribute in conveying direction. In both conditions, the participant's task is to identify the orientation/direction and choose the keys corresponding to the right/left orientation/direction.

Investigating design of running man in emergency signage

This paper presents a theoretical description of metacognition in architectural design. The assumption is that metacognition happens when an architect expresses self-awareness and that’s likely to happen when he/she acquires a global view of the design process. Self-awareness does not necessarily mean high confidence rating, so this needs to be measured separately. During design, architects are immersed in a sequential decision-making process with limited links/references backwards and forwards to immediately adjacent design actions. At certain points during design, possibly evoked by a stimuli or a challenge, architects rise to a higher order awareness where they construct a global view of parts or the whole of the design process, and understand, in parts or in full, the structure and causalities underlying their design decisions within the context of the design process and perhaps within a wider context that is situated in the social world. This is where architects are likely to declare or express their level of confidence about the actions they have made thus far and set longer term strategies on how they are going to manage their design decisions.

Metacognition in architectural design

There is debate about when children can think counterfactually about doubly-determined events (McCormack et al., 2018; Nyhout, Henke, & Ganea, 2017). We tested whether counterfactual thinking about past events, requiring mental time travel, was harder than thinking about ongoing events. 68 3- to 6-year-olds saw a box which lit up when activator blocks were placed on it. In doubly-determined current trials, a red and a blue block (both activators) were placed on the box which lit up. Children were asked “If we had not put the red block on the machine, would the light still be on now?” (Correct “yes”). In doubly-determined past trials, blocks were removed before questioning. 
Five- and six-year-olds found both versions easy. Three- to four-year-olds performed poorly. Four- to five-year-olds performed better than chance on current trials (p=.001), but at chance on past trials (p>.999). We relate this to Hoerl and McCormack’s (2018) dual systems perspective.

Counterfactuals and mental time travel

There is an ongoing debate in the scientific community regarding the nature of cognitive flexibility. It is either seen as a general ability, a task-dependent skill, or a dimensional construct composed of reactive flexibility and spontaneous flexibility. To develop the discussion, we compared first-graders’ performance on four cognitive flexibility tasks. Two tasks focused on reactive flexibility (rule-switching and verbally-cued-switching), and two targeted spontaneous flexibility (role-shifting and divergent thinking). We also used three tests of executive functions (cognitive inhibition, working memory, and processing speed) to evaluate their link with cognitive flexibility. Significant positive correlations emerged between all but one of the flexibility tasks, regardless of whether they targeted a reactive or spontaneous use of flexibility. Similarly, confirmatory factorial analysis did not provide evidence for the dimensional structure of cognitive flexibility. These results suggest that different flexibility tasks may tap into distinct aspects of a task-dependent ability, rather than a generalized competence.

Cognitive Flexibility Or Flexibilities? New Insights From A Classroom Study

The approximate number system, involved in the estimation of large numerosities, forms the basis for the development of mathematics ability. Although researchers have investigated the cognitive underpinnings of the approximate number system, the specific underlying cognitive abilities involved need better theorisation. There have also been conflicting findings regarding the relationship between the approximate number system and mathematics ability. Therefore, the current study had twofold objectives: 1) To investigate the cognitive underpinning of the approximate number system. 2) To test whether the approximate number system can predict mathematics ability. A total of 31 primary school children participated in the study. To measure approximate number system and mathematics ability, non-symbolic number comparison and arithmetic fluency tasks were used, respectively. Using repeated measure ANCOVA and multilinear regressions, results revealed that only inhibitory control played a significant role in the approximate number system. Furthermore, the approximate number system could not predict arithmetic fluency.

Premium content

A Comparison of Human and Machine Performance in Object Recognition Using The ObjectNet Image Set

Next from CogSci 2023

Behavioral estimates of conceptual structure are robust across tasks in humans but not large language models