Canada

We tackle the problem of an agent interacting with humans in a general-sum environment, i.e., a non-zero sum, non-fully cooperative setting, where the agent&#39;s goal is to increase its own utility.
We show that when data is limited, building an accurate human model is very challenging, and that a reinforcement learning agent, which is based on this data, does not perform well in practice. Therefore, we propose that the agent should try maximizing a linear combination of the human&#39;s utility and its own utility rather than simply trying to maximize only its own utility.


CogSci 2022

Reinforcement Learning Agents for Interacting with Humans

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

AAAI 2026

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

We tackle the problem of an agent interacting with humans in a general-sum environment, i.e., a non-zero sum, non-fully cooperative setting, where the agent's goal is to increase its own utility.
We show that when data is limited, building an accurate human model is very challenging, and that a reinforcement learning agent, which is based on this data, does not perform well in practice. Therefore, we propose that the agent should try maximizing a linear combination of the human's utility and its own utility rather than simply trying to maximize only its own utility.


poster

Join us for CogSci2022! Our 44th Annual Conference will be centered around the theme of diversity, with an accompanying range of keynotes and invited speakers who themselves highlight the diversity of our field. Our three symposia are Cognition across Diverse Populations, Cognition within Diverse Societies, and Global Challenges for Cognitive Science. The conference is being offered in a fully hybrid environment for onsite and online attendees.
 
 [**The Program at a Glance can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/07/Cogsci-2022-PAAG-Detailed-July-12final-1.pdf)
 [**The Detailed Program can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2022/08/DeArmond-CogSci2022-program-v113-final.pdf)

This conference requires a registration to attend. Please use the link below to secure your ticket. The website will be available to all registered attendees on July 25th. 

[Register to Attend](https://cognitivesciencesociety.org/registration/)

Registration Is Required

The 44th Annual Meeting of the Cognitive Science Society highlighted research on the theme Cognitive Diversity in addition to the full breadth and diversity of research topics offered by the society’s membership. 





Previous research indicates that using the internet in knowledge related tasks increases overestimation. We attempted to replicate this finding and extended previous research by explicitly manipulating the standards that participants used for the explanatory knowledge task in order to reduce the metacognitive bias. We conducted a 2x2 within-subject experiment with N = 166 participants. Replicating previous findings, the results show significantly more overestimation in Internet than in No-Internet conditions. However, with an alignment to external standards participants elicited more accurate metacognitive judgments. We conclude that explicit standards may be an important factor in knowledge-related activities involving the internet because of their effect on metacognitive judgments. On a theoretical level, this has implications for determining the basis of overestimation in knowledge tasks with the internet. On a practical level, providing external standards could be a feasible aid for buffering against this bias, for example in the educational context.


An Alignment of Standards Enhances Metacognitive Judgment Accuracy in Explanatory Knowledge Tasks with Internet Search

We present a user error model that simulates a user's errors using an eyes and hands extension to cognitive architectures. We developed a complete interactive cognitive model that performs a spreadsheet task. This model is compared with an existing cognitive model that performs the same task in a different spreadsheet tool. Also, the predictions are compared to human data (N=23) on the same uninstrumented interfaces. The comparison suggests that the interactive cognitive model moves us closer to having a user model that can directly test interfaces by predicting human behavior and performing the task on the same interface that users interact. The error model also allows exploration of error detection, error correction, and different knowledge types.


Including Errors and Errors Correction in a Cognitive User Model

In our study we examined the ability to distinguish between verbal and visual representation in preschool children. We developed tasks in which children had to learn new signs denoting different types of representation and then use the new symbols to compare different formats of imagination (inner speech and visual images) in story characters, and then apply these symbols to describe their own representation in solving different tasks. We will present the results of testing children between the ages of 4 and 9, which show that the majority of preschool and school-age children successfully learn the symbols and use them to distinguish verbal representation from visual representation in characters. We will discuss the further development and application of this technique to the study of preschool children’s meta-cognitive abilities. The reported study was funded by RFBR, project number 20-013-00698.


Understanding of differences in verbal and figurative representations by children 4-6 years old.

There’s increasing evidence from studies of human performance in artificial classification learning tasks that a plurality of mechanisms or strategies are available to learners. A recent investigation from our lab group found that a linearly-separable category structure was harder to learn than a comparable non-linearly separable structure; and furthermore, that there were qualitative individual differences reflecting different paths to learning. In the present experiment, we take a deeper dive into performance on the LS category structure using a bank of test phase measures to more fully reveal what each individual has learned and represented. We identify a systematic set of profiles to characterize individual learners and demonstrate novel evidence on the nature and role of traditional psychological constructs: exemplars, prototypes, and rules. In particular we show classic prototype effects arising not as a broad-based phenomenon but only from a specific path to learning.
Keywords: categorization; tacl: prototype effect; individual differences


Paths to Learning in Traditional Artificial Classification Tasks

The results of research by Hélie, Turner, and Cousineau (Hélie et al., 2018) show that category representation (rule-based and information integration categories) may influence performance in the visual search categorization task. In the present experiment, we replicated their results using other types of categories that contained discrete features: categories based on verbal rules and prototypes. We found that after learning the verbal rule as well as after learning rule-based categories target-present and target-absent trials are not much affected by display size. The effect of display size is bigger for prototypes, but unlike the results of Hélie et al, it does not collapse for target absent trials. We explain these results by the fact that the categories with discrete features have a stronger and more conscious representation, which is easier to transfer to the visual search task. The reported study was funded by RFBR, project number 20-013-00698.


The impact of categorization rule on categorical visual search

Categorization is a critically important aspect of cognition that undergoes development. Previous studies found that children of age four acquired categories through multiple features, forming a similarity-based representation, whereas adults relied on a single feature, forming a rule-based representation. One possible interpretation is that children cannot inhibit the distraction of multiple features. We designed a categorization task with all features covered in the present study. Children were asked to uncover one feature at a time until they could categorize the items. As a result, we minimized distraction and encouraged rule-based learning. Nonetheless, even when given incentives to uncover fewer features, children learned the categories with significantly higher entropy, as calculated by the number of features they uncovered, than adults did. Such a result suggests that inhibition failures cannot fully explain categorization in children. Instead, children’s tendency to over-explore (and thus over-sample features) may also contribute to early category learning.


The role of exploration in the development of category learning

Navigation is a process that humans use to get from A to B. Landmarks used during navigation and wayfinding can address different sensory modalities. We examined landmark information in four different variants: as a written word, as a spoken word, as a picture, or as an odor. Our 51 participants were separated into four groups. Each group received one specific variant of landmark information integrated into a learning and wayfinding video of a virtual maze with 12 intersections. At each intersection, one landmark information was presented. To assess how well the relevant landmarks could be distinguished from unknown distractor items of the same condition, the experiment concluded with a recognition phase, where 24 stimuli were presented (12 landmarks + 12 distractors). Relative frequencies of correct responses and mean response times were measured for wayfinding and recognition. Odors lead to similar correctness in wayfinding compared to the more common landmarks (pictures, written and spoken words), even though requiring longer response times. We stepped away from the traditional but limited view on landmarks towards a more holistic (i.e. including all senses) view of human orientation. Implications for future scientific research are being discussed.

Landmark Modality in Wayfinding: Does it Make a Difference?

Previous research has shown that proportions, ornateness, and lighting of indoor architectural spaces affect observers' mental states (Negami & Ellard 2021). However, most studies are limited to verbal self-reports, and focus either on photographs, or CAD-modeled rooms in which certain elements, (e.g. ceiling height), are experimentally manipulated. 
We study human psychological and physiological responses to historical indoor sites reconstructed in virtual reality (VR). Such spaces are often designed to evoke affective responses -- for example, sacred architecture is meant to evoke feelings of calm. Using drone footage, we record 3D geometry, visual and auditory sensory data of an indoor space.
We recreate the space in VR, and record humans’ eye-movements, heart-rate, galvanic skin response, and reports of affective states, during free exploration. We propose a cognitive model that interprets physiological responses as information-foraging, and identify the correlates of reported changes in affective states with specific properties of architectural space.


Human Information Seeking in Architectural Spaces Simulated in Virtual Reality

The “Direction and Orientation Strategy Questionnaire” (DOSQ) is a novel assessment based on giving directions around your hometown and how one remains oriented in real-world environments. The DOSQ produces three sub-scale scores: use of allocentric strategies, use of egocentric strategies, and sense-of-direction. Through an online survey, over one hundred Texas A&M undergraduates completed demographic questions, DOSQ, Santa Barbara Sense-of-Direction scale (Hegarty et al., 2002), Wayfinding Strategy Scale (Lawton, 1994), Questionnaire on Survey Representation (Pazzaglia & De Beni, 2006), Spatial Anxiety Scale (Lawton, 1994), and familiarity measures using the Texas A&M campus. Using factor analysis, we found that the DOSQ sub-scores were highly related to other measures of strategy use and sense-of-direction, establishing its convergent validity. DOSQ sub-scales were not related to the spatial anxiety or familiarity measures, establishing its discriminant validity. The DOSQ provides a method for briefly assessing spatial thinking strategies that support navigation through real-world environments.

Convergent and Discriminant Validity of the Direction and Orientation Strategy Questionnaire

From previous work we know that mechanical reasoning is associated with spatial skills, and that it could even be possible for mechanical reasoning training to lead to meaningful gains in spatial abilities. However, it is unknown whether the connection with mechanical reasoning applies to different types of spatial skills or is specific to only one or a subset of them. To answer this question, we presented participants with six different tasks, four of which corresponded to spatial assessments that were organized along a 2 (intrinsic/extrinsic) x 2 (static/dynamic) matrix of factors, and two evaluations that measured mechanical reasoning (gears and pulleys task and DAT-5 mechanical reasoning test). Three hundred adult participants from Chile responded from their personal computers to the stimuli presented on Open Lab, an online platform. The results shed light on the link between mechanical reasoning and the different subtypes of spatial skills.


Premium content

Downloads

Next from CogSci 2022

An Alignment of Standards Enhances Metacognitive Judgment Accuracy in Explanatory Knowledge Tasks with Internet Search

Similar lecture

Achieving Equilibrium Under Utility Heterogeneity: An Agent-Attention Framework for Multi-Agent Multi-Objective Reinforcement Learning