Shawon Sarkar

This study examines generative AI (GenAI) adoption among K-12 mathematics and science teachers using data from the RAND American Educator Panel (n ≈ 1,000). We investigate: (1) the extent of GenAI use for instructional tasks, (2) differences between mathematics and science teachers, and (3) barriers to adoption. Results show that science teachers report higher GenAI use, particularly for instruction. The most common applications are lesson planning (38%) and generating student assignments (31%), while few teachers (7%) use GenAI for assessment. Differences between subject areas are smaller for administrative tasks and professional learning. Overall adoption remains below 50%. Key barriers include insufficient training, time required to learn tools, and unclear district policies, whereas concerns about output quality and restrictive policies are less frequently reported. These findings highlight subject-specific adoption patterns and emphasize the need for clearer guidance and targeted professional development to support effective GenAI integration in K-12 education.

As generative AI reshapes education, the AmplifyGAIN R&D Center leads national efforts to build AI readiness across the K-12 sector. Funded by the Institute of Education Sciences, the center supports educators, edtech professionals, and policymakers through professional development, policy engagement, and open dissemination. To date, AmplifyGAIN has trained over 1,000 participants via webinars, professional learning communities, and partnerships with state agencies, focusing on instructional integration, personalization, and assessment. Central to its approach is capacity building, with educators serving as co-designers through the COALESCE framework. The center contributes to national discourse through conference presentations and research dissemination, while supporting state-level policy development in collaboration with education agencies. Its Innovation Science for Education Analytics (ISEA) program addresses the AI talent gap by training interdisciplinary fellows in computational methods and ethics. AmplifyGAIN further promotes equitable access through open platforms, resources, and public repositories. Together, these efforts position the center as a leader in advancing responsible, scalable AI integration in K-12 education.

Artificial intelligence is reaching science classrooms, but how teachers use it remains unclear. We report results from a national survey of 566 U.S. public-school science teachers (RAND American Educator Panel, 2025). We defined generative AI as tools that create content or automate tasks using large data patterns. About half of teachers reported any use of GenAI; roughly one in four use it weekly or daily. Use concentrates on teacher-facing work—planning lessons (40%) and creating assignments or assessments (32%)—with smaller shares for in-class support, differentiation, and routine paperwork. Only 5% said they teach students to use GenAI as a learning tool. About a quarter said GenAI has changed how they interact with students, most often through more personalization, better support for varied needs, and more ways to explain difficult ideas. Teachers also reported limited student use overall. Taken together, GenAI currently functions more as a planning assistant than a classroom learning tool. We close by outlining implications for professional learning, classroom-ready resources, and policy.

The COALESCE framework extends DBIR principles to AI-powered educational technology development, positioning educators as active co-creators. This study evaluates COALESCE through its application in designing an AI-assisted lesson-planning tool. Our findings show how deep educator involvement supports trust, ownership, and domain-specific improvements in AI-generated content. We discuss implications for participatory AI design and offer a replicable model for context-aware, educator-aligned innovation.

As generative AI (GenAI) models are increasingly explored for educational applications, understanding educator preferences for AI-generated lesson plans is critical for their effective integration into K-12 instruction. This exploratory study compares lesson plans authored by human curriculum designers, a fine-tuned LLaMA-2-13b model trained on K-12 content, and a customized GPT-4 model to evaluate their pedagogical quality across multiple instructional measures: warm-up activities, main tasks, cool-down activities, and overall quality. Using a large-scale preference study with K-12 math educators, we examine how preferences vary across grade levels and instructional components. We employ both qualitative and quantitative analyses. The raw preference results indicate that human-authored lesson plans are generally favored, particularly for elementary education, where educators emphasize student engagement, scaffolding, and collaborative learning. However, AI-generated models demonstrate increasing competitiveness in cool-down tasks and structured learning activities, particularly in high school settings. Beyond quantitative results, we conduct thematic analysis using LDA and manual coding to identify key factors influencing educator preferences. Educators value human-authored plans for their nuanced differentiation, real-world contextualization, and student discourse facilitation. Meanwhile, AI-generated lesson plans are often praised for their structure and adaptability for specific instructional tasks. Findings suggest a human-AI collaborative approach to lesson planning, where GenAI can serve as an assistive tool rather than a replacement for educator expertise in lesson planning. This study contributes to the growing discourse on responsible AI integration in education, highlighting both opportunities and challenges in leveraging GenAI for curriculum development.

This study introduces a practice-informed AI nudging system designed to support responsible, context-aware instructional decision-making during lesson planning. Grounded in a hybrid intelligence framework, the system integrates multiple sources of human insight: individual teacher preferences, collective behavior patterns, and pedagogical research, with computational support from large language models to generate timely and explainable nudges. A co-occurrence network of instructional strategies, derived from annotated teacher-AI interactions, guides contextual nudge generation using behavioral heuristics. Initial analysis shows strong alignment between system recommendations and teachers’ pedagogical goals and values, as well as increased diversity in effective instructional practices. By balancing professional autonomy with evidence-based guidance, this work advances responsible AI in education, contributing an explainable, value-sensitive system that fosters sustainable pedagogical improvement in real-world settings.

A recommender system is tasked with effectively analyzing a user’s preferences and interactions to provide personalized recommendations. This calls for extracting and connecting various heterogeneous data while preserving their temporal relations. Graph neural networks (GNNs) have proven to be highly suitable in recommendation systems for connecting different types of user behavioral signals. However, they inherently lack ability to capture temporal aspects of underlying data. This shortcoming prevents them from explicating and utilizing task information, which is shown to be instrumental in many information retrieval applications. To overcome this limitation, we propose a new Task-based Graph Neural Network model (TGNN) focusing on identifying users’ underlying tasks within their temporal multi-behavior, specifically in each session. The model consists of three modules: (1) a sequential meta-path module that captures a temporal sequence of users’ behaviors; (2) a graph neural network layer that models the relationships between different information items and users into task representations; and (3) a recommendation layer that utilizes a collaborative filtering method to generate top-N recommendations based on the model’s comprehension of users’ tasks. The novelty of our approach lies in understanding users’ tasks through their temporal behavior, enabling more accurate personalization. The results of evaluative experiments on three publicly available datasets demonstrate the effectiveness of our task-based recommendation model compared to 10 baselines and indicate a promising research direction for task-oriented recommender systems.

Despite the considerable advancements in modern search systems for assisting users in search tasks of varying types, support for complex tasks that call for multi-round interactions remains challenging. Identifying users’ tasks is essential to understanding their evolving information needs and search goals during search sessions to simulate and achieve real-time adaptive search retrievals; thus, it is a crucial research thrust in interactive information retrieval (IIR). While a series of descriptive and formal models have been proposed to characterize complex information search sessions, only a few focus on leveraging dynamic task features in search personalizations to support users in different task stages in an adaptive fashion. This preliminary study presents a heterogeneous graph neural network model for extracting and representing tasks to better understand users’ interactive search processes by connecting tasks with search interactions. Our approach’s novelty lies in our application of task representation learning, which enables systems to extract hidden task information from users’ search behaviors. The results of our evaluative experiments on TREC Session track data highlight the value of our proposed task representation model and illustrate a promising research direction on task-oriented intelligent systems.

The importance of tasks in information retrieval (IR) has been long argued for, addressed in different ways, often ignored, and frequently revisited. For decades, scholars made a case for the role that a user’s task plays in how and why that user engages in search and what a search system should do to assist. But for the most part, the IR community has been too focused on query processing and assuming a search task to be a collection of user queries, often ignoring if or how such an assumption addresses the users accomplishing their tasks. With emerging areas of conversational agents and proactive IR, understanding and addressing users’ tasks has become more important than ever before. In this paper, we provide various perspectives on where the state-of-the-art is with regard to tasks in IR, what are some of the bottlenecks in deriving and using task information, and how do we go forward from here. In addition to covering relevant literature, the paper provides a synthesis of historical and current perspectives on understanding, extracting, and addressing task-focused search. To ground ongoing and future research in this area, we present a new framing device for tasks using a tree-like structure and various moves on that structure that allow different interpretations and applications. Presented as a combination of synthesis of ideas and past works, proposals for future research, and our perspectives on technical, social, and ethical considerations, this paper is meant to help revitalize the interest and future work in task-based IR.

For users working on a complex search task, it is common to address different goals at various stages of the task through query iterations. While addressing these goals, users go through different task states as well. Understanding these task states latent under users’ interactions is crucial in identifying users’ changing intents and search behaviors to simulate and achieve real-time adaptive search recommendations and retrievals. However, the availability of sizeable real-world web search logs is scarce due to various ethical and privacy concerns, thus often challenging to develop generalizable task-aware computation models. Furthermore, session logs with task state labels are rarer. For many researchers who lack the resources to directly and at scale collect data from users and conduct a time-consuming data annotation process, this becomes a considerable bottleneck to furthering their research. Synthetic search sessions have the potential to address this gap. This paper shares a parsimonious model to simulate synthetic web search sessions with task state information, which interactive information retrieval (IIR) and search personalization studies could utilize to develop and evaluate task-based search and retrieval systems.

Colonialism has had direct impacts on the transmission of American Indian and Alaska Native (AIAN) knowledge systems. Spaces of higher education continue to create challenges for Indigenous students, especially through the disconnection between Native knowledges and the dominant knowledge being taught. These challenges are reflected in the educational attainment disparities between AIAN and non-Indigenous students in the United States and Canada. To address these issues, we emphasize the significance of promoting ethical approaches to Indigenous research and embracing Indigenous ways of knowing. We examine how combining a directed research group (DRG) with a Knowledge Family approach can shift Indigenous experiences with higher education knowledge production. We describe the structure and goals of the DRG Knowledge Families program, which provides support and resources for Indigenous students while fostering meaningful relationships between Indigenous and non-Indigenous researchers. The DRG Knowledge Families approach integrates research opportunities for Indigenous students, focusing on Indigenous knowledge systems and research methodologies. Ultimately this approach aims to create research spaces that value Native knowledge, center community needs, and support the success of Indigenous students.

The varieties of information seeking behavior encompass a range of practices and constructs such as the realization of an information need, selecting the nature of information, as well as information sources. Most of the past works have studied various constructs of the information seeking process, i.e., information, information need, and information sources individually. However, a person forms and re-forms his or her information seeking strategy based on continually shifting values of these dimensions associated with information seeking. This preliminary study conducted a survey with 15 search scenarios and multiple-choice characteristics completed by 114 Amazon’s Mechanical Turk workers to find out more about how these constructs play a role in people’s preferences regarding information seeking strategies. The study took an exploratory and inferential research approach to investigate how different forms of information and information needs might lead to different information sources by building binary classification models. The results show that the choice of sources can be predicted (with 80% accuracy) if the information need, representation, and form of information are apparent.

In this exploratory study, we examine how American Indian and Alaska Native (AIAN) governments and organizations are using social media to share critical health information about coronavirus disease 2019 (COVID-19) with their citizens. Through a thematic analysis of 119 public Facebook posts made by Tribal governments and organizations, we identified three broad categories and 13 subthemes. Tribal governments and organizations created risk communication material for their respective communities that fell under (1) risk reduction,(2) meeting community members’ needs, and (3) staying connected to community and culture. Our findings suggest that through social media AIAN communities and organizations played a crucial role in disseminating reliable culturally adapted risk communication and vital community information to Tribal citizens during the COVID-19 pandemic. Such communication included clear illustrations, posts and messages about the importance of masking up, social distancing and washing one’s hands; mandated border closures; and suggestions for maintaining a sense of connectedness with community. By doing so they are filling a gap that ensures their communities receive the relevant information they need to mitigate and manage risks. In order to understand how to better meet community needs, more work is needed to improve the wellbeing and visibility of AIAN people in the areas of health disparities, technology, social media, and the many impacts of COVID-19.

Complex search tasks that involve uncertain solution space and multi-round search iterations are integral to everyday life and information-intensive workplace practices, affecting how people learn, work, and resolve problematic situations. However, current search systems still face plenty of challenges when applied in supporting users engaging in complex search tasks. To address this issue, we seek to explore the dynamic nature of complex search tasks from process-oriented perspective by identifying and predicting implicit task states. Specifically, based upon the Web search logs and user annotation data (regarding information seeking intentions in local search steps, in-situ search problems, and help needed) collected from 132 search sessions in two controlled lab studies, we developed two task state frameworks based on intention state and problem-help state respectively and examined the connection between task states and search behaviors. We report that (1) complex search tasks of different types can be deconstructed and disambiguated based on the associated nonlinear state transition patterns; and (2) the identified task states that cover multiple subtle factors of user cognition can be predicted from search behavioral signals using supervised learning algorithms. This study reveals the way in which complex search tasks are unfolded and manifested in users’ search interactions and paves the way for developing state-aware adaptive search supports and system evaluation frameworks.

Information need is one of the most fundamental aspects of information seeking, which traditionally conceptualizes as the initiation phase of an individual’s information seeking behavior. However, the very elusive and inexpressible nature of information need makes it hard to elicit from the information seeker or to extract through an automated process. One approach to understanding how a person realizes and expresses information need is to observe their seeking behaviors, to engage processes with information retrieval systems, and to focus on situated performative actions. Using Dervin’s Sense-Making theory and conceptualization of information need based on existing studies, the work reported here tries to understand and explore the concept of information need from a fresh methodological perspective by examining users’ perceived barriers and desired helps in different stages of information search episodes through the analyses of various implicit and explicit user search behaviors. In a controlled lab study, each participant performed three simulated online information search tasks. Participants’ implicit behaviors were collected through search logs, and explicit feedback was elicited through pre-task and post-task questionnaires. A total of 208 query segments were logged, along with users’ annotations on perceived problems and help. Data collected from the study was analyzed by applying both quantitative and qualitative methods. The findings identified several behaviors – such as the number of bookmarks, query length, number of the unique queries, time spent on search results observed in the previous segment, the current segment, and throughout the session – strongly associated with participants’ perceived barriers and help needed. The findings also showed that it is possible to build accurate predictive models to infer perceived problems of articulation of queries, useless and irrelevant information, and unavailability of information from users’ previous segment, current segment, and whole session behaviors. The findings also demonstrated that by combining perceived problem(s) and search behavioral features, it was possible to infer users’ needed help(s) in search with a certain level of accuracy (78%).