Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Extending Interactive Science Exhibits into the Classroom using Anthropomorphized Chatbots and Bloom's Taxonomy

Yousuf Golding

TL;DR

The paper investigates extending public science exhibits into classrooms through anthropomorphized Generative AI chatbots, aiming to improve accessibility for marginalized learners. It evaluates feasibility using prompt engineering with GPT-3.5 and explores Bloom's Taxonomy-driven question generation to support assessment. The study demonstrates basic feasibility of combining persona-based dialogue and taxonomy-aligned questions, and discusses design considerations, limitations, and pathways for future field evaluations and inclusive classroom implementations. The work lays groundwork for future in-field assessments of virtual exhibits and highlights the potential for GenAI to broaden science education access, including for historically underrepresented groups.

Abstract

This study explores the use of Generative AI chatbots for transforming public science exhibits into virtual experiences that can extend the engagement of exhibits into the classroom. The broader goal is to increase accessibility of science exhibits, especially for those marginalized in STEM due to various factors, including cultural barriers. We hypothesize that turning exhibits into first-person anthropomorphized chatbots with a personality, like quirky-talking asteroids or comets, can increase engagement and learning. The paper mainly explores if such techniques are possible using Generative AI (e.g. GPT) via prompt engineering alone. The research includes an investigation into the possibility of integrating interactive assessment via question-generation using Bloom's Taxonomy. Initial results indicate that it is possible to combine these techniques. As such, it lays a foundation for future classroom evaluations of such chatbots to gauge their overall efficacy in extending the reach of science exhibitions. The paper concludes by discussing extensions of the research to fully evaluate effectiveness in virtual field-trips. We also include a brief examination of additional ways to enhance student motivation towards learning via chatbots.

Extending Interactive Science Exhibits into the Classroom using Anthropomorphized Chatbots and Bloom's Taxonomy

TL;DR

The paper investigates extending public science exhibits into classrooms through anthropomorphized Generative AI chatbots, aiming to improve accessibility for marginalized learners. It evaluates feasibility using prompt engineering with GPT-3.5 and explores Bloom's Taxonomy-driven question generation to support assessment. The study demonstrates basic feasibility of combining persona-based dialogue and taxonomy-aligned questions, and discusses design considerations, limitations, and pathways for future field evaluations and inclusive classroom implementations. The work lays groundwork for future in-field assessments of virtual exhibits and highlights the potential for GenAI to broaden science education access, including for historically underrepresented groups.

Abstract

This study explores the use of Generative AI chatbots for transforming public science exhibits into virtual experiences that can extend the engagement of exhibits into the classroom. The broader goal is to increase accessibility of science exhibits, especially for those marginalized in STEM due to various factors, including cultural barriers. We hypothesize that turning exhibits into first-person anthropomorphized chatbots with a personality, like quirky-talking asteroids or comets, can increase engagement and learning. The paper mainly explores if such techniques are possible using Generative AI (e.g. GPT) via prompt engineering alone. The research includes an investigation into the possibility of integrating interactive assessment via question-generation using Bloom's Taxonomy. Initial results indicate that it is possible to combine these techniques. As such, it lays a foundation for future classroom evaluations of such chatbots to gauge their overall efficacy in extending the reach of science exhibitions. The paper concludes by discussing extensions of the research to fully evaluate effectiveness in virtual field-trips. We also include a brief examination of additional ways to enhance student motivation towards learning via chatbots.
Paper Structure (86 sections, 4 figures)

This paper contains 86 sections, 4 figures.

Table of Contents

  1. This study explores the use of Generative AI chatbots for transforming public science exhibits into virtual experiences that can extend the engagement of exhibits into the classroom. The broader goal is to increase accessibility of science exhibits, especially for those marginalized in STEM due to various factors, including cultural barriers. We hypothesize that turning exhibits into first-person anthropomorphized chatbots with a personality, like quirky-talking asteroids or comets, can increase engagement and learning. The paper mainly explores if such techniques are possible using Generative AI (e.g. GPT) via prompt engineering alone. The research includes an investigation into the possibility of integrating interactive assessment via question-generation using Bloom’s Taxonomy. Initial results indicate that it is possible to combine these techniques. As such, it lays a foundation for future classroom evaluations of such chatbots to gauge their overall efficacy in extending the reach of science exhibitions. The paper concludes by discussing extensions of the research to fully evaluate effectiveness in virtual field-trips. We also include a brief examination of additional ways to enhance student motivation towards learning via chatbots.
  2. High school and middle school students benefit from learning about science via hands-on exhibits and demonstrations in public-science establishments such as museums and centers like the https://chabotspace.org/ (CSSC), Oakland. Field trips enable students to develop interest in science, which may lead to improved learning or improved science literacy (Behrendt et al., 2014) and improvements in science-related test scores (Whitesell, 2016)
  3. An obvious question is how to extend the reach of such exhibits into the classroom using digital technology. By providing a more cost-effective alternative to in-person visits, virtual exhibits could enhance the reach of such centers (Behrendt et al., 2014). Ideally, given the value of hands-on or participatory demonstrations (Ekwueme et al., 2015), virtual exhibits would attempt to incorporate features that could encourage engagement with science. Also, AI tools have the potential to improve student success and engagement, particularly among those from disadvantaged backgrounds (Sullivan et al., 2023) and those marginalized from STEM due to cultural exclusion (Cook, 2023).
  4. To that end, our hypothesis is that AI-powered chatbots, if designed correctly, might provide a suitable basis for “virtual exhibits”, given their interactive nature. However, interactivity in of itself isn’t guaranteed to be engaging. We therefore explored potential mechanisms for elevating engagement. Research revealed two potential mechanisms: persona-based chat (Dwivedi et al., 2023) and interactive assessment via questions that exploit Bloom’s taxonomy of learning (Adams, N. 2015)
  5. Our longer-term research question, if we had the time and resources, would be to evaluate how effective the proposed chatbot-related ideas are in engaging students and enhancing learning outcomes in the field. However, this would require a full field test with the participation of students and their teachers, and possibly the involvement of a science center like the CSSC.
  6. Given our limited resources, we chose instead to evaluate the feasibility of using Generative AI (GenAI) chatbots to achieve the following:
  7. The use of chatbots in education is not new. Literature surveys suggest broad applicability to a range of educational circumstances (Pérez et al., 2020). Additionally, impacts upon some educational outcomes have been shown to be positive (Rong et al., 2023) such as the ability to provide per-student personalized learning (Winkler et al., 2018). According to Vazquez-Cano et al. (2021), a well-designed chatbot can make learning more continuous and automatic. The use of chatbots within the context of “micro-learning” has shown them to be effective in enhancing motivations towards learning (Yin et al., 2021). In a study to explore learning English vocabulary (as a foreign language) results showed that vocabulary gains in the test group were significantly higher than in control groups (Annamalai et al, 2023).
  8. Chatbots can be improved in terms of educational engagement by equipping them with human-like features by incorporating anthropomorphic design features (Dwivedi et al., 2023). Anthropomorphism can take the form of embodied designs, such as human 3D representations, or disembodied interfaces, such as text-only interfaces, but nonetheless still imbued with human traits, like “personality”. Many forms of anthropomorphism for chatbot designs have been attempted (Janson et al., 2023).
  9. These stylistic modifications can include personification (Pizzi, Scarpi, & Pantano, 2021) to create what is sometimes called personality-adaptive chatbots (Ait Baha, Tarek, et al 2023). Due to recent technological developments such as generative AI (Bommasani et al., 2021), adapting chatbots to include stylistic influences is possible via creative use of inputs (prompts), such as “write me poem in the style of grime rap”. This lays the foundations for enabling personified bot dialogs.
  10. Research by Annamalai et al., (2023) revealed that chatbots in the classroom supported competence, autonomy, and relatedness, which are aspects of Self Determination Theory, a psychological framework for developing motivation. In that study, students placed value on the use of chatbots in helping with assessment within a blended-learning approach. In terms of assessment, the development of meta-cognition has been shown to enhance learning outcomes via the use of questions derived from Bloom’s Taxonomy (BT) (Sudirtha et al., 2022)
  11. BT includes six categories in the revised edition:
  12. We performed experiments in QG initially with all six categories, but constrained the hybrid testing of QG with personified chatbots to only the first two as this was easier to evaluate and to control during AI generation. Also, for reasons that will become clear, the use of higher-order categories introduce certain complexities in the way the final solution might need to be designed and delivered into the classroom. Note that we did not explore the many criticisms of BT (Kompa, 2017), except for one of them: that learning is non-linear and doesn’t necessarily follow the neat ascent of the taxonomy. This criticism was of interest because of the capabilities of modern GenAI chatbot techniques in accommodating non-linear learning paths (see later).
  13. AI has recently undergone a transformation due to the invention of Generative AI (GenAI) (Garrido-Merchán et al., 2023). Our study utilized the latest GenAI technology (GPT-3.5) that powers ChatGPT. This technology can capture and retain contextual information throughout interactions, leading to more student-relevant conversations. Unlike previous-generation chatbots that follow fixed learning paths (or decision trees), ChatGPT can engage in open-ended dialogue. This seems more compatible with our goals to provide a more engaging experience as it allows the student some degree of autonomy.
  14. Moreover, its adaptability allows it to accommodate different language styles, and even write and debug computer code, making it a valuable tool in educational settings (Baidoo-Anu & Owusu Ansah, 2023; Tate et al., 2023). One such adaptability is the ability to incorporate stylistic influences, such as the use of personas to add a particular voice (“personality”) to the chat, thus aiding our goal to incorporate anthropomorphic features into the user experience.
  15. The tests in this work were conducted by prompting GPT-3.5-turbo and GPT-3.5-turbo-instruct. Both were used because of our use of the legacy mode in the OpenAI playground, which supports the former. We also used ChatGPT (3.5) to generate example text passages to compare with passages scraped from the NASA educational guide (e.g. https://www.jpl.nasa.gov/edu/teach/activity/modeling-an-asteroid/). Due to limited time, we only explored a single subject, namely the topic of comets and asteroids. We chose this because the author has prior experience in explaining and demonstrating asteroid formation to student visitors of a public science center (CSSC). This made anecdotal assessment easier. All results of the methods mentioned below can be found in https://github.com/yooleee/chatbot-research.
  16. ...and 71 more sections

Figures (4)

  • Figure 1: Example Template of Few-Shot Learning from Parnami et al.
  • Figure 2: Bloom's Taxonomy: Original (left) and revised (right) -- https://joanakompa.com/2017/02/07/why-it-is-time-to-retire-blooms-taxonomy/
  • Figure 3: Data Used to Populate Prompt Placeholders
  • Figure 4: Potential User Interface Components (Conceptual Only)