Widening Perspectives on Learning and Research Opportunities
Locations of grocery stores in Puerto Rico. Created by Normandie Gonazales, graduate student at University of Puerto Rico.
During my 2023 summer internship at The Learning Partnership, a private company providing science, technology, engineering, and math programs for K-12 schools, I developed an AI-supported tool to enhance high school students’ spatial reasoning skills and their understanding of local issues in Puerto Rico. The tool aimed to bridge the gap between software tools and learning by integrating an AI-tool that helps with map visualizations and data relationships, designed specifically for students with no prior experience in Geographic Information Systems (GIS). In this piece, I’ll dive into the details of my design process, my learnings, and the broader implications for PhD students considering internships with small and mid-sized private companies.
Design process and what I built
The design process included evaluating existing tools within a curriculum as well as reviewing related literature. The curriculum provides students with research questions related to food deserts in Puerto Rico, which is a local issue that several districts on the island face. Insights into the challenges students face were provided by my internship host who had observed students’ interactions with GIS tools for several years in previous research. Challenges he noticed with current tools included complexities of software tools to explore relationships between data points and visualizations, complexity of GIS-specific concepts, and a lack of real-time assistance when students encountered difficulties with GIS tools. The students would lose track of original research questions because they would get confused trying to use the tool.
To address these challenges, we envisioned a tool that would use natural language to support interactions with spatial data, provide concise definitions and explanations of GIS terms and concepts, and support students in visualizing spatial relationships with ease. This, we thought, would enable students to focus on exploring relationships between data and local issues, instead of focusing on learning how to use a complex tool. This rested on the principles of a design framework that guides the design of technology-supported, content-driven inquiry tasks. The Learning for Use model is structured into three essential phases: motivate, construct, and refine. The motivate phase aims to spark students’ curiosity and create a demand for knowledge, which is accomplished through the curriculum’s focus on local issues. Construct activities immerse students in hands-on experiences and observations, fostering new knowledge through communication, which is accomplished through interactions with our AI tool. Finally, the refine phase encourages the application of understanding and metacognitive reflection, which is accomplished through final deliverables such as chatbot logs and journal entries.
I developed a prototype (see Figure 1) integrating these design principles. Specific features included: (1) answering questions about local issues with reliable spatial datasets; (2) controlling map visualizations using natural language commands (for an example, see Figure 2), and (3) scaffolding to answer original research questions. As the tool evolved, these features grew in complexity. Feature 1 expanded to include local demographic data, while Feature 2 incorporated both written and voice commands. Feature 3 became more sophisticated, allowing the model to track ongoing conversations with the ability to help students reflect to answer research questions.
Fig. 1. Interface of tool.
Our initial assumption was that the tool would primarily support individual curriculum completion. However, through the user-centered design process, it gradually transformed into a more interactive tool, engaging pairs of students in collaborative problem-solving dynamics. These students not only interacted with the tool but also with each other, sparking debates about what to ask and how the tool functioned, occasionally even challenging the tool’s output.
Fig. 2. Adding layers of data to a map via natural language commands. In this example, students asked for data on stores in different cities as well as a store density layer.
What did I learn?
Through the design process, I learned many topics and skills, such as guiding student inquiry, basics of Large Language Models (LLMs), web development, and multimodal interaction.
First, it was valuable to observe the challenges students face when learning a tool, as well as the importance of social interactions and collaborative learning. While the tool provided substantial support, collaboration among students sparked deep interactions in terms of what questions to ask, interpret responses, and hypothesize how the tool derived answers.
Learning the basics of LLMs and their applications enabled me to understand their potential and limitations in learning domains. While they could provide valuable assistance, there were numerous moments when the LLM struggled to grasp the complete context of the task at hand. It took a lot of back and forth to get the model to understand the dynamics of the learning experience and how it could serve as a valuable helper for students to gain an expansive understanding of local issues. Ensuring that the model provided relevant and useful data was also a valuable experience. The tool guided students through the complexities of spatial relationships with demographics and supermarket data, providing direction and responses with data while fostering critical thinking and not only answering with some numbers. It became apparent that the tool’s role was more akin to guidance and scaffolding the process of inquiry, rather than serving as a way to directly provide answers to the students’ questions.
Finally, building a multimodal web app was a valuable experience as it highlighted the intricacies of individual learning preferences. A significant takeaway was the realization of the tool’s role as a facilitator for collaborative learning rather than a tool for individualized map exploration.
Reflection on the experience
As I reflect on my internship, there are a few important insights for other PhD students as they approach internship recruitment season. The insights are to expand your horizons and approaches to networking, think expansively about your goals, and consider intellectual freedom.
During the first couple of years at Northwestern, I focused on pursuing internships with bigger tech companies (i.e., the usual FAANG). While I secured a few interviews with Meta, Google, and Apple, I did not receive any offers. This experience taught me that there are many other companies out there that offer valuable learning opportunities.
I approached networking with a sincere mindset, sharing that I wanted to do work related to Puerto Rico and that I am trying to learn from people. I always shared my interests in doing work related to my community, and rather than solely seeking tangible benefits, I aimed to learn from people. Networking, I realized, is not just about what you can get but also about what you are about and what you are seeking to learn. I cannot prove that my internship at The Learning Partnership was the result of this mindset, but I do believe it played a large role.
I have been trying to do work related to issues the Puerto Rican community faces, as well as accessibility, and AI for several years now. If you told me three years ago that I would have been doing an internship that overlaps all three of them, I would have been a little skeptical. It turned out it was possible. I just needed to think more expansively about how to do this kind of work. As in research, when approaching learning experiences like internships, it’s crucial to consider a diverse range of organizations, projects, and approaches.
Lastly, what I believe is one of the most significant advantages of interning with a smaller company was the intellectual freedom it offered. Although I did feel lost at times, due to not knowing how to go about what I was envisioning, the opportunity to explore and implement my ideas was incredibly rewarding and energizing. I cannot contrast this with bigger companies, but the freedom provided at The Learning Partnership was extremely beneficial to my growth as a researcher.
In conclusion, my journey in developing an AI-powered learning tool at a private company was a positive experience. It allowed me to understand the foundations of LLMs, web development, and the nuances of guiding student inquiry. Furthermore, my internship offered valuable lessons in expanding my horizons, networking with a sincere mindset, thinking expansively about my research goals, and embracing intellectual freedom. These lessons, I believe, are valuable takeaways for any student considering an internship in today’s dynamic and competitive internship landscape.
