The Integration of Artificial Intelligence Education and Ecosystems Science for the Upper Elementary Grades

• In Event: Teaching Artificial Intelligence in K–12: Innovative Learning Designs in an Evolving Landscape

Abstract

OBJECTIVE
The [GAME] project is investigating a key question in computational thinking for upper elementary science education: how can we create deeply engaging learning experiences integrating artificial intelligence and life science for upper elementary students with immersive problem-based learning? With a focus on ecosystems science, [GAME] enables students to collaboratively learn and use AI tools by engaging in problem-based learning (Hmelo-Silver, 2004; Saleh et al., 2019). Additionally, we leverage principles from collaborative game-based learning (Buffam et al., 2016; Clark et al., 2016; Mott et al., 2019) as students are virtually transported to New Zealand and asked to investigate practices of population study to address a conservation issue: the yellow-eyed penguin population is in decline, but we do not know what the main threats are.

THEORETICAL FRAMEWORK & DESIGN FEATURES
In [GAME], we engaged in practices of co-design with teachers (Matuk et al, 2016; Severance et al., 2016) to design and develop the curriculum and immersive problem-based learning environment. Based on the co-design process with our teachers, we developed the following driving question: “What is causing the decline of the yellow-eyed penguin population, and how can we use our knowledge of science, computer science, and AI to help without causing additional harm?” The immersive scenario places students in the role of programming an autonomous RoboPenguin, a robot disguised as a penguin, to collect data and determine threats and solutions (see Figure 4.1). We address areas of computational thinking (e.g., reasoning about information, data, and perception) and AI education (i.e., AI planning, machine learning, computer vision, and AI ethics).

FINDINGS & NEXT STEPS
Research with two teachers from two different schools represented by grades 4 and 5 focused on AI planning, computer vision, and machine learning suggest that students generally like the PrimaryAI experiences, and found it to be fun and realistic, with one student saying, “It’s really cool how someone engineered a RoboPenguin to take pictures of the yellow-eyed penguin without it being scared.” Pre- and post-test results showed overall improvement with AI competencies but identified specific strengths and weaknesses with respect to AI planning, machine learning, and computer vision. For example, a paired samples t-test based on mean comparisons of computer vision pretest scores (n=40; M=11.6, SD=3.0) and posttest scores (n=40; M=16.2; SD=3.0) showed significant increases in learning, t(39)=9.6; p<0.001, and a high effect size, Cohen’s d=1.5. The computer vision assessment measured sub constructs for such areas as Pixels & RGB (example items: How does a computer tell two pixels apart?; What are the 3 colors represented in the stored values of each pixel?) or Feature Extraction (example items: If you were a computer and you would have to come up with a rule for edges making a rectangle, what would it be?; How many unique features are depicted in the image below?). Next steps in PrimaryAI involve refining our assessments and approaches, to include key insights into local adaptation and relevance for additional contexts represented by a diverse range of learners.

Authors

• Krista D. Glazewski, North Carolina State University
  Presenting Author

• Anne Leftwich, Indiana University
  Non-Presenting Author

• Katie Jantaraweragul, Indiana University
  Non-Presenting Author

• Cindy E. Hmelo-Silver, Indiana University
  Non-Presenting Author

• Adam Scribner
  Non-Presenting Author

• Bradford Mott, North Carolina State University
  Non-Presenting Author

• James C. Lester, North Carolina State University
  Non-Presenting Author