The National Institute of Education invites suitable applications for the position of Research Associate on a 12-month contract (renewable) at the Office for Research.

Project Title:

Data and Theory Driven Artificial Intelligence to Boost the Science of Learning (AI4SoL)

Project Introduction:

Many primary school students struggle with mathematical word problems due to challenges in:

• Comprehending text, especially those with complex sentence structures or unfamiliar vocabulary.
• Identifying key information (particularly on how the numbers in a word problem are related) and selecting appropriate solution strategies.
• Applying structured problem-solving approaches and transferring learning to future problems.

Structure inquiry using Polya's Problem-Solving Processes covers four steps:

• Step 1: Understand the Problem: students need to understand the problem. This means reading the problem carefully and thinking about what is given and what the problem is asking.
• Step 2: Make a Plan: students think about how to solve the problem. There are many ways to solve problems, and students need to choose a good plan. In the Singapore primary math classrooms, the model method is used to plan solution. Students need to first select a model to best represent the problem and then complete the model.
• Step 3: Carry Out the Plan: Students follow the plan made to solve the problem. Here they do the calculation based on the model worked on in Step 2 to solve the math word problem.
• Step 4: Review/Reflect: students take a moment to check the answer and make sure it makes sense, and also to reflect what they have learned in the process while solving the problem.

Traditional teacher-guided structured inquiry leads students through the four steps to help them address the challenges in word problem solving, but its scalability is limited by classroom constraints such as teacher workload and time. Teacher-guided structured inquiry in a classroom setting cannot adapt to every student’s pace of learning nor could the teacher provide timely feedback to every student.

While self-studying worked examples allows students to learn at their own pace, such learning lacks immediate feedback when they are overwhelmed by vast amount of information designed in the worked examples. They may hence focus on mimicking the steps in a worked example without full understanding. Self-study may be also constrained by students’ motivation and self-regulation (e.g., monitoring and regulating one’s own learning).

The use of educational technologies is increasingly becoming more ubiquitous in mathematics education. With artificial intelligence (AI) being integrated into the development of educational technologies, for example Intelligent Tutoring Systems (ITSs), AI for education (AIED) applications could provide real-time error detection and personalized feedback to students. But the development of ITSs is often costly and limited (e.g., explicit development knowledge model and the design of hints/scaffolding using a rule-based approach). The optimization of adaptive feedback requires big data being collected for training the algorithms of the ITS. The use of ITSs may also lead to students’ over-reliance on ITS adapting learning progression and hints based on big data, subjecting students to potential data biases, limiting their learner agency and therefore under-optimizing their learning and transfer.

The recent widespread availability of pretrained Large Language Models (LLMs), such as ChatGPT, shows renewed potential for AIED: LLMs may provide timely personalized feedback to students in a more natural way (e.g., being dialogical) and on a scale, without a need to build explicit hints or knowledge models. 

There are still challenges to overcome in adopting LLMs in math teaching and learning, making it a promising nascent field in AIED research. For example, the extents to which LLMs can accurately solve math problems and assess students’ work, e.g., their conceptual understanding, etc. The effective pedagogical use of LLM for math teaching and learning is also emerging to be an important area of research.

Specifically in math education, mathematics learning requires both conceptual and procedural knowledge, and students learn mathematics through sense-making of these types of knowledge through problem-solving (e.g., word problem solving). This requires students to access and/or construct their own relevant mathematics knowledge, create representations of said knowledge, and map their representations to the knowledge. Besides using these steps to problem-solve, mathematics learning also requires students to communicate their problem-solving strategies and solutions. From a socio-constructivist perspective, co-constructing knowledge requires a dialogic exchange between teacher and students, and feedback from teachers is essential in mathematics discourse. Based on Thurlings et al.’s models of feedback processes, most feedback in computer systems is cognitivist in nature. The advancements in LLMs appear promising in bridging this dialogic gap in feedback and learning via computer systems.

Education Study 1 aims to test the efficacy of ChatGPT in teaching mathematics word problem solving through dialogue in structured inquiry. Specifically, this study investigates the extent to which a ChatGPT-supported structured inquiry application can:

• Improve problem-solving accuracy and conceptual understanding.
• Foster independent learning through scaffolded inquiry.
• Facilitate transfer of problem-solving strategies to new problems.

Findings could contribute to the growing literature on LLMs in education within the field of AIED and implications of design and development of LLM-based learning applications (e.g., through prompt engineering). Furthermore, the use of process data as a study instrument could contribute to both methodology (introduce system process data to support findings from research on technology-education interactions) and design (system designs that leverage LLM to enact educational practices (e.g., dialogic practice) that work.

Requirements:

Academic qualifications:

• A Master’s degree in Education, Psychology, Sociology, Learning Sciences, Learning Environments, or a closely related discipline.

Work experience:

• Prior experience in teaching Primary Mathematics, preferably in local school context.
• Prior experience in literature review, data collection and conducting quantitative analysis.
• Prior experience in writing academic or professional publications and conducting qualitative analysis.

Desirable interests, skills, and attributes:

• Strong interest in teaching and learning, especially in school-based research.
• Strong interest in effective instructional practices and learning design in classroom settings.
• Good communication and organizational skills
• Good interpersonal skills, with the ability to work collaboratively with researchers, teachers, and students.
• A self-motivated and proactive team player who is also able to work independently when required.

Responsibilities:

• Conduct literature reviews (summarizing, coding, and organizing relevant literature).
• Perform different statistical analyses for the findings obtained.
• Contribute to the design and development of research materials, such as mathematics-related learning tasks and instruments.
• Assist with data collection in schools or other research settings.
• Carry out other research-related duties as assigned by the Principal Investigators.

Application

Applicants (external and internal) will apply via Workday. We regret that only shortlisted candidates will be notified.

Closing Date

Closing date for advertisements will be set to 14 calendar days from date of posting.

Hiring Institution: NIE