How the project is organised
Workstreams
The early project work is organised around three connected strands. The first establishes a platform for teacher-configured pedagogical tutors. The second explores simulations, games, and students as creators. The third investigates directions that need research and feasibility work before they become commitments.
The strands share infrastructure and inform one another. An interactive artefact developed in one strand can become material for a classroom study in another, while evaluation work helps determine which AI capabilities are suitable for a task.
Strand A: pedagogical tutor infrastructure
The first strand develops the environment in which teachers prepare and deliver AI-supported physics activities.
Its central design pattern is a paired activity: a guided tutor in the conversation and a prepared workbench containing the problem, document, simulation, experiment, table, graph, or other representation being studied. The tutor can refer to that material rather than behaving like a generic chatbot.
Teacher preparation
Teachers can create classes and activities, choose curriculum material, define a teaching goal, select an activity structure, and review the student experience before sharing it.
Authoring is intended to support teacher judgement rather than automate it away. Where an AI co-pilot proposes content or configuration, the teacher reviews and applies the proposal.
Student activity
Students join a teacher-prepared activity using a group code. Within the activity they may converse with a tutor, use a simulation, inspect documents, record observations, complete a checklist, submit an image, or work with another structured representation.
Teacher and research review
Teacher-facing views help educators review how groups used an activity. Research views support analysis across approved studies and roles. Access to those views is distinct from student access.
Safeguards and boundaries
The strand includes group-based access, central management of AI services, reviewed interactive artefacts, explicit teacher control, and environment-specific deployment and testing. Privacy and consent requirements remain study gates rather than assumptions embedded only in software.
Strand B: simulations, games, and students as creators
The second strand explores what changes when learners do more than consume an AI-prepared activity.
Simulations and virtual laboratories
Interactive physics artefacts can support prediction, parameter exploration, procedural rehearsal, data collection, and comparison between a model and a real phenomenon. AIPLA studies several forms:
- Phenomenon simulations, where students vary parameters and observe relationships.
- Procedural virtual laboratories, where students follow and reflect on an experimental sequence.
- Structured workbenches, such as tables, calculators, notes, charts, documents, or concept maps.
- Hybrid activities, which combine several representations with a tutor.
The aim is not to replace physical experiments. Virtual and computational tools are useful when their role in relation to real equipment, measurement, and uncertainty is made explicit.
Student-as-creator
A further direction asks whether students can use AI assistance to build, test, and critique their own simulations or games. This changes the learning task from operating a representation to constructing one.
Student-created artefacts require stronger review, provenance, and safety controls. The work therefore builds on the same versioning and reviewed-artefact approach used for teacher-prepared activities.
Strand C: investigation and scoping
The third strand examines questions that are promising but not yet mature enough to treat as delivered classroom tools.
Beyond a single model class
Different tasks may require different capabilities: textual reasoning, diagrams and images, audio, structured data, or interaction with a simulation. The project evaluates capabilities per task rather than assuming one model is best for everything.
Beyond chat
Chat is useful but not universal. Voice, concept maps, branching structures, shared group work, interactive diagrams, and physical data collection may be better interfaces for particular learning goals.
Models of student understanding
One research direction is whether a student's developing conceptual structure can be represented and compared with a curated reference. A visible concept network or “teachable agent” could make that structure an object students inspect and revise.
This direction raises substantive validity and privacy questions. A model-generated account of student understanding must not be mistaken for a neutral or complete measurement.
Standards and assessment material
Collections of physics tasks can support evaluation of AI capability, analysis of assessment design, and exploration of new question types. Rights, provenance, answer-key quality, and expert validation are prerequisites for using such collections responsibly.
How the strands connect
The shared platform makes it possible to test several activity forms without building a separate application for each one. The research questions determine what is worth studying; teacher dialogue shapes the activity design; capability evaluation sets technical expectations; and classroom findings decide what should be refined, changed, or stopped.
This is a portfolio rather than a promise that every explored idea will become a permanent feature. The progress page distinguishes current platform capabilities from provisional research directions.
- Content status
- Current
- Maintained by
- AIPLA project team
- Last reviewed