15 April 2026

Tuesday 26 May at 10.30am

CRADLE Seminar Series 2026: Seminar #5

In this seminar, the University of Oxford’s Kelly Webb-Davies will outline Voice-First Written Assessment – a two-stage assessment model designed in response to the evidentiary crisis that generative AI poses for educators assessing students’ ideation and reasoning.

Institutional responses to generative AI in written assessment have often framed it as a problem of authorship

This seminar argues it is more fundamentally a problem of construct representation and linguistic access: even before generative AI, requiring performance in prestige academic prose risked distorting or suppressing students’ best thinking, and as a result undermining valid inference about disciplinary reasoning. Now that fluent prose can be generated completely independently of a student’s ideational labour, the already unstable link between polished prose and the thinking it was assumed to represent has been entirely severed. Detector-led and disclosure-based responses cannot repair this evidentiary chain.

Voice-First Written Assessment (VFWA) is a response to this evidentiary crisis. It is a two-stage assessment model that reconfigures where and how construct-relevant evidence is secured and interpreted. Stage 1 is ungraded and produced under observed conditions, establishing an authenticated baseline of ideation. Student choice of linguistic repertoire or modality provides a more interpretable representation of students’ reasoning. Stage 2 permits unrestricted, optional digital tool use reflecting contemporary writing conditions, and enables linguistically and cognitively accessible methods to refine written work.

The model integrates validity theory, cognitive models of writing processes, and translanguaging scholarship, claiming that where prestige academic English is treated as proxy for intellectual effort, construct-irrelevant variance and underrepresentation are predictably introduced. Translanguaging is therefore not simply accommodation, but is a validity condition: a means of protecting evidentiary access to reasoning.

VFWA establishes minimal core conditions that make it adaptable across subject areas. It has attracted emerging institutional uptake across multiple disciplines and educational sectors. The seminar will present the evidentiary logic and operational design of the model alongside implementation examples and an overview of planned pilot applications.

Join us at 10.30am (AEST) in person or online to learn more about VFWA

About Kelly Webb-Davies

Kelly Webb-Davies is Lead Education AI Consultant at the University of Oxford’s AI Competency Centre, where she works on integrating AI into higher education in ways that enhance communication, reduce bias, and expand access to knowledge.

Drawing on a background in linguistics and academic English teaching, she is particularly interested in developing more inclusive and accessible educational environments by exploring how AI can support learners facing linguistic barriers and neurodivergence.

Also coming up in May

Entangled Intelligence: Is AI changing the way students think?

Wednesday 13 May at 2pm