From the Student Success edition of October 20, 2026
City St George's, University of London pilots Duologue, an in-house AI platform that leads Socratic conversations to assess understanding continuously instead of detecting AI-written essays
City St George's, University of London; Times Higher Education (publisher) · Higher education assessment and academic integrity / AI literacy · London, United Kingdom
- Publisher
- Times Higher Education
- Original publication
- Times Higher Education opinion, published 19 October 2026, last updated 20 October 2026
- Source retrieved
- 2026-10-21
What happened
Baba Sheba (City St George's) and Mike Bond (Bond & Coyne) describe Duologue, an AI platform built and piloted at City St George's in which the AI leads a Socratic conversation probing what a student knows, so assessment happens throughout learning rather than at the end. Game-like labs, often with two or three students at a screen, teach how prediction algorithms, tokens and classifiers work. Development was mainly funded through the university's Digital Innovation and Transformation Office, with module input from the Raspberry Pi Foundation, Sage Publishing and Wolfram. A safeguarding layer once flagged an innocent juicing-diet reference and needed local recalibration. The authors say the platform is still in pilot and they are wary of overclaiming.
Why it matters
A developer account of replacing detection with continuous, conversation-based assessment, published as U.S. and UK institutions debate in-person exams versus AI-integrated assessment. The previous run opened it but deferred it because the page carried a 20 October date; it is not archived.
Evidence and measured results
Developer description and design rationale. No outcome, validity or reliability data; the detector false-positive point links to regulator casework, while the claim of bias against non-native English speakers is stated without supporting detail.
Limitations and uncertainty
Opinion by the developers, one of whom is an agency partner; single-institution pilot with no published results. Conversations with other universities are reported, not verified.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-21; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Essays and AI detectors no longer give institutions confidence about what students understand.
- Stakeholders
- Pro vice-chancellor for education
- Academic integrity lead
- Learning design team
- Digital transformation office
- Student welfare and safeguarding lead
- Discovery questions
- Which modules rely most on take-home essays?
- Would the institution build or buy a conversational assessment tool?
- How would conversation evidence be weighed in marks and appeals?
- Who reviews safeguarding flags?
- Value hypothesis
- Interpretation: continuous conversational assessment could give earlier evidence of understanding than end-of-term essays, if its judgements are validated against academic marking.
- Potential engagement
- A bounded low-stakes pilot design: module selection, marker-agreement study and safeguarding review process.
- Unsupported claims
- Reduced misconduct
- Valid or reliable grading
- Learning gains
- Lower marking workload
Pre-sales engineering
Role takeaway
- Fit
- Fits institutions with in-house learning technology capacity and a willingness to redesign assessment.
- Architecture and integration
- In-house platform with AI-led Socratic dialogue, game-like labs and a safeguarding layer, plus partner-contributed module content.
- Prerequisites
- Learning designers and technologists
- Module content and rubrics
- Conversation logging
- Safeguarding escalation procedures
- Constraints
- UK context and institutional build
- No published validity evidence
- Partner IP in some modules
- Classifier needs local recalibration
- Security
- Protect conversation transcripts as sensitive student records, restrict access to safeguarding flags and document retention for appeals.
- Proof of value
- Agreement between platform judgements and academic markers
- False-positive rate of safeguarding flags
- Student experience survey
- Consistency across student groups including non-native speakers
Delivery
Role takeaway
- Work
- Pilot conversational assessment in low-stakes modules with marker comparison and a safeguarding review route.
- Dependencies
- Academic staff buy-in
- Learning design capacity
- Platform engineering team
- Student welfare services
- Ownership
- Education or digital education office, with module leaders owning assessment decisions.
- Skills and adoption
- Staff training on interpreting conversation evidence and student guidance on how the dialogue is used.
- Governance checkpoints
- Assessment board approval before any weighting
- Safeguarding procedure review
- Equality impact review
- End-of-pilot evaluation
- Acceptance criteria
- Proposed: substantial agreement between platform and academic judgements on a sample
- Proposed: every safeguarding flag reviewed by a person within two working days
- Proposed: no assessment weight until validity review passes
- Proposed: student experience measured against the essay baseline
- Risks
- Unvalidated judgements affecting grades
- Over-sensitive safeguarding flags
- Staff resistance
- Bias against particular student groups
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
An in-house conversational assessment platform with a safeguarding classifier needs local tuning, logging of conversations and human review of flags.
Governance
Who approves, reviews and stays accountable for outcomes?
Continuous AI-led assessment raises questions about marking validity, appeals and how conversation evidence is weighed; the article does not say how results feed grades.
Security and privacy
What data, permissions and controls need testing?
A safeguarding layer monitors student language for concerning content, which implies sensitive data handling and welfare escalation procedures not described in the piece.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Some staff were sceptical that a chatbot can stand in for a seminar; workload effects on academics are not reported.
Procurement
What should contracts, pricing and exit terms secure?
The authors argue for building inside the university rather than buying plug-in products; buyers should weigh build capacity and partner IP contributions.
Operating model
Which teams own the service once it runs?
Shifts integrity work from detection and discipline to designing conversational assessment, which needs learning designers, technologists and academics working together.
Publication history
- 2026-10-20Student Success · Issue 436 resources
Stable resource ID: the-city-st-georges-duologue-socratic-ai-continuous-assessment-pilot-20261019