T127 Reflection 3
Capstone reflection on designing a Generative UI system for Harvard's data fluency course.
The pressing challenge that the TLL team is facing with the transition between the Discovery and Design phase of the Harvard-wide data fluency course is content that is relevant and helps bridge the siloed departments across the schools. Staff across Harvard struggle with having time to do another course where skills are usually learned on-the-job. Though the first instinct is to design a top-down course, I think that cutting-edge solutions like Generative UI could be employed here.
My capstone’s primary goal here is: Address the problem of scaffolding the right information to the right user just-in-time and limiting redundancy. Do it in a way learner-created artifacts help inform decisions to bridge the difference in dictionary definitions used across Harvard schools.
The primary focus will still be the users. Generative UI is essentially an interface that renders more structured support (worked examples, guided steps, glossary panels) for novices and progressively strips them away as the learner model indicates competence.
The Generative UI that I’ll be designing also copies ideas from the concept of ‘Learnersourcing’. I’ll be designing a system where answers to dynamic questions help build a Harvard-wide dictionary/wiki. This in turn allows for a richer content base that could be cross-referenced by learners in the future. The primary concern will be quality control, so there must be a tight ingestion pipeline that’s clear and editable (open-source), so anyone can inspect it.
This generative interface aims to display content most relevant to a user’s case (e.g. a data handler in the GSAS admissions office deals with different sets of data than one in the GSE administration office). When it comes to data, the limits of conversational AI or rigid modules might hinder deep learning and contextualization.
This capstone aims to be a proof of concept. A way to supplement learning that’s currently neither contextualized nor just-in-time. The whole design will be open-sourced. I will design it to be portable for the Harvard LXP schema in case there’s an interest in using it in the Data Fluency course in the future.