Aide is a conversational agent that employs AI to highlight strategies used in mis- and
disinformation.
Aide is part of an ongoing project that I lead at the
Transformative
Learning Technologies Lab based at Columbia University. Aide
is a react.js application that makes use of OpenAI's node.js package. User inputs are stored as
variables that are passed as part of custom-made prompts. These prompts are then sent as
requests to Open AI's API, which uses GPT-3 models to generate human-like text. GPT-3 is a
pre-trained language model built upon 499 billion tokens extracted from the web (Dale, 2021).
My role in this project: besides conceiving, designing,
and developing the
initial prototype, I am currently leading a team of 6 junior reesearchers involved in research
to assess Aide's
effectiveness as a learning tool as well as its UX and UI. During Fall 2022, we worked to devise
a study with 11 participants who tested the tool and underwent task-oriented interviews.
The design process involved
user interview,
low-fidelity
prototyping, and an
intensive, fast-paced
learning journey of the intricacies of
React.js. Below are some
images
that illustrate the initial stage of development, going from the initial idea of producing a
game-like experience to the refined, working prototype of an AI-based conversational agent.
image: super low-fi prototype
image: low-fi prototype using Google Slides
image: working prorotype developed with Figma
Click
here to view the
prototype. Here and here are conference papers
associated with this project.