SELF-DIRECTED · 2 min read
Augmented Information Foraging
An early predecessor to my serendipity project, this prototype imagines a side-scrolling Wikipedia with AI-assisted way-finding
This React prototype uses an LLM to determine logical flows between Wikipedia pages, enabling a horizontal infinite-scroll experience.
This React prototype uses an LLM to determine logical flows between Wikipedia pages, enabling a horizontal infinite-scroll experience.
Browsing With a Subtle Guide
AI that respects human agency
I started this project in 2024 imagining a side-scrolling Wikipedia that enabled users to scroll right indefinitely. Users would be able to keep scrolling right as long as they like and AI will pick the next page they go to. The key is that Wikipedia would work as normal, but AI kicks in as soon as the user presses the right arrow key.
The right navigation shows related pages the user can choose from. The user can also just hit the right arrow key and the AI will pick for them.
The right navigation shows related pages the user can choose from. The user can also just hit the right arrow key and the AI will pick for them.
For tablets, the side navigation becomes a drawer.
For tablets, the side navigation becomes a drawer.
AI-Generated Information Scent
The navigation is generated by the LLM that uses a simple pipeline. It ingests the Wikipedia page content, then applies three categories to it: related subjects, broad perspectives and tangents. After that, it comes up with links and summaries that fit those categories.
This is the start of a concept I'm drawn to: AI that works on behalf of the user, but gives them agency. The AI creates a semantic bridge that categorizes paths forward to give the user a scent of where the current page might lead.
The AI produces related links that it fits into three categories: related subjects, broad perspectives and tangents
The AI produces related links that it fits into three categories: related subjects, broad perspectives and tangents
A failed experiment and recovery
I started with a pretty crazy idea centered around trying to control Wikipedia with hand gestures.I created a React prototype and experimented, but it never quite worked the way I wanted. But one part of the design was interesting: link previews.
A gesture-control prototype envisioned users raising a hand to highlight links. When users closed their hand, the app would generate cards containing Wikipedia's link summary data. It didn't quite work the way I was imagining.
A gesture-control prototype envisioned users raising a hand to highlight links. When users closed their hand, the app would generate cards containing Wikipedia's link summary data. It didn't quite work the way I was imagining.
HOVER for tablets
While the gesture-controled Wikipedia navigation idea didn't quite work, the link highlights were interesting . The user can click the "link preview" tab, which changes the app state to preview mode. The first few links are highlighted and their descriptions appear on the right.
The result is a contextual menu that reflects the content currently in the view area.
A React prototype shows how as the scolls, the top links highlight, and the right nav becomes a menu of page links that are contextual.
A React prototype shows how as the scolls, the top links highlight, and the right nav becomes a menu of page links that are contextual.
Evolution
How would this work on a mobile device?
As I started to think about how this concept would work on a phone, I came up with an idea: What if left and right swipes used AI, but in different ways. A right swipe would lead to Wikipedia pages that are conceptually connected. Left swipes would "zoom out" to broad, but related subjects.
The idea was exciting, so I wrapped up work on this and created SerenWikity.
