Play here

---

Algorithms have changed our lives. From social media to nuclear defense systems, algorithms define our lives. But recently, their use on social media has helped spread mis/disinformation and increase polarization.

The project

A recent study that me and some friends were looking at proposes (at least from what I understand) the use of debunkers and negative samples of information to find mis and dis information across social media. We decided to do a workshop for our school’s yearly event called Humans of PCDS which analyzes how a theme is important in our lives. This year the theme was misinformation and algorithms which fits this topic perfectly.

Together we worked to create a presentation for our workshop. I worked on the demo to show how this debunker idea could work. My goal was to show how an algorithm like this would work for a non-technical audience.

The game

To do this, I created a game. The objective of the game was to correctly identify as many posts as possible. You are given one post that you know is true (from the debunker). You are then given a list of 6 reported posts. You have to identify which are true and which are false based on the debunker’s post.

Development

I used sveltekit with shadcn-svelte, my current favorite development stack. I used tailwind to line everything up for all the pages, and shadcn for the ui elements. I ran into some issues making it fully modular but that was just due to the server’s response not containing all the correct elements. I found that surprisingly you can get data from $lib/server using the import syntax even when deploying to a serverless platform (cloudflare pages in my case). I used a component for rendering the posts, and binded the postId. I did this so that I could use a $derive block based on the postId for the current post and avatar. Originally I was running into an issue where getting the dummy json was making the server’s response time slow. To fix this, I just offloading the avatar collection to the client. When I was finishing up the project I noticed that you could see if posts were true/false on the client. I used Valibot since it was a lightweight form validator that was perfect to just remove a field not in the client schema.

The presentation

The presentation went well, but most people only played the game once sadly (there are 10 unique levels). I learned a lot about ui design (replaying should be easier), and also to add a tutorial for my projects in the future.

We talked with our workshop about the use of this specific strategy on social media. Most people didn’t want it as the deciding factor for whether they viewed content on social media, but were fine if it up-weighted posts which had a high probability of being true.

Debunker based algorithms represent an interesting step for algorithms on social media, but further adoption remains to be seen.