A game-metaphor of our public conversation online
It starts with a link. A source of information. Process it into a tweet-sized sentence. Filter it algorithmically according to some type of similarity to your belief system & values. And that’s when you enter the digital public arena to engage in conversations and debates, empowered by your quotes, ready for the daily clash.
Phrase Fighter is a playful exploration of that clash, a game-metaphor of our public conversation online. It is a classic “space shooter” game with phrases algorithmically curated and filtered by some of the great and disturbing voices of our times: AOC, Beyoncé, Elon Musk, WHO, and QAnon. After you submit a link, the algorithm offers you three phrases from it that sound most similar (semantic similarity) to a sample (tweets, lyrics, etc) from each filter.
But in the end, none of it really matters because it’s all about shooting the opponent’s words. The player invites an opponent by sending an invitation link, initiating the match: two phrases against each other, where the goal is to eliminate each other’s word. You can listen to the words falling in a computer voice that feels almost like a terrible freestyle rap.
How does the semantic similarity work?
It uses very succinct sentences to make the comparison. The filter for the American politician Alexandria Ocasio-Cortez (AOC), for example, uses “Healthcare and free tuition for all. Housing as a human right. Tax the rich. Our system is fundamentally broken. Inequality in this country is immoral. Working-class America deserves a living wage.”
And for the conspiracy group QAnon, a couple of sentences like: “Pizzagate is real, stop the child trafficking ring. Hollywood Liberal Elite Harvesting Blood for a drug called adrenochrome. The deep state is behind it. Ask the Q.”
The player’s choice of phrases is extracted by comparing the “closeness” of each of the sentences in the article with the chosen filter. GPT2, a text deep learning model, is utilized to get sentence vectors for each sentence. This reduces each sentence to a set of numbers that we can measure against. Two sentences that are semantically similar are achieved by measuring the distance between the numbers representing the filter and the article’s sentences. The top three closest sentences are then picked by the game.
Settling debates with computer games
The project highlights the way algorithms filter our sources of content and sets the stage for the clash between phrases. As a research-based project, we also speculate scenarios where debates and arguments take place in a computer game environment. Navigating the similarities between these different aesthetics - social media public conversation, computer game, and political debate — we remix, envision impossible mashups, and investigate their impact. What we talk about when we talk about anything.
Invite a friend (or enemy) and play at phrasefighter.wtf
Built using p5js, peerjs, matterjs, vue, gpt2, howler, bootstrap.
Made by Imagination of Things. 💛
