Bad News: Ai news network

Watch the video to see a dumb AI generate dumb news headlines.

Short version:

This dumb AI writes dumb news headlines based on current events.

Long version:

Contemporary political and cultural opinion is increasingly driven by emotional reactions to often-dubious, context-free information. Financial incentives on the Internet and on social media mean that the information that gets promoted is that which best drives “engagement.” Emotion — particularly outrage — is a powerful driver of engagement, and media companies seeking views see few downsides in promoting rage-inducing material even when that information contains no truth nor value to society.

If generating outrage is more valuable than reporting truth, media companies will soon adopt tools designed to most efficiently create rage-inducing content. The AI News Network is such a tool.

AI News Network is designed to produce an endless stream of news-like information that is tuned to elicit emotional reactions. Though an Internet connection allows the system to write about people and events currently in the news, the sentences are randomly created through a simple algorithm that has no connection to reality.

The device itself employs graphics and technology reminiscent of the (very) early days of the Internet as the tool doesn’t rely on modern technology like machine learning. Its simple AI can hardly be called intelligent and could have existed when computer networks first emerged. It isn’t modern technology driving our addiction to dubious information and partisan rhetoric; it’s our human desire to react to anything sensational.

All the information from the AI News Network is false, but if it engages us, do we even care?

How was this made?

The program that generates and displays headlines is written in Python. The source code is here. Each hour the program fetches real up-to-date headlines from RSS feeds. The program then categorizes each word in the headlines — for instance, proper nouns and adjectives each go into separate categories. The program then randomly reassembles these words into predefined headline structures, creating sometimes-plausible, often-inane headlines.

The program runs on a Raspberry Pi and the output is displayed on an old CRT monitor. It’s a real object. It’s currently in the IDEO Tokyo studio.

Before Twitter got weird, there was live feed of generated-headlines available via a Twitter bot. You can still check out the archive of headlines on that site (if it’s still in business).