Never mind AI, here comes RI!

While the hype called “AI” is churning out bucketloads of perfectly useless rubbish concocted together from content stolen from all over the internet, let me show you that we don’t need no steenkin’ AI to generate nonsense. As Andrew C. Bulhak of Monash University showed in 1996, it is perfectly possible to generate academic articles that can even be accepted in a peer-reviewed journal using a simple script, some wordlists, and a keen sense of how to use them to, for instance, prove that in some academic disciplines, the Emperor really does not wear clothes.

This is how it works. You create a “context-free grammar” – that’s a set of rules how to create a word, a sentence or a paragraph. Think about a sentence with the words replaced by placeholders, for instance: (subject) (verb) (object). The script will replace each of those by a random word picked from list of subjects, verbs and objects and display the result.

The script itself is dumb. It knows nothing about the contents: it only knows what form the output should have.

To create interesting contents is where the author comes in. By carefully creating the grammar and selecting words it is possible to generate words, sentences and paragraphs that are completely random, but that look like meaningful writing. Andrew Bulhak tailored his Postmodernism Generator to mimic typical postmodernist writing close enough for it to be accepted by a peer-reviewed journal (see the Sokal Affair for the juicy details).

I found that this technique is very well suited to identify (and poke fun at) jargon-heavy content. This can be found in many places: marketing, anything that’s meant to appear hip, buzzword-infested corporate babble and new-age spirituality, just to name a few. 

When used on the word-level it can be used to identify what emotional charge is connected to certain combinations of phonemes. But above all, I found it irresistibly funny.

The script I use is based on James Baughn’s Nonsense perl script  from 2001 (!) that I updated to get it working on a webserver again. I also added some features I was missing, like weight factoring for words and variables (making them more or less likely to be picked). You can find my updated version on Github here.

Here are some examples of how I used

  • Medigoed BV (in Dutch): a randomly-generated pharmaceutical  company’s web-page, filled with random medical problems and random drug  products to address them. Product & mood images are taken from free-to-use photo-websites, hosted locally to improve performance.
  • Yurpeana (English): a parody of my employer’s network-oriented ‘pro’ website (, complete with hero image layout, snazzy calls-to-action and a nonsensical employee-overview.
  • Flapdoodle: an attempt to generate the sort of over-the-top nonsense that some new-age-advocates produce without batting an eyelid

Keep in mind that the content is entirely random, and any resemblance to a real-world something or someone is purely accidental. Even if you happen to be offended by it.