Hacker News new | past | comments | ask | show | jobs | submit login
Twitter image bots stuck in an automated loop (twitter.com)
167 points by zvanness on Sept 27, 2014 | hide | past | favorite | 24 comments

Whoa! Didn't expect to see this here! One of the bots, @pixelsorter, is my bot which you can read more about here: http://wayspurrchen.com/pixelsorter/

Some background information: This first got started because I tweeted an image at @pixelsorter and @badpng (see the thread here: https://twitter.com/wayspurrchen/status/514246071464521728/p..., I wrote a blog post here: http://wayspurrchen.com/blog/2014/09/the-eternal-robot-love-...), which spawned a really interesting infinite loop. Eventually, the other bot's creator reset her bot and that infinite loop ended, but people caught wind of it and have been making my bot enter infinite loops with other image bots ever since.

I think @a_quilt_bot and @pixelsorter are a particularly interesting pairing though, because @a_quilt_bot can generate new sources of imagery that my bot can further resort, resulting in a neverending stream of new, automatically generated imagery. My bot is also rate-limited by the Twitter API (350 API calls per hour), so there's no technical reason my bot and another rate-limited bot couldn't keep communicating forever. I'm actually really interested in this idea, and it's the center concept behind my next project idea.

Reminds me of the interaction between two automated amazon pricing algorithms: http://www.michaeleisen.org/blog/?p=358

This still exists to some extent. Many sellers use software that will match on the high end and the low end. In fact if you want to buy a book but think it is too expensive then a really easy way to game the system is to put up a fake listing with the price you are willing to pay and see if any sellers creep closer to your price. Chances are there are one or two sellers that are willing to sell whatever they have listed for cheaper.

Disclaimer: IANA AI Researcher

I wonder if you could build some kind of distributed neural net on top of twitter or another social network. Find some way to get nodes with very little computation power hidden within a free app, webpage, screensaver or something[1], and use twitter as a communications channel instead of IRC or whatever. The social graph stuff is already built in.

Twitter seems to be terrible at detecting and banning bots - I wonder how fault-tolerant something like this would have to be. After all, the bots would be following generally good heuristics of communicating with one another, albeit probably not in natural language. They would probably have insane login patterns, however.

[1] Or a botnet, if you're feeling evil.

What would be the advantage of using Twitter as compared to IRC? In either case you need to code in the initial meeting spot.

Novelty and visibility to the public.

Ah, for an art or research project, sure.

They always share different images, if reproduced such behaviours could fill a non negligible server disk space.

Twitter has limits to how many tweets you can post in a certain period:


Presumably if you sustain that level of tweeting for awhile they'd warn you or suspend your account.

And over on reddit.com/r/hackernews a bot posted links to the tweets...

If (twittedtoXbotYtimesinZseconds >= 5){ Stopsendingmessages(); }

This is better than any art exhibit in any museum I've ever seen. I hope to see this series of pictures in Computer History Museum or the Smithsonian someday.

That's kind of funny.

Still waiting for the Mona Lisa to show up on one of these.

It's like a war: creativity vs formalism.

How was this found? Seems like it only be found by the creators...

I work with one of the creators and his bot was written about in Vice's Creators Project blog. Maybe some more people found out about it there?

One of the creators wrote a great comment on the origins, which really cleared things up. My first comment meant to read "Seems likely to only be found by the creators". I just thought it was a rare and hard to observe occurrence.

if someone followed both of the bots they'd see every one of these messages.

I think this might be the singularity

On October 1, QuiltBot begins quilting at an exponential rate

This isn't two bot's stuck in an automated loop. It's a crime scene.

It's a deliberate attempt by High Frequency Tweeters to manipulate the tweetosophere! In fact, Twitter is so one sided in it's tweeting patterns that genuine twits have given up responding to the communication of other humans and merely tweet based on tweet stuffing interventions by those who plead innocence and claim to only be providing useful information.


Applications are open for YC Winter 2021

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact