67

A (now suspended) Twitter bot under the handle PZFeed Ebooks (@PZbooks) posted this tweet on November 11th, at 11:44 AM. Note the date carefully, that's November 11th, 2015.

enter image description here

It reads:

BREAKING: Death toll from Paris terror attack rises to at least 120 with 270 others injured

And yet, there are some news articles posted just today saying the exact same thing, with at least 120 dead, such as this article, this article, and many others.

And yet, despite the shootings happening on November 13th, 2015, the Twitter post was made on November 11th, 2015. That's right, a full 2 days even before the shootings happened.

If you click on the link to the tweet it's clear no part of the page has been modified, and the tweet itself has been replied to and retweeted countless times.

What could explain how it predicted the death toll from the Paris terror attacks a full 2 days before the shooting went down?

Oddthinking
  • 140,378
  • 46
  • 548
  • 638
yuritsuki
  • 1,965
  • 2
  • 15
  • 25
  • 10
    This sounds like the [Texas Sharpshooter Fallacy](https://en.wikipedia.org/wiki/Texas_sharpshooter_fallacy): if you make a large enough number of random predictions, eventually one of them is likely to be right, but your method still has no predictive power. – jmite Nov 16 '15 at 21:09
  • Similar to http://www.mirror.co.uk/sport/football/world-cup-2014/world-cup-final-fifa-corruption-3853975 – Viktor Mellgren Nov 17 '15 at 13:51

1 Answers1

124

The name PZFeed Ebooks is suggestive that

  1. it is a bot running on some kind of Markov chain, which will tell the software what kinds of words are followed by other kinds of words. These are commonly found on Twitter with the "ebooks" suffix in honour of an early notorious one: Horse_ebooks. There is freely available software for making such bots. The idea that it is a Markov chain bot is consistent with the other tweets from the account.

  2. It is basing its corpus on the PZFeed Twitter account

So, here's the tweet it's riffing on, from 2014:

AFP: Death toll from Nigerian mosque attack rises to at least 120 with 270 others injured http://tinyurl.com/og6lodu

enter image description here

The bot's software knew that an incident like "Nigerian mosque attack" could be replaced with another incident like "Paris terror attack", which the PZFeed account had used to describe the Charlie Hebdo murders. That's all.

The remainder of your question is whether this constitutes a "prediction". Human beings are pattern seekers and it's very easy to see some sort of mystery to this single tweet, but really, millions of silly things are posted to Twitter every day, and a robot-generated comedy account would seem to be about as far from willful prediction as it gets.

Another "psychic Tweet" which comes to my mind is the young man who apparently saw Benedict resigning and Pope Francis getting elected in his sleep. This tweet is exciting because out of all of the predictions constantly posted to Twitter, it came true. But there is no way to measure how true a Twitter prediction is going to be until it actually happens. Popperian science can only maintain a cool skepticism about predictions with no reliable methodology behind them.

yuritsuki
  • 1,965
  • 2
  • 15
  • 25
Avery
  • 44,313
  • 16
  • 183
  • 179
  • 78
    Looks like twitter is so big that the [infinite monkey theorem](https://en.wikipedia.org/wiki/Infinite_monkey_theorem) becomes applicable to news headlines. – CodesInChaos Nov 16 '15 at 06:43
  • 44
    There was also the case of a Twitter account predicting the last soccer World Series final result. What went rather unnoticed were the hundreds of similar Twitter accounts *which predicted every other possible outcome of that match.* :P – fgysin Nov 16 '15 at 07:29
  • 34
    @fgysin Actually, it was not done by a hundred accounts, but instead by a single account posting all possible options, and deleting the incorrect one afterwards. [See this Skeptics post](http://skeptics.stackexchange.com/questions/22345/did-the-twitter-account-fifndhs-predict-the-exact-world-cup-results-ahead-of-tim/22346#22346) for more details. – March Ho Nov 16 '15 at 13:13
  • 1
    Ah, I seem to have misremembered this. Thanks for correcting the mistake. – fgysin Nov 16 '15 at 13:18
  • @CodesInChaos funny, that's how Twitter often feels. Except instead of characters it's monkeys picking random words. – enderland Nov 17 '15 at 03:30
  • 2
    "Human beings are pattern seekers" http://wmiii.net/gallifreyan_embassy/always_seeing_patterns_12.jpg – Pharap Nov 17 '15 at 10:32
  • @accurate: What is the source of the image you added? – Oddthinking Dec 02 '15 at 13:21
  • And look at this Tweet: "BREAKING: Fake dollar bills were being thrown into Shanghai square where stampede killed at least 5 in Jerusalem synagogue attack." What does China have to do with Jerusalem? –  Dec 02 '15 at 14:07
  • Just for clarification, the bot's software didn't "know" that one thing could be replaced by another. It searched its database for words likely to come after "attack rises to". It's not as sophisticated as you seem to think. If I start a sentence "I ate a—" you will know that "banana" or "cake" are likely responses, "fly" is unlikely but possible, and "the" is not possible. Markov Chaining is basically that. – AmbroseChapel Mar 18 '16 at 00:16