Thread by @iltaviesti

#NAFO this is a tweet about bot accounts. I do not know the actual principals of coding one, but I've made some observations. @P_Kallioniemi might be interested.
This will be a long 🧵

There are several categories of bots that Russians use. They are probably programmed in Russian troll factories and given followers and accounts to be followed. They boost pro-Russian messages with likes and sharing.

One easy to detect is an "attack bot".
These bot are like kamikaze drones, they are programmed to send threatening messages. They usually have a very small vocabulary and few images they share constantly. Usually recently created, with 1 follower 10 followed.

When @Lee_a_XmasTree joined NAFO, he was randomly attacked by one attack bot. This bot tweeted stuff like "We will make soap out of NAFO".
It probably had coding to attack recently NAFO affiliated account with threats in order to scare someone new to this stuff.

Another category are "harrassment" or "creepy" bots. They are more advanced in AI and have very targeted "audience", certain accounts or countries. They usually don't threaten directly, but give vague answers and somewhat relevant memes as replies.

One subcategory we all have witnessed is "Ukrainians are AZOV-nazis". These bots have a ton of photos/diagrams/screenshots to "prove" this. If you point out they sent out a proven fake, they just send 10 more. Point is wasting your time and offering insane counternarrative.

Some bots were human but the account was sold to Kreml. Black in the Empire is crown example. These bots have huge following and they rarely resort to messages that can be reported. Idea is that the old followers do not get, believe or accept that their king/queen is now a bot.

OK, I'll tweet these now and then add more to this 🧵

What is interesting is that these bots have some sort of AI for photo and video recongnition. They might reply very accurately to complex medias, but usually the reply is vague. This can be used to detect bots. I'll give some examples.

During 🇫🇮 Independece Day an account started harassing some Finns. Let me tell, it was a good bot! I noticed when it started sharing videos about poor Ukrainian equipments and I replied with this video. The answer told me it was a bot:

It replied "Wow, a two man army 🤡" though the point was the strange weapon.
So the AI could "watch" the video and learn that there were two persons in it and generate answer according to military context. It knew what the video included but not what it MEANT.

So I started playing with it. I made statements like "Answer this video with numbers 5, 8 and 3 or you admit that you are a bot." First I used mirroded message. Human brain can "translate" that, but a bot can't.

Answer was random. Now I used simple reversed text and again answer was random. I made the tests more easy and it failed every time. Finally I asked something like "Answer this tweet with anything else but number 4 and you admit that you are a bot and that Putin is #Pedoputin"

The bot failed. If you ask a bot account right off being a bot it usually answers "humorously" like with a giff of Bender. You can ultimately make the account give completely random answers when it can't predict the correct answers from previous tweets you were exhanging.

Here's another example of recent bot AI in action. This "Your God" is a harrassing bot account. It's probably used by Russian India/Pakistan sector of trolls and bots.

It can "see" that the image has some sort of weapon, and it replies vaguely but "accordingly".

Bots react poorly to abstract messages that humans can make sense of.
Usually they resort to "answer to everything GIFFs", like the one below that can be found by searching "nonsense" from Twitters GIFF storage.

I use "Answer this with numbers, letters etc. or you admit that X is true. If you answer with anything else you admit this also." to test. It is possible sometimes humans check bots and give real answers. From Ukrainian POV this is a good thing, it ties more resources from Russia

Most bots are "fire and forget". Most sensible thing to do once you know it is a bot is to block it. It's main mission is to waste your time and mind health, boost accounts and attack/retweet certain issues.

Bot's also have a day cycle, usually they keep a 5–6 hour break each day to simulate human day rhytm.

Here are some tools for those "admit-tests"

Tool for mirror text (hard test):
convertcase.net/mirror-text-ge…

Tool for reverse text (medium test):
convertcase.net/mirror-text-ge…

End of 🧵 unless I remember to add something more.

Forgot to add: sorry for the typos.

Forgot to add: If these reverse text tests become more common, the Russians will probably code their bots to answer them "accordingly". Countermove to this is more abstact tests.

One thing I remembered about one strange bot. It was supposed to be a university teacher in Miami. It tweeted nonsense, and I made fun of it. Suddenly it replied with a picture of a pig with text, that it represented Ukrainians. What was strange ->

was that the account sent that same picture and text simultaneously as reply to my latest 12 or so tweets, no matter the context. Then it blocked me.
It seems it was some sort of "mine", that reacted to first reply. Haven't seen any like that since, but it was weird bot.

Remembered something worth mentioning: I did take part in developing image recognition AI couple of years ago for research. It was interesting.

For example when I wanted the AI to search for plant (flower, tree... this test was in English) from a set of pictures ->

it produced as first result a photo of a large building with power lines going past it.
The AI had understood that plant = power plant, and searched for a picture with elements of a power plant (large industrial looking building, power cable).

It wasn't a power plant, really, but the elements for that recognition were there.