The concept of humor is alien to the machine, but it sometimes produces very funny pictures.
Since May 2018, the publicity ” Abstract Humor ” began to exist by itself: instead of people, the creation of content in it was done by the neural network. It combines funny pictures from popular communities and publishes them on the page.
The creator of “Abstract humor” is the 20-year-old Russian Artem Chirkov, who studies in Canada. talked to the programmer about how he automated the work with the community, what is waiting for the neural network in the future and learned about his other projects.
Thanks to the neural network, public administrators came out “to retire”
The first version of the bot Chirkov wrote in 2015 in Python, when he was just beginning to learn programming. The algorithm was able to download pictures from popular communities and combine them, cutting off one of the lower parts, and the other – the upper one.
I showed my friends, they liked it, and somewhere in a week we had a bot that constantly stuffed the public’s offer with cut pictures that we selected and posted the best ones.
In 2016, the student finalized the system. It began to represent a few scripts: one was designed to search and crop text in pictures, the second helped avoid too often accessing “VKontakte” via API, the third kept pictures from public, and the fourth contained the main code of the bot.
Despite the innovations, the system could not take into account the context of the images, so they still had to be selected manually. According to the publisher, because of “viewing meaningless memes,” the moderators faced “burnout.”
In addition, the bot of Chirkov was focused on memes with black and white font Impact, and they began to go out of turn. Then he began to look for ways to fully automate the public.
The public was in a half-dead state, until in May 2018 I found a way to put the neural network on the recognition of text and the matanalysis to filter the results. Now the public works completely autonomously, the admin team does nothing for his work, we enjoy a well-deserved pension.
Since May 2018, the neural network of “Abstract Humor” once a week collects information about new memes from the publics “MDK”, “Lepra”, IGM and “Eaglet”, analyzes them and stores information about the zones with the text. After that, every day a script starts, which glues the zones and plans eight publications. The developer also instructed the neural network to check the result, manually selecting a thousand “suitable and unsuitable” images.
Chirkov said that the structure of the neural network was not created by himself, otherwise “he would not have mastered such a volume of work”. He used the finished development of a team of researchers – a neural network, which can determine and evaluate the content of the text.
According to Chirkov, the difference compared to the old approach is “huge”: without a neural network, half of the pictures were “with unreadable text,” and now about 5%. However, the updated system has its drawbacks.
The concept of humor is alien to the machine, so it’s nice to observe the public when two unrelated pictures are joking. If this happens, then it becomes both ridiculous and scary at the same time.
Chirkov believes that his public is a clear example of future automation. According to him, even if such a sphere as humor can be “at least somehow automated”, then everything else is possible.
We are the vanguard of robotization. The program has displaced me from the post of administrator, so the robots will gradually displace us all from work.
According to Chirkov, the bot has its own understanding of the context: it analyzes secondary factors such as position, size, color, and also knows how to separate the sets from panchlins. “In short, he is trying,” – explained the publisher.
Creator – 20-year-old Russian from Canada
Chirkov said that he was born in Perm and was carried away by programming before 18 years, and the last two years he spent in Canada, where he studies for a programmer.
On the question “where are you from?” The answer is still Perm, but I probably will not stay here to live. Rest here is good, Russian culture is the best, our boys are the brightest, but it’s not very good to work here.
According to the student, until his adulthood, he always wanted to become famous and become a “boy-genius,” but since then, maximalism has “lost.”
Programming was for me a way to combine something that I like: computers, games, the Internet and something that brings money. When I finally realized how to move from that dead point – this is when you know how to program the solution of mathematical examples or basic algorithms, but do not know how to apply knowledge in practice – then I had a huge motivation to continue programming.
Chirkov recommended the programming language Python to all novice programmers, since it was on him that he was able to “finally do something useful”.
If I could give advice to novice programmers, then I would say to think about some area where this skill can come in handy, games, websites, bots, anything, and learn programming on simple projects, otherwise there is a risk of losing motivation forever.
Another project of the developer is a service for searching for acquaintances by subscription “VKontakte”
Chirkov considers his biggest project “Abstract Humor” (and the second public “Even more abstract humor”), but this is not the only thing that the developer created on the basis of VKontakte. The other work of the student is a service for finding people who are close in spirit on the basis of subscriptions to VKFinder groups and publics .
According to Chirkov, human subscriptions often accurately reflect his personality and interests. In addition, the developer’s friends often confessed that they found themselves a friend or couple in some kind of public, which prompted him to create a service.
VKFinder launches the search right in all public places and finds people with similar interests in the same city or country where you are. Strangely enough, but it works, many people there have found people with whom they are already familiar in reality, that is, the points of contact really are. As an experiment, I also met through him with a man, played indie games and took to my group on DD (Dungeons & Dragons, – note ), where it is usually very difficult to find people.
According to Chirkov, due to the fact that the service does not require reposting, only a little more than 600 users use it. In addition, on the day the project was launched, VKontakte prohibited links to it as unsafe, so next time the developer plans to use the platforms of other social networks, for example, Telegram, Twitter or Facebook.
The future of neural networks
While neural networks can not stably generate funny pictures, but, according to Chirkov, it’s only a matter of time: over time algorithms will learn to understand the context and “progress will go.” The developer recalled that there already exist similar projects: for example, a neural network of students from Stanford, which generatesstrange memes based on old templates, only occasionally producing ridiculous results.
Neural networks are constantly developing, very promising sphere, I do not think that we are separated by a chasm from machine humor. One day it will happen, I’m sure. And even more so if neurobiologists or psychologists deduce an approximate formula or algorithm of humor, then they will immediately follow the implementation of this algorithm.
The best memes from the neural network
If you do not see the entries from “VKontakte”, try reloading the page.