The Painting-guessing Song That Hits The Screen In Moments, And The Soul Painter’s Skills For Clearing Levels

Some time ago, the "Guess the Painting Song" mini program became very popular in the circle of friends. It was like a painting assessment with the participation of all the people. It was really like this.

When you turn on your phone, all you see are the "win streaks" shared by friends, or the "soul paintings" presented after failure.

The gameplay is actually very simple: you are given a word, such as "strawberry" or "Mona Lisa", and you must draw it within 20 seconds for the artificial intelligence to guess correctly.

It sounded like it was a "you draw and I guess" game at an offline party, except that the person opposite who was responsible for guessing the drawing had turned into a somewhat "picky" AI.

When I tried it for the first time, I was confident and full of confidence. However, I was stuck in the "envelope" in the first round. The feeling of frustration was really ridiculous and indescribable.

Gradually, I noticed a certain change in the style trend in my circle of friends. At first, it was a "show off skills" trend, but later it evolved into a "large car rollover scene".

There are cases where someone drew a speaker, but the AI ​​guessed it was a table; another person drew a dragon, and it turned out to be an abstract pattern that was incomprehensible.

While everyone complained, they couldn't help but start over again and again. This feeling of love and hatred mixed with it is probably the most attractive part of this small program.

From "showing off skills" to "rolling over", the style of the circle of friends has changed dramatically

At the beginning, those who could easily draw a carrot or pass the level by drawing a strawberry were all very proud to post screenshots of their consecutive wins, and they were eager to tell everyone in the world that they were super talented in drawing.

But soon, this sense of superiority was shattered by the cruel reality.

Netizens who originally thought that "my paintings are quite similar" have posted their "failures" on social media, with words full of helplessness and confusion.

For example, there was a netizen who drew a sheep with a super high level, but the artificial intelligence failed to recognize it at all; there was someone who drew a three-dimensional sound object with rich and detailed details, but the artificial intelligence gave a wrong judgment.

What's even more exaggerated is that when faced with situations like "Teddy Bear", "Animal Migration", and when faced with difficult topics such as "Dragon", many people's brushes stopped directly on the screen, not knowing where to start painting.

For a time, the circle of friends became a forum for "soul painters" to complain.

Everyone began to wonder if their painting skills had really deteriorated?

In other words, the logic of this artificial intelligence is simply not in the same channel as the aesthetics of us ordinary people.

The more similar the painting is, the more likely it is that it will fail? Revealing the recognition principle of AI

After experiencing countless failures, thoughtful netizens have noticed a kind of "metaphysics". Those paintings with casual graffiti and simple lines showing simple features are more likely to be guessed correctly by AI, while those works that put serious attitude and carefully crafted details have repeatedly failed.

What is the principle of this?

Does AI prefer "abstraction"?

In fact, behind this is AI The problem of learning mechanism.

The underlying technology of "Guess the Painting Song" is formed by relying on Google's QuickDraw data set for training.

This data set contains more than 50 million hand-drawn sketches, all of which are simple doodles from ordinary users around the world.

In other words, what AI learns is what are the core and most common contours and features that most people use when describing an object.

For example, when you draw a painting called "Mona Lisa", in the AI ​​database, what it remembers may only be a simple outline of a face shape and an arc that shows a smile. It is not the eyes and shadows that you have so carefully drawn.

So, if your drawing is too complicated and adds personal style, it will exceed the "cognitive scope" of AI, and it will naturally be unable to guess.

Secrets for soul painters to pass levels: turn complexity into simplicity and capture features

Once you understand this principle, the "secret" to pass the level will be readily apparent.

If you want to achieve good results in this game, you must not entertain the idea of ​​creating a flawless sketch.

The most critical step is to quickly extract the core features of the object.

Take "strawberry" as an example. You don't need to draw every seed and leaf. You just need to use an approximate heart-shaped outline and put a few small dots on it, so that the AI ​​​​can basically recognize it quickly.

For a difficult task like the Mona Lisa, the simpler the painting, the better.

An oval-shaped face, plus a smile line that seems to exist but not seem to exist, and two simple landscape outlines in the background. Generally speaking, this is often easier to achieve a successful result than depicting exquisite facial features and hairstyle on the painting.

Faced with such abstract words as "animal migration", there is no need to bother drawing specific animals. Just draw a few simple arrows and wavy lines. These symbols of "movement" may get passed by luck.

Keep in mind that in this game, the enemy you face is not an art teacher, but an artificial intelligence that operates mechanically only by referring to the "reference answer." Only drawing in accordance with its established thinking mode is the most effective response strategy.

More than just games: every stroke we draw has meaning

Although we often complain when we fail, the original intention of this small program is not just to bring people a sense of joy or cause them to lose control of their emotions.

Developers hope to collect massive amounts of hand-drawn sketch data through this gamified approach.

When hundreds of millions of users participate in it, every trace left by each individual on the screen becomes precious training data, helping the computer to better understand how humans draw and how to summarize and present objects.

Every time we "draw" seriously, we are actually contributing to the future development of artificial intelligence. Periods cannot be missing.

Not only does the machine no longer just recognize photos coldly, but it also allows it to start trying to learn humans' unique visual logic, and it also has abstract capabilities.

Therefore, when we "wail" ourselves in the circle of friends and fail again, we are actually participating in a very meaningful technological evolution process at the same time.

Looking back, "Guess the Painting Song" is very popular. On the surface, it is everyone's enthusiasm for novel gameplay. If you dig deeper, it is a collective release of curiosity about AI technology.

It is a relaxed and interesting way that makes everyone among us realize the charm of human-computer interaction.

The next time you play again, you might as well put aside the burden that comes with achieving the effect of a painted object, be bolder and use the simplest lines to express it, and enjoy the fun of competing with artificial intelligence for wisdom and courage.

After all, in this game, drawing "likely" is not the goal. However, drawing so that the AI ​​can "understand" it can be considered a real success.