As an author of AI vertical self-media, a large part of my work is to "lead the way" for AI, and then tell human readers that your work and that work will be replaced by AI.
For a long time, humans seem to be standing in a weak position without any help. Perhaps the future is like some pessimists imagine that we have AI drivers, AI salesmen, AI poets, but human beings have become poor beggings. By.
Until today, I first discovered that there is a work AI that can't surpass human beings for a period of time, and this victory belongs to the second element - the work that will not be replaced by AI, is a cartoonist.
- Even the comics don't understand, what about destroying the world? -
The reason why cartoonists don't replace AI is simple, because a professor at the University of Maryland conducted a study and eventually found that AI couldn't understand comics at all.
The above is a very simple four-frame comic. It is very easy for humans to understand: the kitten is thinking about the creative material, then discovers the puppy, asks the puppy to tell a joke, and the puppy says "you are very beautiful" leads to small The cat is very angry.
In fact, in the last picture, the puppy did not enter the mirror, and "you are very beautiful" was originally praised. It must be connected with the "joke" in the previous picture to explain the kitten's emotions.
For AI, it is simply too difficult to understand the information presented outside the picture.
In an experiment at the University of Maryland, the researchers built a data set of 1.2 million comics and extracted the text from each frame. Using the LSTM model, I hope that AI can have a coherent understanding of the comics. .
About LSTM (Long-Term and Short-Term Memory Network) has been introduced a lot. The model is characterized by the concept of memory, which can process and predict longer elements of time series. Although it performed well in long texts, machine translations, etc., LSTM was completely defeated in watching comics.
After a lot of training, the researchers gave AI a set of comics that they had never seen before, asking AI to understand and predict the text information or the content of the picture in the next picture. As a result, the results of AI were a mess. The correctness of human predictions usually reaches 80% accuracy.
- Visual narrative? Why is it difficult to artificial intelligence?
Strictly speaking, comics are a kind of "visual narrative" - ​​hiding information in images. The same is visual narrative. It is easier for AI to understand movies than to understand comics. The protagonist of the movie is human, and the face is long. It is very simple to train AI to read facial expressions and recognize emotions. Not to mention the movie will have a detailed script.
But the biggest feature of comics is that they are not visually coherent. Like the four-frame comics above, the third picture puppy is still in the picture, and the fourth picture is gone. Humans can quickly understand that a puppy named Calming Dog will leave with a calm sentence. For AI, reading such information outside the picture and text is really difficult for artificial intelligence.
Secondly, the different comics have different painting and narrative styles, which is also a difficult point for AI training. In a simple four-frame comic, the scenes of each grid are the same. In other comics, this may be a fighting scene, and the next one is an angry face. I can understand the four-character comic AI, and then look at the comics with a lens switch, which is also a fog. As for the style of painting, different cartoonists have very different ways of depicting faces. When they are replaced by AI, the difficulty will increase.
Another point is that the visual narrative is based on the concepts of "logic" and "common sense." For example, the kitten said a joke, the puppy said "you are beautiful", understanding the plot requires the logic of "you say that I am a joke = you say that I am ugly". For example, the common mouse stalks in the Doraemon also need the basic knowledge that "cats are usually not afraid of mice." These things are simple for humans, but AI does not have these common sense and logical concepts, and we can't instill these concepts into the brain of AI like the encyclopedia.
- The big hand in the world of chess, it is also an episode in the rumor -
In fact, combined with the victory of AI in the field of Go and the failure of comics, we can see that the performance of AI in the field of complete information and information is completely different from that in the field of incomplete information.
Complete information is originally a part of economics, meaning that participants can understand all the information of the entire market. Here, we can look at the data set of a job. In the work of Go, all the information can be summarized into a data set: the rules of the game, the play of each step. In the comics, we can at most put the detailed annotations on the screen and extract the text information. The logic that can be hidden from the picture text, common sense, etc. can only be said to be unspeakable, and cannot be provided to the AI.
The worst thing AI does is read between the lines.
By analogy, AI plays a dark game in drama, after-speech, and four-nation military chess (a kind of flagship game that includes deception), including love, and so on. It is full of incomplete information, deception and anti-spoofing, interpretation of images, common sense and logic. The performance will not be too good.
In this way, AI is a bit like the original Tomoko in the "Three-body". It does not hide its own ideas, nor can it understand the concept of concealment and deception. Therefore, we really don't have to be afraid of the victory of AI. It will be the worst-known colleague in the office and the passer-by in the harem. The outstanding ability of one can not make up for its incomplete information. Short board. What's more, imagery, analogy, irony, and metaphor are the best tools that human beings are good at.
I believe that the best way forward in the world must be for humans and AI to do their jobs and do what they do best. They are still able to provide a lot of help to humans in things such as visual narratives that AI is not particularly good at.
For example, using the generation of anti-neural networks to create characters, using supervised learning + convolution networks to color lineups, and even developing a comic reading app that automatically magnifies text. These are not illusions, but the reality that is happening. When these complicated mechanical labors are done by AI, we can also invest more in what we are good at: use the incomplete information environment to tell more stories and keep the world as interesting as it should be.
Transfer from: brain body
Scented Candles,Candle Fragrance Oils,Vanilla Candle,Sandalwood Candle
Jiangsu Raymeel Home Decoration Co., Ltd. , https://www.raymeelhome.com