Explore the 2022 AI trend.Can you really make a story in the machine? --Wirelesswire News

In this industry, it is often said that "if AI takes human work, it is the AI researcher who is the first to be deprived of the job."

In fact, when breakthroughs occur in the AI field, it often seems that their research has almost been wasted.In the past, it was a year -on -year that was disappointed that it was wasted, but now it has been shortened to quarterly.

In addition, it was common sense in the AI industry in recent years that relatively shocking news caused relatively shocking news during the year -end and New Year holidays, but this year's year -end and New Year holidays seem to have few impactful announcements compared to last year.rice field.

At the end of last year, Openai announced AI to support Web browsing, but did not talk much.

In some indicators, despite the achievements that exceed the GPT-3.Openai also announced Dall-E, which can freely generate images from words a year ago, but since the implementation was only partially disclosed, volunteers from around the world who were dissatisfied with it independently Dall-.The development of E clone implementation and pre -training models, which finally became the shape.

Depending on the point of view, Openai's concealment has led to a year of understanding of the world as a whole, but from another point of view, the discovery of Openai in the past year is the world's general researcher level.It can be said that it has penetrated.

Until a few years ago, it was a big topic just by saying that "image recognition has gone significantly more than the past", and "combining image recognition and reinforcement learning is thought to be impossible until the past.I was able to make a complicated decision. "It is the era of AlphaGo, Alphazero, and Muzero.Some said, "AI to manipulate words more than human beings appeared."It is a major revolution in the natural language field of the technical called Transformer, such as Openai's GPT-3 and Google BERT.The leading role of last year's topic was Dall-E and CLIP, which is clearly applied to the GPT-3.

Furthermore, several new techniques with the performance of Transformer are tested, and Clip and Transformer have been tested other than sentences and images, and have developed into voice recognition, voice synthesis, video recognition, and video generation.。Recently, there have been studies that directly create three -dimensional objects from words.In addition, research has begun to have a faster result (like deep learning), which is more complex mathematical formulas (like deep learning).

The interest in the research world is, "How far can AI can do?", But the theme of "what can do with AI" has not yet been delved into the industry.

There are two causes.One is that the understanding of AI has not yet penetrated the business person in general, especially among business people, which are particularly working in real business.The other is that a business person who once learned the power of AI is trying to conceal its effects and methods with all his might.

This is a completely correct adaptation in capitalist competition.

In the past, it was called "know -how", but now it is just a model (AI) or (AI) architecture.

Of the research at universities, "practical" is being steadily eliminated, and the rudder has been turned into more academic directions, while the AI social implementation is gradually spreading under the surface of the water.This penetration is asymmetric and non -continuous.

In other words, a company that seemed to be old -fashioned until yesterday has been able to get overwhelming competitiveness without anyone known to anyone.

What is the meaning of this?

For example, let's say you are doing a rice shop in a shopping street in front of the station.There is a rival convenience store on the other side of one, and there is a large supermarket a little away.You have to survive in this world with a few capital and a limited store area as a weapon.

Supermarkets and convenience stores can purchase vegetables in low and large quantities in the background of great financial strength.Moreover, high -quality brand rice can be purchased from a farmer with a dedicated contract.

If you are a business genius and you can prosper your own rice shop with a completely new idea, that's great.However, you have retired when you are old, and your young son who gave up the store doesn't seem to be interested in business and his parents.

In such a case, what if Doraemon appeared and "TV where you can see the news after a week"?

Of course, no one else has this.A week later, if a mysterious pathogen can be known faster than anyone else when the stockpiling of rice in Japan is depleted, we will consider a large amount of rice inventory now.right.One week later, a rally riot occurs nationwide.Will the son who made a lot of money pride to someone the existence of this TV?

If you are not smart, you may be proud.But if you have a little wisdom, you shouldn't fix such a great tool to others.

This "great tool" was "information" in the 20th century.The information that only you have, hidden information, it is still worthwhile, but in the modern age, AI that insights "future information" has value.

It may sound like a rush, but I want you to think about Go and Shogi AI.Or mahjong or poker AI is fine.What AI is necessary is "future prediction".

The simplest poker in this is to predict what will happen if this card is cut.The reliable prediction is given to some extent equally.A combination of your hand and your hand.And the expressions and light mouth of other players.

What it means, whether it is strong, natural, and the most likely future from limited information.

Of course, what you can do is a separation.However, AI can be seen in a short time to the point where humans can never think of.

To beat the poker, just predict only a minute later.In other words, there is a "TV where you can see the news one minute later".Go or shogi would be enough to see the news at least 24 hours later.

The use of AI is that this is "one week later", "one month later", "six months later" and "a year later".This is not just about the prediction of numbers.It is a story about which branch to grab from the countless future led by the chain of judgment.

If you want to know the image of future predictions by AI in about two hours, you can watch the movie "All You Need Is Kill" starring Tom Cruise, the original of Hiroshi Sakurasaka.

In this work, the hero who plays Tom Cruise repeats death many times.When you wake up, the same daily life is repeated.Just like repeating the game.Try various options and grasp the possibilities of a variety of options.

Similar productions are in the Doctor Strange in "Avengers/Infinity War" and "Avengers/End Games".The Avengers, who have been desperate, will do their best to play a thread of the possibility of victory, only one 14 million, one 14 million, which Dr. Strange has seen.

And if you can know the future, you will definitely be more advantageous than strangers.AI cannot predict a perfect future, but can present the best options at the moment, calculated back from the future prediction.

Unlike many misunderstood facts, the options indicated by the AI in this case can show the path to the future as many times as humans fall.

Even if AI presents the best options against his intuition, it will explain the path to the future of "why?"

In Star Wars, R2D2 is drawn as a very smart robot.Looking back on the movie nearly half a century ago and a little dissatisfied, the current AI is smarter.

R2D2 is very wise, but when Luke is distressed in Episode 5, the C-3PO tells Luke's best friend Han Solo, but ignores his solo.

With the current AI, you should be able to draw a map, "Which place, and what probability you are surviving?"This can be easily created by the method called Bayes estimation.

And the "map of survival probability" should be more accurate the more accuracy of the search activity, and at least the location of Luke (aside from life and death).

The search for the American nuclear submarine Scorpion, which was distressed in 1968, was done so.And only one submarine (unfortunately it was that remnant) from the wide Atlantic Ocean was found.

This map has many parts that lead to the current concept of AI.In fact, AIs such as Go and Othello constantly update something like a "victory probability map" and predict the future.

In this field, discussions are no longer on a different stage.In other words, it's not "do or not" but "which is good?"This is the largest business know -how in modern society, and the better, the more you will not be able to introduce the details as an example.It's a problem if you know your rivals.

Now, based on this background, the purpose of this article is to think about what AI can't do yet and what you want to do.

Last year, the "AI drawing from words" was certainly shocking.It was shocking, but it looked like half or part of what you needed.

The picture is certainly interesting, but just interesting.From that point on, I began to use the illustrations of my blog and articles on the AI drawn.I no longer need the myriad of irritable photos taken by stock photos and stock photos.Obviously, the number of cameras has been reduced a few years ago.

Previously, a slightly nifty photo used in your own presentation was to take it yourself.So, in my "Planning / Presentation Course", I mentioned a single -lens reflex camera as a "tool to buy".

How about now?You can now take good photos with your smartphone camera.However, more, the paintings drawn on AI have become easier to get what they want.

For example, let's write a manuscript on the theme of "Human dawn".This is the abstract painting drawn by the latest AI under the theme of "Human dawn".

Of course, this is the result of choosing what I liked among some output results, but other works seem to be a dawn of humanity in some context.

However, these pictures are not the text.It is just an illustration attached to the text.

It is an attached item to convey the image of the sentence more accurately, and if you only present this and get a manuscript fee, you will be angry at the editorial department.

Regardless of whether it is useful for readers, at least the manuscript that the editorial department needs is accompanied by a story.At present, interpretation of the phenomena that is happening, the next action that comes from it, or hints, are required for all manuscripts.

When I tried to draw a picture from the word "story generation device" in the same AI, this came out.

Indeed, it's definitely a device.However, it will be difficult to interpret this picture as a "story generation device".It is exactly an "illustration".

So why not add a little more explanation of this word?For example, like this."AI to create a story | Can you write a story in a machine? Can the story impress humans? How far can you do a machine that doesn't understand love and emotions?"

Then it was such a picture that came out.

Indeed, I think it's quite tough while drawing it yourself.There are things like a flowchart, something like a two -frame manga, and something that humans have been incorporated into the center of the machine.

However, it must be noted that this is a kind of fantastic landscape that AI, which has learned the relationship between human language and images, is created with poor knowledge.

This machine (AI) does not create a story.For now, for now.

As a matter of fact, it is not so difficult to generate a story in a machine.This is because the nature of the story is often "a contradictory description of the event."

I don't even need an imadoki AI to make this.

In the book I wrote about 20 years ago, I introduced a "love simulator" program that models the personality of multiple people and has a chain reaction, such as fighting and falling in love.

It is a program that is not surprisingly unexpected that a human emotional model is interacting, and the same thing is the same thing unless the theme of handling is human emotions.Life), it is also held in BOIDS of Craig Reynolds.Recently, it is the same as an infectious simulator.

This certainly can do something like a "story".However, it is a different matter whether it is a "line of events", whether it is "interesting to read".

Here, it is necessary to divide the structure of the "story" into two.Let's call it "facts" and "truth".

One is the "line of the event", a record of the chain reaction.It can be paraphrased as "facts".

However, in terms of one fact, if the narrator felt, if it was called the "truth", if the "truth" was introduced to the generated story, the author's position must be clarified.

Who is the author, or is it from the god's perspective?No matter what position the author is, the most important thing is to imagine the audience's feelings.

Otherwise, it will never be an "interesting story."I think there is a problem with this part.

In other words, even if it is easy to generate "facts (construction of inconsistent events)", it is not easy to think of the method of sublimating it into "truth" that sympathizes with readers.

Even so, the only possibility is to use the typical illusions often used in tarot fortune -telling, gambling, movies and games in the past.

Humans find meanings without permission, even if they look at a line of unrelated facts."Don't get caught by a traffic light from a while ago." "Don't look at the numbers 5 today."

In psychology, this is called "montage effect".

Machines can easily post a row of inconsistent facts without intentional (pursuing the truth).Humans can get the illusion that the list of facts shown "can be seen" like a story.

This is a game called "SIMS" by Will Light.In this game, the player becomes a god's point of view, creates a room himself and lives a "sim" in a human figure.

The shims live, fall in love, quarrel, and leave.Players can intervene in the Sim's life as a god, but can also see the shim's life without intervening at all.

In other words, like the goldfish in the aquarium, they look at the behavior of the shims.

In order to interpret these shim's behavior and express the production that sympathizes with the audience, it seems that the current AI is still unsatisfactory.

More specifically, the ultimate form of AI should be a machine that automatically generates interesting movies, comics and games that enrich the lives of the users.

If you think of it as a consumer AI goal, it seems that there are still many issues to be solved.However, I guess the path of the research was visible until last year.

Since the GPT-3, people have forgotten that the image recognition has had a great breakthrough, and since Dall-E, only one picture was automatically generated.

In the future, we will have to face AI from a different perspective.Not only is it convenient, but it is a secret that human beings have fundamentally demanded, such as "what is a good life", "what is a rich life", and "what is interesting".

Now that people around the world have a "painting AI", the next step will be AI to make a story in a true sense.Interestingly, more and more researchers who specialize in "AI are specialized" should be in trouble.

Before doing AI research, AI researchers will be forced to study what intelligence is and what is human.And it is not an information science or engineering, but rather a literary and artistic world that has been asking the question.

There is a "mechanized intuition" brought by deep learning beyond what can be explained in logic, which is essentially a science civilization that has been pursuing the "world that can be explained by logic" for tens of thousands of years.I think it's a sign to do.

The understanding and value of the wisdom of the other mankind who have been pursuing the "world that cannot be explained by logic alone" for tens of thousands of years and separated from science on the way, that is, culture, culture, and art is re -understand.Isn't it evaluated?

And unexpectedly, at first glance, "the distinction between humanities and science is only in Japan", it seems that the disadvantage of the fusion of literature is suddenly reborn as an effective means.

I think this year will be the beginning of such an era.