Mechanical Shakespeare: Are machines capable of literary creativity?

It was night, the lights of the Boryspil highway flew past the windows of a taxi. The driver turned off the music, which was unbearably pressing on my brain after a hard flight, and began to speak in order not to fall asleep.

First, of course, about politics, “brought the country”, and everything like that, then about something personal. I also did not want to switch off right in the front seat, so I tried to listen to him.

“... And then the end will come to all of us,” fragments of the phrase reached me. - Or rather them, not me. I securely insured. When they are all: taxi drivers, minibuses, even trams thrown out into the street, I will not be there. I will sit in the warmth, drink coffee and laugh out loud.

“Why-why are they thrown out?” - I asked sleepily.

“Have you heard about Uber?” What they do with the drivers is only a rehearsal, yes. Soon, very soon they will launch their autopilot. It will be cheaper, safer, cooler! All these mediocrity is waiting for work on the construction site. Or bomzhatnik. But not me, I'm smarter than them.

- Yes? - I rubbed my eyes.

- Yes! I started writing a novel! Extreme! You know, I have here, ”tapped a finger on his temple,“ there are so many plots sitting around. I drive it to different people, each blurts out something. And I have a memory since childhood - God forbid everyone. Who I just did not go here: a transvestite who has already performed three operations cannot determine everything; an elite prostitute - all the way around chattering a Sumerian cuneiform; the lover of the minister of one was driving, he told so many things ... Names just change and write - there will be a blockbuster!

-BUT ...

- The only thing that I am writing for now is not very. I downloaded a couple of textbooks, read after work. Every evening for training I write down what I heard during the day. By the way, can you tell something interesting, but today, for some reason, the golyak is full. I can even make a discount.

- What I have is interesting ... I am a programmer, I am flying here from ...

- Hey, programmer, - the driver interrupted me, - you can not tell about your life. It is the same for all of you, uninteresting. Tell me another is better: these robots, which learned to drive cars, will they start writing novels soon? Honestly. I will not be offended, do not be afraid. Just if soon - in my plan there is no point, you need to go to the programmer then, then to repair these robots to someone else, set up ...

I, frankly, confused. On the one hand, as far as I knew at that time, it seems that no algorithm has yet been able to write even an approximately meaningful text, let alone an artistic one. And on the other hand, to assume now such responsibility for someone else’s life, to say that the writing profession is safe, and then, a couple of years later, to wait for this taxi driver with a trim at his house?

“Come on,” I said to him after an awkward pause. I'm going to google, read, comprehend. And write an article. Maybe someone else will be interested in who has the same problems.

Then I paid off and went to bed after a hard road. In the morning, of course, I forgot about my promise. I remembered a little later, and there already tortured my conscience: I had to really write.

I hope, my casual friend, (I never asked what your name is), you believed the sleepy passenger and now read this article. And if not - to hell with you. Someone come in handy.

So here.

Pessimists of the past

At first glance, the task of writing artistic works in a software program looks simple:

1. Create a Super Mega- Artificial Intelligence .
2. Make him write a novel.

Well, no, you say, it's a robot! White plastic garbage! He only knows how to execute commands, where to create it? He does not understand either human emotions or motives, what can he write about at all? About electric boilers?

With this opinion, you will not be left alone. Also thought one very smart woman: the daughter of a famous poet, and at the same time the first programmer in history - Ada Lovelace.

She argued that a computer, even a very powerful one, is not capable of creativity, or even of what we call "intelligence." After all, any machine is just a thoughtless performer of the algorithms embedded in it by man. That is, even if the computer ever writes a poem or a novel, the copyright will still belong to the coder who programmed it for it.

Of course, we can understand it. Before the eyes of the noble lady was an example of the father, the famous rake and rowdy: only such a temper of character can have a real poet. If her favorite, logical and accurate machine is doomed to become something similar - it is better to destroy it right now, so that later you do not have to send troops to the past.

On the other hand, its key argument about “the program is only executing predetermined algorithms” has recently been cracking at the seams.

It all started with the fact that hundreds or even thousands of people started working on complex software systems. And programmers who understand what absolutely every function does are usually not found in large projects. That is, due to the cooperation of different people, the programs are already one level higher than what one person is capable of creating. Imagine a novel with a thousand authors. That's the same thing.

Further more. The latest achievements of machine intelligence are based on algorithms that no one has ever written. Programs create them themselves, based on data that they are fed in incredible quantities. It got to the point that the question "how does the program work?" Its creators just shrug. Back to human knowledge, mysterious zeros and units are simply not converted.

Ada's successor, Alan Turing, was a little more optimistic. He even devoted a few lines to proving that an immortal soul can live not only in a human body, but also in any other vessel. For example, in the fighting humanoid robot. For such passages he was not much loved in his homeland, in conservative England.

Even though hacking Enigma, he seriously influenced the results of World War II, gender preferences turned out to be more important for the public than scientific achievements, and in the end the “winning people” brought yesterday's hero to suicide.

Programmers Turing is known primarily for its “machine”, on the basis of which all existing programming languages ​​have emerged, but popular culture is more familiar with its other creation - the famous Turing Test.

The opinion that you can assess the “intelligence” of the robot simply by talking to it captured the minds of leading science fiction writers of the middle of the last century and got so tired of everyone that when a test was passed a few years ago, no one noticed at all.

The fact is that specially trained lexical generators, in addition to “pretending to be human,” are actually not able to do anything. And for everyday problems, systems that did not try to copy human behavior turned out to be much more useful, but simply solved the tasks assigned to them.

By the way, a similar revolution occurred a century earlier in automobile and aircraft construction. As soon as the designers stopped trying to create a “metal horse” or a “wooden bird,” they immediately managed to build mechanisms that turned out to be much more useful than the products of blind evolution.

The man, too, to be honest, is so-so a computational mechanism. Being designed for a specific purpose - survival among wild animals and their own kind, the human body in some areas hopelessly lags behind the most primitive mechanisms. What is the point of trying to imitate what can be done better?

That is why, after Turing, the concept of “artificial intelligence” settled only in science fiction novels, while programmers set about solving specific problems facing the economy.

Missiles flew into space, stocks were traded on the exchanges. Computers successfully calculated the optimal train and aircraft schedules. While the romantics were reading the rules of robotics, simple workers of the reel and punch cards introduced algorithms of varying degrees of complexity into tube machines.

It would be logical to assume that the problem of writing literary works can be solved in a similar way. Like, let's write out the algorithms according to which a writer or a poet makes words from letters, throw them into the mouth of a computer and voila - keep a new poem. Or a poem. Or a line of text at least, no?

Out of habit, programmers turned to experts, expecting to get a clear set of rules for business logic. Once they also went to physicists, railway workers and even biologists. The latter were so-so friends with mathematics, but among them were many physicists, with whom it was easier for programmers to talk with.

But with the philologists came out full of seams. Formal features of the text, they said, of course exist. Poetic size, genres and subgenres, styles of speech and writing, after all. But if to place the stress on the machine it is still possible to teach (after a long prescription of examples manually), then the philologists refused to formulate the features of Baudelaire in mathematical terms tightly.

Like, in any figures the real art can not be driven. Do not need! For then who will pay salaries to us, philologists? Go short, dear programmers, calculate the trajectories of nuclear missiles. We will need much more in the near future. So they went.

For a long time, the plot was the only component of an artistic text that was somehow suitable for algorithmization. Indeed, in most cases in the work were the main characters who performed certain actions. For example, they moved from point A to point B. Just like rockets, the programmers thought and took on various plot generators .

Some, frankly, even gave good results. Someone managed to arrange them into whole books. Only here, Ada Lovelace, having critically squinted his eyes, would have noticed that such programs are nothing more than variations of the “plot hat” when the author writes on the pieces of paper first the names of characters, then events, and then in turn draws out pieces of paper, describing in literary form what happened and with whom. And if there is any creativity in this process, it is solely on the part of the author, who writes on these leaves. And from the side of the machine there is only one random number generator.

They added fuel to the fire of mystics: creativity, they say, is a meditative process, a kind of communication between the writer and spirits. If there are any rules, then only in their etheric world. Until we teach our computers to connect there via HDMI, no creativity from computers should be expected.

This position is also shared by the authors of the "creative writing" textbooks. Most of the tips there are about how to get yourself to sit at the table, relax and start “ listening to your broccoli .” After she dictates the text to you, all you have to do is correct the mistakes, throw out all adverbs and send the manuscript to the publisher.

Well, the method may be effective, but it is absolutely unsuitable for writing a program. After all, to program something, you need to know how this something works. Is not it?

In fact ... not really. But more about that later.

World as rules and representation

If you dig deeper, the problem of formalization arises not only in the literature. The philosophers of ancient Greece also loved to define everything that they encountered on their way. At first, it worked out well. But it turned out that the greatest problems arise with simple concepts that are understandable even to a child. Good, evil, beauty, deformity, virtue, vice, God.

After several anecdotal cases like “man is bipedal without feathers”, Plato was tired of working on definitions and decided that the real rules of this world, the so-called Ideas, exist somewhere in a parallel world, and only their shadows reach us. The human soul, having access to this “Ideal world,” can easily distinguish between them, and our thinking, limited to the material world, cannot understand this greatness.

Talk about ideas for a good wine, however, no one forbids. Although no, it still prohibits.

Wittgenstein, one of the leading philosophers of the twentieth century, once categorically stated: "What is impossible to say - it's better to be silent."

No, he was not an enemy of free speech. Simply speaking, he noted, in relation to our thinking, is frankly secondary. It can help in basic communication and even the transfer of some part of knowledge in printed form, but the main thing that happens to us is poorly converted into words.

The maximum that we can - is to give names to certain facts, elementary units of this world. In contrast to "concepts", "facts" describe what we see, without unnecessary generalizations. A specific table stands in a particular room at a specific temperature, atmospheric pressure, and level of illumination. There is no Idea Table in the room, and therefore there is no point in talking about it. The whole world around is just a configuration of atoms and waves into which the sickly human imagination tries to cram terms that it does not itself understand.

The child comes to the father and says: “Dad, I came up with the word karchapan. What does it mean?". Such Wittgenstein saw all the philosophers of the past. But the "facts" do not suffer from such problems. We already know what we see. It remains to be understood - why?

The language of the description of "facts" seriously claimed mathematics. In orderly series of formulas, they tried to explain everything: from the interaction of atoms to unhappy marriages. Only when the model became a little more complicated, the formulas immediately lost their elegance and turned into ten-story notations of tensors and “strange attractors” (the only term I still remember from the course of functional analysis).

Thus, it turned out to be even more difficult to translate the avant-garde of modern mathematics into Boolean logic than artistic texts. Moreover, the mathematics itself, as Gödel showed, was not so omnipotent. In short, a complete fiasco.

So complete that a part of the philosophers finally went into all serious things and created a strange teaching called “poststructuralism”. The problem of describing the world in text format, they decided very simply. The text, they say, does not describe, but generates this world. The original teaching of Wittgenstein is the opposite: if nobody wrote about this, it really does not exist .

Programmers would really like this point of view, because typing text into a computer is much easier than making millions of different sensors collect information about the world. Only here the postulates of the theory turned out to be so close to utter nonsense that, except for the creators themselves, no one, it seems, understood them. Therefore, after the death of Foucault and Deleuze, the current slowly but surely dissolved into a rhizome e.

Finally disappointed in the philosophers, the programmers turned to representatives of more mundane specialties. For example, to familiar biologists.

Be like a brain

The twentieth century, in addition to breakthroughs in cybernetics and ways to kill each other, also brought good results in understanding how people work. More and more processes that previously could be explained only by intercourse with spirits are now attributed to the brain. Before a complete understanding of how our mind works, it’s still very, very far away , but in basic things, like the structure of neurons, science is already well-oriented.

So, apparently it turned out that the famous Cave of Ideas is, in fact, just an area inside our skull. The brain, receiving information about the world from the sense organs, writes it to the “hard disk” in an intelligible form only from neurons and connections between them.

About the “reading” of such a record we are not talking yet, but we know that the key characteristics of our “cave” are: a) abstraction, that is, combining the features of various objects and phenomena under one “key” and b) constantly updating data based on the incoming information.

The data storage system of the "brain" type is several orders of magnitude better than that created by man so far. And the only, though very unpleasant, disadvantage is that, both in language and in mathematics, all our knowledge and all our life experience are converted very, very badly .

There is even a funny situation when a top-class specialist who is able to perform complex tasks cannot explain to students how he does it. The poor lads can only watch the Master and try to reproduce his every movement. How to translate it all into machine language, the question is not even worth it.

It would seem another dead end. But the scientists, scratching the back of their heads, thought. And why don't we remove intermediaries who are unable even to agree among themselves, and not release our programs directly to the outside world? Let them take - and learn! Is seductive, is not it?

The first attempt was the collaboration of Warren McCulloch and Walter Pitts in 1943. It is worth noting that none of them was a programmer. McCulloch, a neuropsychologist, a neuropsychologist, a philosopher, a poet, had a rather complicated relationship with mathematics. But it was compensated by Pitts, a wunderkind mathematician, whom Russell himself was delighted with. 15 , MIT, - .

. , , ( «») ( «»).

, . — , , . , , .

«A logical calculus of the ideas immanent in nervous activity» , . . , , .

, . , , - , , .

, : , . : .

. . ( ) 1969 «». , , , , , , «» . «XOR», , , , .

, : , , , .

, , . , : , , 80%. , 90-95%. , , , . , , « ».

: — , . «», , .

, AlphaGo . , , — .

. , «» , , . , — . — .

. .

, , ? , , ?

-, . , , .

-, , . , , . , : , , .

, , . , , «», - . , .

, , , , . , , , . — , . .

, . , , , .

, , . , . , — .

, . .

, . — . , — .

— . . , , , , . , , — «» , , , — .

(, , , , . - -: ).

— . . « » — , , .

, , , — , . , , .

, : « , , », , , , , - , .

, — , , , «» , «» , , , , , . , .

( . — “ , ”. — “”?

: “” “ ”. , .)

, , ? , . .

. , , . , ( ) .

. , . — , .

, , . , , , .

-?

— . — , , , . « ?». (, «») : « - ?».

2015 . , , , - . , , — . ? , .

, . -, . , , 1616 - .

— . , . — , , , . , «» - - .

-, . , »%* $^ , , — , .

- . , , . - . «» , .

, , , . - , , , . , .

, , TED .

. , — .

( , ) , , , . .

-, . , , , — . .

— . , , - . « », .

: « ?», « ?», « ?». , .

, , . -, , , , . -, - - , , , — .

, . : ? , . ?

— - . , , , .

, , . IBM Watson , Google . -.

— , - - . — , .

— «» . , , .

— , . — . , — .

, : , . , , , — . , , .

, . , , — . « » — , .

— . - — , «» . , , . , — . The bigger, the better.

— ? ? ? , . . , . For example, such:

So, the main problem of today is not the generation, but the analysis of the texts. As soon as the machine learns to understand and even in some sense “empathize” with the creation of people, highlighting ingenious works among gigabytes of nonsense and graphomania, the problem of generation will no longer be.

Why, publishers are now ready to pay millions for a program that will be able to choose masterpieces from thousands of opuses that appear daily at Amazon samizdat, and it is desirable that it do it faster and more efficiently than poor philology students who are doing it now.

The development of the industry, therefore, will move from two sides : generation neural networks will be improved to extract more information from the data (RNN, LSTM), and recognition neural networks will approach the true understanding of the text, and, as a result, the true understanding of the surrounding world. At the human level, but rather greatly surpassing ourselves.

***

In the end I want to note one very important thing. Most of the results described in this article are very new. This is not even about decades: the above-mentioned article by Andrey Karpati was published in May 2015!

Right now we are at the next round of explosive growth: the number of scientists employed in the industry is growing, companies spend billions of dollars to keep up with the race of Artificial Intelligence. Technology also does not stand still: hardware companies, one after another, produce iron , intended primarily for machine learning.

And therefore, even at the moment there is not a single system capable of creating even a little meaningful prose, the chances that such a system will appear in the near future are very high. (Programmers threat, by the way, also did not pass ).

September 11, 1933 at the annual congress of the British Academy of Sciences, the best nuclear physicist of that time, Ernest Rutherford, categorically stated that everyone involved in the extraction of energy from the transformation of atoms is charlatans. And all their research is nothing more than another round of pseudoscience.

The very next morning, the young researcher Leo Szilard described the principles of a chain reaction, thanks to which in a few years the first nuclear reactor will appear.