"Soon, pilots will not be able to cope without the help of artificial intelligence": an imitation of "air battles" was held in China

The Chinese Air Force instructed its best pilots, aces, to engage in "skirmishes" with artificial intelligence, which leads to the enrichment of the experience of both "opponents" necessary for conducting air battles.

As in the case of the recent experiment in the United States, AI in most "battles" won out over humans. However, the pilots gained new experience and became more prepared for battle.

AI has demonstrated proficient flight control skills and flawless tactical decisions, making it a valuable adversary to hone our capabilities

- said the commander of the air brigade, Du Jianfeng.

"Battles" are computerized, which allows you to reduce risks and significantly save on costs. In August last year, the AI ​​won five consecutive rounds of simulation battles in virtual F-16 Vipers.

As noted in the publication Interesting Engineering, the continued development of AI will primarily lead to an increase in the capabilities of software designed for combat aviation:

AI can help pilots in combat by figuring out the best options for action at speeds beyond the reach of the human brain [...] We are approaching a time when pilots simply cannot cope without the help of AI.

According to some reports, pilots of the fifth generation J-20 fighters were involved in the training.

Comments

[mojohed2012]

Where did the Chinese pilots become Asami? Where did you fight?

[Blackmokona]

The concept of AU with WWII has depreciated long ago, there are almost no conflicts in which you can fill the required number of frags and now these are just the best pilots

[Troll]

Not depreciated, but yes, it has changed.

[credo]

Quote: mojohed2012

Where did the Chinese pilots become Asami? Where did you fight?

For example, on a simulator. This is also not an easy matter, and not everyone will be the winner.

[NIKN]

The AI ​​has won five consecutive rounds of simulation battles in virtual F-16 Vipers.

And at what level did you play? You need to start with an easy one, and then complicate the mission further :))

[voyaka uh]

You shouldn't laugh.
There was a full video - over an hour of this competition.
At first, several AIs fought in turns with each other.
And the winner fought an American Air Force instructor pilot.
AI won 5 out of 5 melee matches.
The pilot failed (using all his experience of deceptions, traps,
non-standard solutions) create not a single acute situation.

[NIKN]

How can be calculated and optimized, making a key decision is difficult. Will you be able to programmatically formulate the decision to ram? What does this technique have to radically change the situation in the situation, but can you predict two battering rams?
Well, that's me, so far, there is no AI in its concept, there is optimization of WB control algorithms, nothing more

[voyaka uh]

AI exists. This is a specialized self-learning
(without the participation of programmers) software.
Programmers only lay down the basic source shortcode.
Further, the machine writes the program code for itself, for situations with which
actually faced in training battles.
The software code grows like a snowball. But modern computers have a lot of memory.
This is no longer a problem.

[NIKN]

Quote: voyaka uh

Next, the machine writes the program code for itself.

It optimizes, but does not analyze, it does not have data on a person's behavior in a particular situation, and a person does not have such data either.

Quote: voyaka uh

under the situation

No more ... I respect you as a specialist, but AI has not yet been created, it cannot yet make decisions outside the algorithm, and if it can, I do not envy you or me.

[voyaka uh]

"she has no data on human behavior in a particular situation" ///
---
AI on the drum, there is in the cockpit of the aircraft with which he is fighting,
human. Or there is the same AI. Or an alien.
He has a mission to shoot down this plane.
And AI learns and copies the tactics of this Aircraft, on the move, in the air
himself changing and supplementing its own algorithm.
The programmers who created the original base code
They have no idea what the code will turn into through the rounds of fights.

[NIKN]

Quote: voyaka uh

AI on the drum

Of course, this is his disadvantage. It is easier to predict him than human.
The future is predicted in the movie "Star Wars", in the back cockpit of the AI ​​and in the front cockpit

[alexmach]

It optimizes, but does not analyze, it does not have data on a person's behavior in a particular situation, and a person does not have such data either.

She has such data. These data are summarized by statistical methods. These generalized data, roughly speaking, is the very AI.

No more ... I respect you as a specialist, but AI has not yet been created, it cannot yet make decisions outside the algorithm, and if it can, I do not envy you or me.

In fact, this is the most vague definition of AI, while given somewhere in the 60s, and at the same time implemented in practice - the ability to make a decision without a previously clearly formulated algorithm for making it. "Philosophers" also call this "weak AI" and it was implemented, I repeat, a long time ago and there are a bunch of methods for its implementation.

[Israel]

Further, the machine writes the program code for itself, for situations with which
actually faced in training battles.

No AI system writes program code in production, all current AI, not the most complex mathematics, various types of neural networks, from “simple” types of super-precise ones, transformers and newfangled graphs. And all the training is the correct input dataset, the correct coefficients, activation and evaluation functions, and a bunch of heuristics.

[xorek]

Quote: mojohed2012

Where did the Chinese pilots become Asami? Where did you fight?

They are traders, most likely Sergei, but they are trying to become a military power ..
Their tactics are well-known .. "If you sit for a long time on the bank of the river, you can see the corpse of a passing enemy ...
Let's see who will swim and how much ..

AI can help pilots in combat by calculating better options at speeds beyond the reach of the human brain

Russian hooligans are still the best in all ranks of the armed forces.
We offered the Chinese joint patrols in the tense areas of the world.
But the Chinese modestly refused ..

[Eug]

For some reason, the first stages of the development of chess computers come to mind ...

[Marachuh]

AI operates according to patterns and algorithms. And a person can think unsystematically. The only significant limitation of the pilot today is high G-forces. If it is possible to bypass them in some way, then the AI ​​will lose to humans for a long time.

[Mikhail m]

Quote: Marachuh

If you can somehow get around them

The problem is just this if. Development is proceeding rapidly, and the time is not far off when AI will be able to do without a pilot, who has become a factor limiting the capabilities of technology.

[Gardener91]

Quote: Mikhail M

when the AI ​​can do without a pilot,

Well, already ... And if a fighter is without a pilot, then maybe the term "close air combat" will lose its meaning?

[Marachuh]

It seems to me that this is a distant prospect. In the near future, I think there will be remotely controlled drones. Will increase the data transfer rate via satellite, something like 7,8,10G. The "pilot" will see a 3D picture, feel everything, but without overload, and make a decision. I heard that now NASA has a data transfer rate of 90GB / sec. And they will hardly dare to give everything at the mercy of AI. A flock of uncontrollable death machines overhead, it’s still a pleasure. Here you will remember the "terminator" and "skynet"

[Gardener91]

Quote: Marachuh

AI operates according to patterns and algorithms. And a person can think unsystematically.

A trained fighter pilot in a complicated combat situation "survives" instinctively. The very notion "thinks" is inappropriate, since actions are performed on a subconscious level and decisions are made instantly. This is partly not the case for actions in the NDE. And so yes, "thought" - lost time - lost.

[Souchastnik]

A trained fighter pilot in a complicated combat situation "survives" instinctively.

Here, in my opinion, lies the catch. Instinctively, this is practically according to the program (laid down by training). This is where the AI ​​can calculate a live pilot. There is only one way out - to act non-standard (not logical).

[Gardener91]

Quote: Souchastnik

Here, in my opinion, lies the catch. Instinctively, this is practically according to the program (laid down by training).

Drill air combat situations, even using standard million and small cart maneuvers! Work them all out to automatism ... AI can tell the direction, speed, set the altitude, etc., but IT will NEVER calculate the next move of the enemy. I bow my head to Ivan Nikitovich and his "My Science to Win".

[apricot]

On a car in an emergency, the arms and legs work independently and the head has no experience of anything like this, so it is impossible to predict the behavior of the car on partially icy asphalt.

[Aleks2048]

AI operates according to patterns and algorithms.

This is the highlight of modern AI based on a neural network, that there are no constant sequences and the algorithm changes from time to time, while unsuccessful options are immediately discarded. So I think that in fact the moment has already come the question only of maintaining communication and scalability. But as soon as the Ministry of Defense invites highly qualified specialists to work, these issues will be resolved, for example, the United States uses Link16. So far, this is very bad for us. But I think that soon they will portray something like that ... some kind of analogue. I think China is already on this path. The world is changing and whoever does not have time to move forward will inevitably fall behind. In modern combat, reconnaissance and communications give too many advantages to be ignored, not by conspiracies and the last, most difficult and most deadly link, in this chain from each non-killer link - the analyst is the place for making decisions and planning both the actions of an individual unit on the battlefield and actions of armies. And it may become so that it will not be necessary to remember

If you sit for a long time on the banks of the river, you can see the corpse of a passing enemy ...

And from the same treatise, fight where the enemy does not expect you.

For some reason, the first stages of the development of chess computers come to mind ...

Today, a computer confidently beats a person, but the most interesting thing is not that a program based on a neural network wins a program in which people analyze the chess games that have already passed.

[VO3A]

Let's descend to the sinful earth ... There is a problem with which automatic systems and AI, including ... This is a reliable definition of the enemy by "friend or foe" systems, especially in conditions of intense interference and the impact of electronic warfare systems ... Here it is necessary to take into account and algorithms for the operation and daily input of information into such systems ...
All wars begin with the erroneous destruction of their aircraft or other equipment in the absence of visual contact in a long-range battle ... or in our case, when using systems with AI ... This is followed by a ban on this type of combat and turning off AI in autonomous mode ... And also fire on the civilian population ...

[voyaka uh]

"AI operates according to schemes and algorithms." ///
----
It is self-learning software. It remembers, copies and optimizes
enemy techniques. Moreover, instantly - in the same battle.
Without the participation of a human programmer.
The Chinese realized what a threat it was. Russians - not yet.

[stalkerwalker]

Quote: voyaka uh

The Chinese realized what a threat it was. Russians - not yet

Lyosha ... Always happy for you. Savvy ...

[voyaka uh]

There are many smart ones. I sat, thought it over, understood ...
The quick-witted is in short supply. The Chinese are not just smart
but quick-thinking. And what is even more dangerous - fast-acting.
For this reason, China is beginning to overtake the United States and dominate all of Asia.
And further in the world.

[stalkerwalker]

Quote: voyaka uh

There are many smart ones. I sat, thought it over, understood.

You just surprise ...

[gsev]

Quote: stalkerwalker

You just surprise ...

The fundamentals of artificial intelligence were demonstrated in popular puzzle books of the 1970s, when the creation of self-learning artificial intelligence systems was hampered by the lack of cheap and small-sized memory. Artificial intelligence in Russia looks for terrorists who evade taxes, helps to track a person's interests and send him targeted advertising. For example, you look at a friend's computer or phone, say that I am interested in this sofa, artificial intelligence considers your photo as a camera or remembers your voice, determines your identity and sends a profile advertisement to all computers you use, if it considers you a potential client. creating a self-learning machine in the game of tic-tac-toe requires about 1 KB of memory, or about 30% of this number. In the end, ATGM fired and forgot systems were created, which better than the operator determine the target and direct a rocket at it. It's just that Russia does not consider the creation of weapons with artificial intelligence a priority compared to financing TV, theaters, sports, and religion.

[VO3A]

Yep, exactly ! The Indians were beaten in hand-to-hand combat with their fists with AI ... It is not harmful to want and dream ... Tell the kids fairy tales, maybe they will believe ...

[VO3A]

Dream on ... There is also the second part of the problem, but it is not the last ... It is very difficult to recognize the target in different conditions, from different angles ... Pi_ndos invented a bunch of ground-based automatic devices with AI ... There was a whole program, full videos ... Where are they all? Not used! This is not a simple automatic system for tracking the take-off of Jewish planes from their airfields and escorting on their last journey ... The distances are ridiculous, the devices are simple ... and you don't need AI ...

[rocket757]

"Soon, pilots will not be able to cope without the help of artificial intelligence": an imitation of "air battles" was held in China

... No wonder, everything goes to that.

[Gardener91]

So far, there are two ways: the navigator is the operator of the control system and the tactical situation monitoring and AI. Everything will depend on the goals and objectives of the aviation branches. And why were only fighters considered? Maybe the scope of AI applications should be wider?

[Pavel57]

How does AI get information? This is the main question, which still comes from computer games.

[t-12]

Just like a live pilot - from the radar and all kinds of sensors (speed, altitude). Well, on the radio from the command.

[sen]

Why does a computer beat a person at chess? It contains many options that people have developed, and he goes through them very quickly. The main advantage of AI is to find the optimal option very quickly and to quickly implement it. Man cannot do that. Of course, at first, AI will be most widely used in aviation, but very soon there will be no weapons left in the armed forces where AI would not be used.

[voyaka uh]

"It contains a lot of options that people have worked out, and he goes over them very quickly" ///
---
This was the case in the first generations of AI.
Today's chess software is not loaded with any variations, openings, tricks - nothing.
They are primitive at first, like children playing chess for the first time. But they learn on the fly - from the enemy.
And in the course of the games they become "grandmasters".
These AIs are confidently crushing the first software that won the world champion 20 years ago.
It's the same with aviation AI

[Israel]

Even when there is a real AI in chess, he will be forced to sort out positions and calculate moves due to the very specifics of chess as a game. In theory, the program can be "empty" without an endgame database, but will fill it up as it learns.
And current programs use well-known algorithms such as alpha-beta pruning, etc., the whole struggle for parallelizing tasks, correct estimates, etc.

[Blackmokona]

The current chess champion does not use the typical alpha, beta, and other clippings. Just a neural network that played with itself a lot

[Israel]

If you look at the phrase itself, the neural network was playing with itself. In the simplest case, an artificial neuron is a nonlinear function (activation function) of a linear combination of input signals.

that is, mathematically, y = F (u), u = the sum W0X0 +… + WnXn
where y is the output, F is our activation function, W are the weights, and X are the inputs. Further, the network is combined from the same neurons.

The whole training of the network from the point of view of mathematics consists in setting the weights and choosing the activation function, estimates, etc.
If we go back to chess and neural networks, the first “boom” was when AlphaZero “tore apart” the computer chess champion Stockfish Alpha Zero won 28 wins, drew 72 games and did not lose a single one (we will return to the conditions of the match).
It was stated that AlphaZero played with itself for only 9 hours, in fact, having trained the neural network, “filling the database” of openings and endgames, heuristics and various cutoffs of variants were probably used during training, AlphaZero was spinning in a full-fledged data center with huge capacities. And what did the enemy have, at the request of Google engineers, Stockfish played without a base of openings and endgames, so Stockfish was limited to a cache of 1 GB, and at the request of Google, not the newest version of Stockfish was used. Well, obviously, Google engineers experimented with different variants of Stockfish and found the most convenient for them.

Although it is worth noting that Stockfish, the current leader, has also become a hybrid using neural networks and classical algorithms.

[Blackmokona]

No, StockFish had a lot more processing power than AlfaZero and adding a base didn't save him.

[Israel]

He counted more moves per second, yes, but the computing power was on the side of AlphaZero and the neural networks themselves are quite a base (based on weight factors), but I will repeat the restrictions that Google engineers demanded (lack of databases, time limit per move, restriction on the cash, the "old" version), they clearly say that the "opponent" was selected for the "novice boxer".
Well, if you go back to the very beginning, chess is an understandable mathematical problem and that a person, that a real AI (when to appear), that the current AI or other programs, solve approximately a lonely, position assessment and counting moves, and since, after all, a person is not a machine, and in the ability to solve "number-crushing" problems, it naturally yields even to a calculator, it is not surprising that it loses to a computer.
For example, another task is to control a passenger car, until more than one autopilot manufacturer says that you can sit in the back seat and fall asleep and have nothing to worry about, and at the same time, many cars are autopilots, hung with cameras, lidars, radars and a bunch of other sensors in some cases. and with information about the general traffic situation, while the average person can only drive a car using two eyes (even one) and side mirrors, while driving in any climatic conditions, roads / off-road, etc.

[Blackmokona]

A trained neural network requires very little power, it needs a lot of training. Therefore, there is a win here, especially the chess players who began to carry nonsense by launching StockFish on their computers and seeing other moves were especially pleased. Only Google had Stockfish much more powerful, and if these chess players gave Stockfish a couple of hours to move, then the move changed to the one in the games shown.
Well, the databases are just prepared answers in advance, since they were not there and it was fair for neural networks. But adding them didn't save StockFish.

[Shahno]

Quote: voyaka uh

There are many smart ones. I sat, thought it over, understood ...
The quick-witted is in short supply. The Chinese are not just smart
but quick-thinking. And what is even more dangerous - fast-acting.
For this reason, China is beginning to overtake the United States and dominate all of Asia.
And further in the world.

Here's what I think. The Chinese should not be allowed into AI technologies. We'll have to somehow resolve this issue. They have become insolent with others. We'll have to pat the head. And they certainly do not deceive the Jews. Go west.
Ps. As a manager, I am not at all ready to be robbed of my profit by these people from the respected narrow-eyed diaspora, who did not pay anything. Whom they are trying to deceive is ridiculous.

[voyaka uh]

"The Chinese should not be allowed to AI technologies" ///
----
It's too late to "drink Borjomi"
China is a superpower. And dictates the terms. Softly but persistently.
Israel is a technological incubator for them. Which needs to be protected.
To pump out in all ways: and gifts in the form of construction projects,
and super-profitable trade, and espionage - the technologies they need.

[ammunition]

Quote: voyaka uh

China is a superpower.

in 1984, the PRC adopted a development plan for the country. The third phase of that plan was supposed to end in 2021. On the centenary of the CCP. That is, right now. The purpose of that plan in the third phase is
"By the centenary of the CCP (by 2021), raise China to the level of an average developed country"
.......
And to the level of "highly developed ...." they were going to the current for the centenary of the formation of the PRC. that is, by 2049. -))))))))))))

[voyaka uh]

Were going ...
But the development went in the American way, fast, capitalist with a significant advance.
communist party graphics.
And this can be seen in all areas: space, quantum physics and quantum communication, fusion,
5th generation of mobile communications.
Not to mention the engineering fields and all kinds of Ali-express.

[ammunition]

Quote: voyaka uh

But the development went ... with a significant advance

Not. Not ahead of the curve. -)) All according to plan. Now the PRC is a moderately developed country (in terms of GDP per capita). They have slowed down their overheated economy more than once .. so as not to get out of control .-)) A comment wrote like - And what will happen when they become highly developed ??? That's it, that's it.

[voyaka uh]

"Now the PRC is a moderately developed country (in terms of GDP per capita)" ///
----
In terms of GDP per capita, yes, you are right.
But this is because there are almost a billion and a half of these souls.
You can't save enough money for all.
But in terms of the scientific and technological level, the PRC is already firmly in the group
most industrialized countries.
The jerk is exactly the same as that of Japan in the 60s and 70s.
Then the Americans got scared too. But Japan didn't have an army
and she was under US military control.
And China is on its own. He is not friends with anyone.

[gsev]

Quote: Shahno

Here's what I think. The Chinese should not be allowed into AI technologies.

AI technology is embedded in Gardner's collection of puzzles. How much time the Chinese will spend on developing hardware and software depends only on them. In principle, the biggest secret is that AI can be created and in order to convey this idea to the leadership of the PRC. If the leaders of the PRC deem it necessary to create AI, they will not lag behind anyone in creating it.

[Pavel57]

Quote: voyaka uh

"It contains a lot of options that people have worked out, and he goes over them very quickly" ///
---
This was the case in the first generations of AI.
Today's chess software is not loaded with any variations, openings, tricks - nothing.
They are primitive at first, like children playing chess for the first time. But they learn on the fly - from the enemy.
And in the course of the games they become "grandmasters".
These AIs are confidently crushing the first software that won the world champion 20 years ago.
It's the same with aviation AI

There are two variants of chess aggregation algorithms based on neural networks - some learn by themselves, playing with themselves, others are taught, starting from the games of the masters.

[Pavel57]

Quote: t-12

Just like a live pilot - from the radar and all kinds of sensors (speed, altitude). Well, on the radio from the command.

And in close combat, where they promise the complete superiority of the AI?