Artificial intelligence: fantasy and reality

Any person who has played computer games at least once has an excellent idea of ​​a simple artificial intelligence from his own experience. In fantastic plots, such an intellect settled even earlier. Since then, he either tortures the heroes of the ancient System Shock or the not so ancient Portal, then assimilates the organic races of the galaxy in obedient cyborgs obedient to the will of the superintelligence in Stellaris, or does something else equally reprehensible and plot-forming.

But if we dig even a layer deeper than the popular artistic image, we will see in the same games a set of cunning software elements occupied with official practical things – such as the adequate behavior of a monster armed to the teeth in the sight of a live player, planning a cunning maneuver on the playing field and other complex actions. , which are usually considered accessible only to the human mind.

Therefore, it is called intelligence, albeit artificial. But what actually hides under this name – and how does it work?

Difficult task

Seventy years ago, in the distant 1950s, when what would eventually become modern computers was just emerging, humankind was over-optimistic in many high-tech industries. It seemed that a full-fledged artificial intelligence was already a stone’s throw, if you only know, finance the development. And there are already robots, and interplanetary atomic vehicles, and apple trees on Mars.

But immediately serious problems arose in full growth. Traditionally, “suddenly” for science, it turned out that scientists do not really understand how what they are trying to emulate on the available electronic and software base. And it’s okay they don’t represent – but they also cannot decide for themselves, but is it necessary?

Well, in fact, if a cool nuclear submarine is not particularly similar to a whale, and a cunning drone drone, although it flies, has almost nothing to do with a bird, is it necessary to sculpt artificial intelligence in a human image and likeness? Maybe it will be enough for them to solve complex problems faster and more efficiently than people do?

Science versus magic

Almost 70 years later, humanity has made great strides in the creation of AI, but the problems described above have not gone anywhere. If the program distinguishes between a bare field and a strip of uncompressed rye and drives a harvester equal to a live driver – is this already artificial intelligence, or just a set of clever algorithms that compare mathematical descriptions of key parameters without any self-awareness? And is it really necessary here at all, is it self-awareness, or is efficiency alone enough?

Moreover, even the question of self-awareness remains open: humanity still has a very poor understanding of how our brain works. Otherwise, polyclinics would have been treated for idiocy long ago. But modern logical theories are perfectly able to break complex actions into simple steps, and mathematics is able to describe these steps in terms that can be reduced to a sequence of zeros and ones and fed to a seemingly completely stupid, but surprisingly effective artificial intelligence.

Harry Potter and the Mystery of the Philosophical Algorithm

Scientists and romantics have endowed artificial intelligence with the traits of a full-fledged human personality for a century already. Character – preferably quarrelsome, aspirations – preferably misanthropic. Otherwise, no conflict will work, neither the reader will be attracted, nor, later, the player.

Practitioners, meanwhile, are betting on fairly extensive algorithms that can give a quick, accurate answer in their field. The early ones, of the last century, usually did monstrously narrowly focused, like expert programs for playing chess or go – but from the point of view of technology, the era is extremely effective. Already thirty years ago, at the dawn of the distant 1990s, fiddling with the consistent simplification of each problem to the stage when it can be calculated with a combination of mathematical, statistical and software tools has yielded results.

Significantly, no one considered these achievements to be artificial intelligence. Every time humanity took a technological peak – such as effective recognition of human speech, or readers that are able not so much to mumble monotonously, but to really speak with expression, this was recorded as the success of the developers of an innovative program – and they immediately stopped calling it artificial intelligence.

Yes, in many ways such programs are the result of striving for perfect computer intelligence. But people still stubbornly call them just cunning programs. And they continue to make them more and more complicated, and the iron – more and more productive.

Computer power

The combination of a modern powerful processor, cunning algorithms and a large margin of space for raw and intermediate working information is already quite enough to get the expected result of acceptable quality. About ten years ago, in 2012, machine algorithms for “deep learning” began to provide solutions to specialized problems, similar to human ones. If a modern computer can calculate a picture of a battle with hellish devils armed to the teeth in real time and in high resolution, it has already been capable of practical tasks for a long time.

In those areas where it is necessary to take into account and correctly process a large amount of similar signs of this iron is already enough. With proper training, a modern computer program processes and takes into account the results of a medical scan of the body better than a person at the stage when you just need to find out whether he is healthy, or sick, and if so, how badly.

In the future, when such systems are thoroughly tested, they will certainly free the modern doctor from mechanical review of patient examinations, immediately giving him preliminary results – already for making an accurate diagnosis and prescribing treatment.

It seems that this is not so much, but the benefit of such a system based on modern artificial intelligence is that it does not care whether seven patients come a week for examination or seventy. She will review everyone – quickly, efficiently, accurately. And an inexperienced trainee can also manage it, unloading experienced specialists. In Russia, such promising developments are seriously and quite successfully engaged, for example, by Rosatom, one of the world flagships of high technologies.

Engineer clause

It is important to understand that all modern breakthrough hardware and software are relatively young. There are no analogues to them in the last century in terms of price-performance ratio. Problem and Opportunity Analysis Papers are out of date. Much – to the point of ridiculousness

The real history of the modern development of artificial intelligence is being written here and now. Where it will come is simply impossible to say absolutely for sure. Modern young specialists have every chance to write down their names in history. Only one thing is clear – commercial hardware and seemingly cool by the standards of the last century processors for the transition to the next generation of effective solutions are critically lacking in speed. Because the medical diagnostics system described above is one thing, and the program of a “smart city” (or even a “smart country”), which must analyze thousands of different parameters, based on them, making both operational decisions and making forecasts for years to come, is one thing. …

A real breakthrough result can be provided by quantum computers of a new technological generation. There is exactly one problem with them – with the expected speed and information density per unit of time, insane by the standards of the old computer hardware, they are still at the experimental stage. Even in order to really play with their capabilities and understand what they really can, and what tasks they can really entrust – you need to at least have the first models in metal.

Rosatom is responsible for the creation of the Russian quantum computer, which together with partners created the National Quantum Laboratory for this. A significant part of the work on the development of a quantum computer at the NKL is carried out by the Russian Quantum Center. The work is designed for a period until 2024 and involves the creation of a quantum computer simultaneously on four platforms: superconductors, cold atoms, ions and photons.

“ Learning artificial intelligence on both classical and supercomputers is a long, expensive and, unfortunately, ineffective task. Quantum algorithms can make a difference, so many people associate the emergence of real AI with the creation of a multi-qubit quantum computer. The minimum program: to accelerate the learning and work of AI with the help of quanta, the maximum program: as a result of this acceleration, to obtain new qualitative effects that will lead researchers to results that we did not even think about , “said Ruslan Yunusov, head of the project office for quantum technologies of the State Corporation “Rosatom”.

Effective manager’s clause

In the modern pursuit of high technology, the bumps are already stuffed more than once. There have already been periods in the industry two times when those who talk about artificial intelligence were perceived as “again these crooks with their computer scam got used to it” – so massively and unanimously leaving both crooks and scientists without money, that it was all dubbed “Winter of AI”. But if you know exactly what you want to get from a modern high-tech system within the framework of well-known solutions, you can get a lot.

Yes, the first computer programs for effective online trading could easily fall into a senseless price increase and other frankly absurd decisions. But modern neuromorphic systems, with their proper tuning and training, ultimately completely “outplay” a person where it is necessary to keep in mind a huge array of variables and quickly take them into account, compare and process. Of course, the live operator decides. But the stage of preprocessing for him is done by a narrowly focused, but quite working artificial intelligence.

Xenophobic clause

The emergence of a new complex system, which at the same time actually decides something and something, and worse, someone, really controls, of course scares a lot of people. Something incomprehensible and alien really appeared before them, such that one can easily attribute the achieved efficiency to the intrigues of an evil mind. And if it also learns like a modern effective neural network!

The creators of OpenAI gave their peaceful brainchild the opportunity to play the equivalent of 10,000 years of fighting in Dota 2 – and led it both to the creation of an effective school of the game (albeit with external restrictions on all the tools available to players to win) and to the uniqueness of this very school of the game.

Cyber ​​veteran of championship games Sebastian Debs from OG openly admitted that OpenAI Five is doing what no one else has done in his memory, and you do not immediately understand why it was done at all. The main problem of a person is the transition from “no, no, some kind of stupidity, probably” to attempts to comprehend why the AI ​​adversary did it at all.

It was then that the xenophobe turned on in their heads and began to pester: what if not in Dota? And if not in units? And if unmanned tank wedges, guided MLRS shells from closed positions and an armada of drones eclipsing the skies?

Glory to the robots, kill all people?

Specialist’s reservation

Employees of less committed organizations, like the same Alexei Fedorov from the Russian Quantum Center, in such cases talk about current achievements that are within the power of artificial intelligence on a quantum computing platform.

And indeed – if the materials of the ancient Nuremberg Trials occupy many tens of terabytes, although they were documented seventy years ago, in the harsh analogue era, then what about the materials of modern high-tech industries – where are the same immense terabytes in a continuous stream? For example, from a nuclear power plant?

Both a high-speed computer and an effective trainable artificial intelligence capable of solving industry problems are needed literally “already yesterday.” That is why they want to see the working hardware of their own design, a quantum computer for 100 qubits, by 2024, and the design office of the digital block of the same Rosatom under the control of Ruslan Yunusov is busy with the issues of the material base – superconductors, cold atoms, ions, photons. What the super-developed civilizations of the distant future used in science fiction and computer games is becoming a reality here and now.

HR clause

A horror story about how much more elitist all professions will become, as soon as inhuman and ruthless computer algorithms come to replace lower-level workers, is wandering around the network en masse.

But in reality, even here they are perfectly engaged in real help to real people. The modern personal computer has saved a person from the need to sweat with a pencil over a slide rule and carefully draw and color graphs by hand. The numbers and formulas for the last forty years can already be hammered into the sheet of the accounting program on the screen, and it will already calculate the results and draw beautiful graphs.

What will change for a person when a program with AI elements can fill the numbers there more or less by itself? Pulling meaningful questions out of the workshop transcript – which I wrote myself? Translate legal documents into human-readable language? Does it itself redirect the request to technical support to a competent specialist – “reading” dozens of letters per second from those tens of thousands that continuously storm any technical support of a large company?

Yes, it will just become easier to work. But an adequate solution in an environment with an excess of variables is still within the power of only humans.

Personnel accounting, drawing up an individual training and retraining program for each specialist, adequate redirection of important intermediate information to the specialist who will cope with it in the best way – all this can really be delegated to the specialized AI in order to get, on average, better results than without its participation.

Security clause

Even more interesting is the combined system of debugged AI pattern recognition and workplace safety requirements. Almost any labor protection standards can now be controlled by a video camera system and a sufficiently powerful computer intelligence, which is able to recognize whether a mask, gloves, a protective helmet, corporate uniform or special clothing are worn – and notify the labor protection service in time about an outsider at the facility or violation requirements by employees and guests.

Modern computer vision works just as effectively in flaw detection – it is already easier to distinguish a good weld from a computer that has moved on than a tired person on the next kilometer of inspection. will definitely go on and on.

Denis Larionov, Head of the Artificial Intelligence Department, Tsifrum (State Atomic Energy Corporation Rosatom):

“Artificial intelligence has been used for many years in solving security problems, for example, for video monitoring of personal protective equipment, defectoscopy and quality control of products, controlling the speed of vehicles and monitoring violations of traffic rules. AI is also used to identify individuals and search for criminals, in anti-plagiarism systems, in the analysis of inaccurate records, filtering mail spam, and identifying information leaks. Promising areas of application of AI for security tasks are searching for anomalies in large data streams, recognizing situations in the natural environment, and cybersecurity. As for ensuring the safety of products that use artificial intelligence technologies, the main thing is that this should not be accompanied by attempts to limit AI. It provides a competitive advantage for businesses,

And futurology!

Modern artificial intelligences, already quite capable, when compared with the achievements of the last century, cannot even a hundredth of what they expected from a cool robot from a fantastic action movie. What’s next? General happiness? Common misfortune? Taking worthless bags of meat and bones to the dustbin of history?

Opinions vary.

Raymond Kurzweil is optimistic. From the point of view of one of the leading minds of Google, humanity is already half a step away from a true technological singularity and shared prosperity. Well, indeed, if computer systems of high complexity more effectively solve complex specialized problems, then from some point it is realistic to create a computer platform and complex AI, engaged in improving computer platforms and those intelligences. It seems clear to the proponents of the idea that this process will show the same explosive growth that the computer industry has over the past forty years, since the early 1980s. This means that the possibilities of high-tech tools available to people will rapidly increase. Kurzweil believes that this will happen by 2045 – however, the forecast looks too optimistic.

James Barratt, a journalist and documentary filmmaker, takes the opposite view. Any highly developed system begins to rake in resources and optimize the struggle for them with competitors. Just remember the uneasy relationship of antiviruses with other programs in the struggle for computer memory.

At some point, it will be relatively easy and profitable for a sufficiently developed artificial intelligence to optimize humanity until it is completely destroyed, after which it will be relatively easy and profitable to build up the solar system not by space cities with imitation of terrestrial living conditions on board, but to mold much simpler and more profitable blocks of a distributed supercomputer … But in any case, already without us.

Most likely, both of them are wrong.

To the heights of the singularity

Neither excessive optimism nor excessive pessimism affects the real history of the development of technology. H.G. Wells lived from a fictional description of armored vehicles and airborne assault forces to the actual landing of a massive assault on French territory. A reusable space rocket, landing on its butt in a cloud of smoke and flame, had to wait a little longer – but it also exists.

And each time it all looked like predictions only with an anecdotal statement “But in the main he is right!” Most likely it will be so with AI.

Does this reverse the promised technology singularity?

Yes, somehow not very much. People are already showing the main signs of its approach. Technological development is certainly accelerating. New results appear faster and better than previous ones. Forty thousand years ago, people lived in caves, five thousand years ago they built the first cities, two thousand years ago they invented a republic and democracy, a thousand years ago they began to work on mechanics, two hundred years ago they rapidly squeezed the world in size with a steam locomotive and a telegraph, a hundred years ago they used in battle, strategic long-range aviation and weapons of mass destruction, half a century ago mastered computers and the atom, thirty years ago – workers personal, twenty years ago – social networks and truly effective search engines, ten years ago – early trained artificial intelligence.

Within the framework of this bad sequence, further only the rapid explosion of computer technology, in which people do not really understand what is happening, and technologies become obsolete almost as quickly as they are introduced.

The harsh truth of the singularity

An effective computer AI system requires neither personal goals nor self-awareness. Only the efficiency of processing the incoming data stream and compliance with the task. A very cool modern robot machine works faster and more accurately than a person, but completely without understanding why he is doing it. Even the principle of neuromorphism, that is, the correspondence of the processes of computer processing of incoming information to the work of a living brain, does not imply either self-awareness or one’s own desires at all – but promises a sharply greater efficiency of work.

What will change when he gets an external cyber repair system, improvements and updates? Nothing special, only the engineer on duty will find it easier to work. What will change when such a system begins to rule not individual machines, but the entire production of the industry – something at the workshop level, something at the plant level, something at the level of the production chain? Yes, too, especially nothing, except for direct economic indicators.

Can that self-consciousness suddenly appear in a rather complex distributed system? It is unlikely – after all, as we wrote at the very beginning, the creators of this system still do not really understand what self-consciousness is and where it comes from. But can such systems more and more effectively solve their main task, and specialized design systems – to develop new hardware platforms and software for such a solution?

Of course yes. If we consider the onset of a singularity the moment when the responsible developer relies on the opinion of a set of expert programs in making a decision – yes, it will come. It optimizes both the accuracy of decisions and the effectiveness of any informational feedback, and the selection of the right expert. The same Russian “Smart Arctic” as an expert system should take into account all important parameters of the region – from the ice situation to the internal economy and external relations of the region.

Does the massive introduction of such systems by countries of the world mean that we will all immediately move into immortal non-humanoid cyber-bodies? Unlikely. Will it become cheaper and more comfortable to live and work – most likely yes, it will, although it is still somewhat naive to expect such an effect from the introduction of AI alone.

Well, conversations about an intelligent AI, which in terms of the level of consciousness will rise on the same level with a person, should be postponed until a later time – when a person will better understand what an AI is all about.

PS We have prepared this text together with the Homo-Science project from Rosatom. It is a scientific and educational platform that brings together scientists and those who are interested in science. And also visit our mini-site supporting Homo-Science, enter contests and win prizes.