Why are we so afraid of neural networks

For those who do not like to listen, but love to read, Samizdat continues to publish deciphering the best podcasts of the Glagolev FM radio. This time, together with the author of the most terrible of them - “Sociology of Ostroma” - Kostya Filonenko think about whether artificial intelligence will inherit our vices and flaws, why social networks make us post photos of young ourselves and why we are so afraid of neural networks.

Some time ago, a challenge started on Facebook: the next mass psychosis (this cannot be called otherwise), when people suddenly began to post their photos of a decade ago and compare them with the current photographs. And many wondered: why is it needed? For what purpose do people do this? And at this moment information surfaced that it seems like this calibration of a certain system of recognition of persons, which should compare the same person at different ages in order to recognize him in all photographs and recognize him in the future. And thus, this artificial mind becomes smarter, bringing the moment of our enslavement. In order to find out whether it is like that or not, we called Ivan Puzyoryov, a specialist in the field of VR, artificial intelligence and digital strategies in architecture.

-Ivan, is the 10-year Challenge really bringing the enslavement of the natural mind artificial?

- Personally, I think yes. It is too simple to not do this, having a Facebook power. I was at a rather large conference in Munich, and a high -ranking speaker from Facebook with a perfectly animated presentation and a well -staged speech in the middle of his report used a slide on which, on a blue background, “More Content, More Power, More PEOPLE” was written on a blue background.

- The word "Power" scares the most.

- Yes. And at this moment people clap, the badges are shaking, the tape is scrolling, and understanding people exchanging glances, because the company with such power, with such a quantity of data that at a large important conference that is published, inserts me into its presentation, scares me.

- Does it scare that they have “Power” for purposes? ” What they need more people in order to know more? What is frightening in this?

- Here it is worth telling about platform capitalism. Facebook has such a significant scale, such a size that, in principle, Facebook citizenship can be correlated with the citizenship of a certain country. The united platform citizenship can definitely give you the result similar to citizenship of the country. That is, we can move, we can exchange information by only combining our citizenship inside the platforms. The way I see it. Facebook is strictly oriented and exists in the world of capitalism. This means that she needs to make money. And when the company is based on marketing, advertising, we can very easily see how, for example, at the same conference, they show one of the options for selling advertising, in which we are a subtle tape, and when we see glasses, advertising points, I lingered a little longer in this picture, my front camera is cut - and glasses are trying on augmented reality on me. Yes, in principle, harmless, but if it is my life? If at this moment I do not want to try on my glasses? Of course, at first they will ask, and then they will not ask how we saw it many times with Google. If you look at their first loading page, it sounded like "we are harmless guys, we will only look for."

- The main difference between the state and any other association is the presence of its own army. What plays the role of violence in corporations?

- This is a worthy question. When we talk about the army, the question of borders always arises in my head. That is, they must be protected. And in general, in the economy being broadcast now, in geopolitics, borders lose such significance. Because - what is the border? The presence of information-that is, where the information is, and how it crosses any feature-this can be a border. Perhaps you saw recently: the largest database, the leak occurred - 700 million accounts, and everyone can check if his data leaked. I looked: mine leaked. All my mail is now in the public domain. It will be necessary to change passwords. This is pretty suitable for the phrase "O Worst New World."

- That is, instead of soldiers and border guns, we have a wall of passwords and protecting code?

- Yes, the Chinese wall is not a physical, under construction wall of cybersecurity that does not work. Once again, I emphasize that, even using the resources of one of the largest economies, it is impossible to limit the flow of digital information from the country or to the country. That is, being there, I could still use the resources that were blocked. Yes, these are dancing with a tambourine, but it is still possible. I do not see a way to stop this, except following the example of North Korea, where the country is simply physically cut off.

When we talk about neural networks and artificial intelligence, there are things that scare us, namely: a misunderstanding of how it works. In particular, there is very terrible news about the fact that Facebook made chatbots who were set opposite each other. They conducted a dialogue, and after some time they began to speak English, but it was their own language. They used English words to somehow communicate with each other. It was not clear for people, but at the same time they informed each other. As the news says, they were turned off, and for me as a person who reads articles on this topic, it became obvious that they were turned off because it became too scary. What will they agree there? Nobody knows except them. In fact, the language that they created, the data transmission method was much more complicated than necessary to get scared.

- Much more difficult than you need to be scared?

- Yes, he was extremely ineffective. About half an hour to sing a song to convey one phrase. But things regarding artificial intelligence, all the most specific and creepy come from China. If the developments that we see in the United States are one way or another try to control other capitalists - like Elon Musk - then in China we see chaos at all. I can tell a story. We ended up in the headquarters of Tencent in Shenzhen. This is a colossal company.

- And what do they do?

- They own WeChat. This is from the main assets. This is a colossal size of skyscrapers, and in one of the audiences they told us about their experience with artificial intelligence. There are several features that you need to know. Firstly, they teach their artificial intelligence in games. Gamification plays an incredible role in the process of education, the formation of man.

I advise you to watch the film "Sinofuturism", revealing everything that happens in China, in terms of gamification. Artificial intelligence, nurtured at games, is trained in strategies to beat a person or other artificial intelligence - why is he released in the city or into the space of the social network/messenger to solve local problems, but with the skills received in the game? Nobody does so. This is a unique experiment. One woman in the audience told us that recently they held a competition among their artificial intellects. They released them “to freedom” - hiding that it was artificial intelligence, passing them for people. Challenge was that the person with whom the dialogue was conducted had to tell the artificial intelligence "I love you." But he had to say it naturally. He had to really fall in love. In this, Challenga - whose artificial intelligence will be the first to fall in love with. This is pronounced with a smile under a very strangely regional presentation in a skyscraper in the center of the country, and I do not understand this presentation at all.

-This is like in the film “Ex Machina”: the subject not only believes that the robot feels something, but also feels something to him, and helps him avoid this. This is Turing Test, right?

- You can say so. I can’t imagine what is happening in the subcortex, what happens there deeper - how many algorithms are now present in the process of communication between people. It is worth recalling the recruiting process and answers Alibaba - all their collection of hardware ties, work with clients; The enemy plays a colossal role, which understands that you wrote/asked. Even in Russia, the question you asked, and the robot is already answering.

- What you say is amazingly converging with what the first guest of Strand's sociology, a specialist in deep Internet, said. He said that for his own security - mental and informational - it is necessary to very clearly understand that on the other side of the screen there is a person, not artificial intelligence. Artificial intelligence cannot have the properties of a real person, and people teach him the properties of real intelligence. We cannot teach him anything good, because we do not have a single altruistic and non -conflict example of behavior. The depravity, originally laid down in a person, can also transmit to artificial intelligence.

- Yes. There was a story when Amazon launched Alex in “Free Swimming” before the physical column exit. Then there were still experiments. They said “study on tweets”-and after some time, I don’t remember the deadline, after a few days it was a robot and a sexist who tried to buy and sell drugs. And here we come to the most basic question: there is artificial intelligence, there is a neural network - what and how is it? In my head it works like this: there are Data sites, you need to teach you something and someone should teach. That is, we need the amount of information to give her neural network, which can say: “Yeah, now I can make decisions based on how I saw.” Or, in simple words, I can explain this way. We will play the game. Explain to me what a cat is.

- This is an animal, small in size, with a long tail, different types of wool, eyes, nose and ears.

- In an amazing way, I can choose another animal for all the parameters that you named, and, unfortunately, we will play this game for about an hour, while we will understand that there is no parameter that would highlight the cat among other animals. Therefore, a cat in your head is a regiment on which all the photos of the cats that your eyes made, all the cats you saw, and then the link and the word “cat”. And every time you see a cat, you feel apofenia. You have a certain “Clue”: you are trying to guess on the basis of all the cats you have seen, and you recognize much better when you saw more cats. Therefore, the term "apothenia" is so important when we are talking about neural networks. Maybe you saw these pictures with strong distortions on which they offer to find a dog. And when she finds the neural network, she experiences apofenia in connection with what she saw.

- A person has hormones, in the neural network - what is this?

- I will think about what analogy to draw here. Who teaches or who is a DATA set is priority in the answer or priority in which neural network or artificial intelligence we get.

-I listened to the performance of Ilya console, an anthropologist from the European University in St. Petersburg, and he said that when we communicate with artificial intelligence, we make much more demands on him than a living interlocutor: when a person knows that he is communicating with a bot, he wants to somehow test him, and when he knows that he was a living person, we behave differently. And what to teach artificial intelligence to communicate on the records of people's conversations is not very effective now: people do not listen to each other, they interrupt. This is to the question of human malformations and how they are transmitted to artificial intelligence: when we program the chatbot, we lay a much greater degree of participation there, much greater empathy than it is required of a living person. Can you somehow comment on it? What is more promising: to create your own language to communicate with them or, on the contrary, to adjust them to us? And what could be more dangerous?

- There is a belief that artificial intelligence has problems with PR. Because "here is a bot, and I am a man, I am better." We want to somehow prove ourselves, ask specially complex questions. I think there is no doubt that this is a process that is not subject to stop. The most important thing is how the country (if we take higher - the scale of the planet) will decide to regulate himself, which will decide to introduce the rules to work with artificial intelligence.

- Like a country and a country as a corporation?

- Yes, sure. And, of course, in the country I do not mean a contour on the map. This is seen as a set of megagorods, special economic zones that are combined. Not in the good old concept - "country". And when these points make decisions about how they work with artificial intelligence, what they make with people recognize whether residents themselves have access or only a big brother has access. And this is already being built by ethics of communication with artificial intelligence: he is my friend or the same unattainable representative of the authorities - and I must wear a special sticker who will not allow me to recognize me.

- Does this mean that we have different race of artificial intelligence, different nationalities - and how can they interact with each other?

“A cool term, but I have not met it before.” No doubt, yes. The difference is as great as the country's language or legislation may differ in the same way. Even the very fact of whether the country says that they are developing or not, is already a certain indicator.

One of the graduation projects that we did at the institute was a study on the future transformation of the planet. Anyone understands that we do not have enough space to grow the necessary volume of products. Half of this is taken by animals, we eat half, and we do not have enough space. How to make your plate be connected with what we grow? For this, a system was proposed in which two artificial intelligence conduct a dialogue among themselves. One represents people who are a customer (I want this), and the other represents the planet (what I have, what can I do now, what kind of space I have, what I can offer now). And with the speed of millions of communications per second, they offer each other and reject something in order to achieve an option in which one will receive what he needs, and the planet will not suffer much. We refused such parameters as “I want carrots”, “I want carrot”, “I want vitaminity”. I want a parameter or some kind of color at ease? And, if in simple words to tell how my plate can be an architect of space? As soon as I took something, it landed. And the direct connection between these phenomena can be achieved by the dialogue of two artificial intelligence, which know the whole picture entirely. This one knows how the window opened, he has a free bed or not, he feels the earth. And the second knows what people wanted.

- That is, roughly speaking, on Mondays I want to eat scrambled eggs with bacon and tomatoes, and you can grow other people personally and under me personally, but under others who do not need it. Here in this sense?

- Yes. The moment of your meal should be associated with the moment of the appearance of a new egg.

Elon Musk quite directly and clearly several times said that the main danger that many ignore is artificial intelligence. What he means is not entirely clear. Or this is an intervention in the structure of society, but it is not very scary. It has always been. Technologies always lead to a change in labor, nature of labor and employment. This is not scary. And scary - this is when artificial intelligence becomes the third party in an armed conflict. What awaits us at this moment is another question. It is clear that this is not the script of the "Terminator" and not the "Matrix". We can already fantasize about much more subtle things.

An example that comes to my mind is from Pelevin’s book “Pineapple Water for a Beautiful Lady”, where there were drones, who not only made the decision to “bomb - not to bomb” on their own - there was a separate intelligence that developed a moral justification for this violence, and there was a large ethical discussion about each stroke, which ultimately led to the fact that, unlikely, it was necessary to bomber. What awaits us in such an alleged conflict? Can this artificial intelligence really be attracted? Yes. A large number of armies and developments are obviously using either combined decision -making, in which both a person, artificial intelligence, or exclusively artificial intelligence decides even about such things as “attacking”. I talk about this so calmly, because although it is unprecedentedly terribly terribly, I am not sure that it is possible to somehow stop. The key point that you should pay attention to is such a term as Dead Man's Switch. This goes well into a connection with a person who is responsible for the fact that this happened. Now, when the drone has decided what needs to be attacking, regulation is introduced in most countries: they are trying to prevent this from happening. But a country that is permitted will always appear: we can use it. And while the dialogs will go, this technology will still continue to be present in combat conflicts. Personally, I imagine it so much. Что является очень важным, так это «dead man's switch». Это значит, что некий человек, который сидит за пультом, может не принимать никакого участия в управлении, но его рука лежит на переключателе. Этот переключатель даже, возможно, не может выключить дрон. Но это значит, что ты несёшь ответственность за то, что делает этот объект. Это очень важно, потому что, если двое разработчиков сделали алгоритм для тысячи дронов и эти тысячи дронов убили миллион человек, эти два разработчика не могут нести ответственность за это. Это недопустимо в нашем мире. So far. It is very important to me.

Listen to other SoundCloud podcasts

— То есть решение проблемы того, кто несёт ответственность: тот, кто изобретает, или тот, кто использует?

- Yes. И, образно, почему я так спокойно говорю о том, что дрон может управляться субъектом, — потому что точно так же, как они могут спрятать неадекватность или неправоту искусственного интеллекта, точно так же на пресс-конференции может быть спрятана неадекватность и неправота человека относительного того, бросить бомбу или нет. Если чуть-чуть затронуть мирную тему, то, когда мы летим в самолёте, отчасти я могу назвать пилота «dead man's switch». В большинстве случаев он не принимает управление в самолёте. Даже посадка на современных самолётах сейчас осуществляется автоматически. Это human object, который несёт ответственность, если что-то пойдёт не так.

- Class. Это очень-очень круто. Я предлагаю рассмотреть три главных страха, связанных с нейросетями. Первый — так называемая проблема вагонетки, когда испытуемому предлагается решить, кого задавить, а кого спасти: на одном пути лежит один человек, а на соседнем несколько. Здесь возникает жуткое ощущение того, что человек, который программирует, играет в бога в этот момент и он должен предложить решение этой моральной дилеммы.

— Это, пожалуй, самый знаменитый вопрос, который задавался к системе, например, управления автомобилями. Когда едет автомобиль и нужно принять решение, кого мы будем сбивать. И на самом деле я не знаю, почему вопрос так долго висел, потому что ответ прям сразу понятен. Компания не может сказать покупателю: «Знаете, она вас может убить в 30 процентах случаев. Если будет выбор между вами и десятью детьми, то, к сожалению, она вас грохнет. В стену влупитесь и умрёте». Нет никакой дилеммы. Я буду спасать того, кто заплатил бабки за тачку. Чувак, это капиталистический мир.

— То есть это, по сути, рынок уже решил, а все моральные дилеммы есть только у людей, которые об этом задумываются, но не участвуют. У искусственного интеллекта нет дилеммы?

— В том репрезентативе на родине, откуда он, в головах нет дилеммы. Странная немного аналогия, зато поможет немного сориентироваться, что в мире-то происходит. Книжка называется «Капитал в XXI веке». Естественно, по аналогии с «Капиталом» девятнадцатого.

Томас Пикетти — автор. Книжка довольно большая, но я скажу одно предложение. Можно не читать её. «Доходность труда ниже доходности капитала». Dot. All. OK. Вторая проблема, второй страх-фантазия — как и любой страх, он порождает огромное количество фантазий. И, собственно, то, с чего началось освоение темы искусственного интеллекта в массовой культуре, а именно — восстание. Когда у нас забастовка шахтёров, мусорщиков — это понятно. А восстание роботов-шахтёров или роботов-мусорщиков — это звучит уже очень страшно и жутко. И, по сути, все художественные сценарии этого — они все апокалиптичны.

- Why is this happening? Это проблема пиара и восприятия нами искусственного интеллекта или она в чём-то другом?

— Я думаю, действительно, базовый получасовой урок о том, как это устроено, на YouTube хоть немного расширяет восприятие и смещает границу этого страха.

— Но тем не менее, например, Стивен Возняк в одном интервью сказал, что, возможно, искусственный интеллект достигнет точки сингулярности, когда он просто станет развиваться сам и уйдёт далеко вперёд человечества и будет относиться к нам, как к своим домашним животным. Моя собака никакой функции, по сути, не несёт. Мне просто приятно, что она есть. Я даю ей лучшее мясо, лучше, чем себе. Самым лучшим образом ухаживаю за ней. How so? Ведь люди к домашним животным относятся по-разному.

— Когда искусственный интеллект обыграл человека в шахматы (назовём этого человека — Гарри Каспаров), тот сказал: «Я создам новую лигу в шахматах: там будут играть человек и искусственный интеллект против человека и искусственного интеллекта». It's cool. Такое будущее мне представляется аккуратно встроенным в реальность того, до чего я хотел бы дойти.

— Почему искусственный интеллект так сильно связан с обманом? Почему те китайцы, которых ты описал, по сути тестируют, сможет ли человек распознать обман или встать в кружево из обмана.

— Смотря с какой стороны взглянуть. Искусственный интеллект ищет, где его обманули, где не кошка. Помнишь, говорили: «Это что, не кошка? Вы меня обманываете? Это не кошка? Нет, это кошка».

— То есть можно сказать, что это не обман, а ошибка? Поиск ошибки?

— Да, и так. Либо, наоборот, поиск верного ответа. Это всё бинарность классическая. Классно бы посмотреть и подумать над выходом из бинарности, когда вместо «я хочу либо великолепно, либо никак» мы имеем третий вариант. Почему не ответить «я ищу что-то новое»? Эта теория называется «медиум дизайн». Есть отличное короткое эссе на эту тему, написал Келлер Истерлинг. Оно как раз о поиске того, как человеку формировать собственное отношение к искусственному интеллекту не как к поисковику какому-нибудь, а к чему-то, что просто есть рядом.