Igor Ashmanov about the future of home robots. Home robots: on the eve of a tornado

Good day, Habrovchane!

This post, which is a reworked ekstexting of Igor Ashmanov’s speech at the Skolkovo Robotics Conference 2015 in the Skolkovo Hypercube on March 21, the Lexi project ( VK , FB ) opens a series of posts about the project, the technologies used, its own developments and experienced team experience. This post is a kind of introduction, a brief overview of the industry and voiced to the reader a number of topical issues.

Attach the full video of the performance to the post:
')

Hello, my name is Igor Ashmanov. I have been developing software for almost 30 years. Now, basically, any Internet projects.



It so happened that I contacted a pair of startups, a couple of projects on robots (Ed. Notes: 1 , 2 ) . I would not like to use the word “invest” in relation to startups in robots, because, in my opinion, there is no investment in this area right now.

For example, in front of me, Dmitry Grishin said here that he is investing in startups with robots. He studies the market, how these startups plan to make money, studies business models, the audience and so on. Formally, everything is correct, everything is in venture science. And it seems to me personally that there all this cannot be called an investment. It is, in fact, just investing money in the investor’s own education, in order to understand the market and understand what teams are there, what difficulties, what problems, where, what and so on. Just buying knowledge about existing and future robotics. Because, in fact, in this robotic industry, in that part of it, which I, in particular, do, are home robots, or remote presence robots - there is no market there and it will not be for a long time . No business models and no buyers. There, while in general, this is not about the market, but about one certain very important, special thing, about which I will now tell.

What is artificial intelligence

We begin by clarifying the general concepts. Here everyone is talking about artificial intelligence. I will say a few words about it so that we speak the same language. And artificial intelligence is something that must now be present in robots .

We must immediately make a reservation that there are two understandings of artificial intelligence .



The first is a domestic, mass, sort of Hollywood understanding, which means: an anthropomorphic robot, a robot that naturally speaks, communicates and so on, while it is kinetically capable, so to speak, capable, it moves, it can take things, do work. He necessarily has an awareness of himself, emotions, some kind of relationship with people, claims to them, love and all that. And in the future, of course, it comes to rivalry with people, to conquer the world, to the terminator with a weapon in his hands, and so on.

This Hollywood notion of robots is pretty much shaping the industry: those startups and those attempts to make robots that are emerging now. Actually, in my opinion, this is utter nonsense, that is, it’s such an illusion that people in the robotic industry are chasing the same way as the Pentagon often makes weapons, say, supernew generation airplanes based on the Hollywood films about these planes. Overcomplicated, too expensive and not working in essence.

There, in the military industry, of course, everything happens more cynically, because we are talking about a lot of money. They order a blockbuster to a well-known director about a super-hyper-mega-plane, an invisible man who hangs in the air and gets up different tricks, then comes to Congress and says, “Have you seen? Cool? Now give money for it, we will be like in the movies. ” And it turns out super-expensive aircraft that does not solve the tasks of the troops, and the promised tricks, in general, performing somehow. Here is the same story with artificial intelligence.

Well, what does artificial intelligence mean from a developer's point of view? Here I am since 1983, that is, for more than 30 years, I have been working in this industry. Actually, he came from Mekhmat directly to the Department of Artificial Intelligence of the Computing Center of the Academy of Sciences to work in 1983. Those who deal with this AI know a more boring definition of AI: this is such a bundle of optimization methods that should imitate human functions . That's all. And, of course, the imitation of human functions is completely different from the way a person performs them. It is clear that cars that make people, and which drive faster than a horse and fly faster than birds, do not copy horses and birds. They do not move their legs, airplanes do not flap their wings (by and large, most of them, at least) and so on.

There is, of course, the direction of artificial intelligence, which is to understand how the human brain works, and repeat it in metal and electrons. This is, in fact, a marginal direction, and from my point of view, a dead end. There is no time to explain in detail, I’ll just say that this is my personal opinion.

Thus, AI is a scattering of tasks for imitating various functions of a person . The task of creating a computer that is aware of itself is not among them, at least among the developers, and not charlatans.

Well, these optimization methods (well, or machine learning there, as is now fashionable), they gradually transfer part of the tasks of imitation of human functions, part of these functions, into the category of solved problems. Before the AI ​​problem is solved, it seems that there is some magic in it. As soon as this problem is solved, it loses all the romance, the whole veil of romance that surrounds it. For example, say, 50 years ago, only very cultured, educated people, teachers, and so on could check spelling. Now, of course, nobody even thinks about this as artificial intelligence. Everywhere this function is, the same applies to the T9 algorithm that you have on your phone, the same applies to search engines and other things. That is, as soon as the tasks of searching for pictures, recognizing melodies in Shazam, antivirus, antispam and so on - and these are all artificial intelligence programs, all recognition programs - as soon as they become everyday, everyone forget that it is artificial intelligence. For users, this is just like a plumbing, there is some useful thing, it works, and that's enough about it. Nobody thinks that the T9, say, in the phone is a rather complicated program of artificial intelligence. Just think, words picks up. And there all the same cunning methods are used, which are used to solve unsolved problems of AI.

Let's say, the problem of speech recognition , good recognition of any speaker in a noisy audience, or at least in a large audience from a distance, and so on, is not completely resolved . Unfortunately, the problem of normal speech synthesis is not solved , that is, the synthesized speech is still easy to distinguish from the voice of a person, it still sounds quite debilityful and without intonation.

Machine translation is not normally done , with acceptable quality, surprisingly, although, generally speaking, all the loud promises of the field of artificial intelligence began with it, it is such an Everest for applied linguistics. And there are two of them in fact, Everest and K2, two peaks, the second top is a dialogue in natural language. And this, too, in general, is still not a very resolved problem.

You all know the online translators who are in Google or Yandex, who translate the pages; you yourself know that they are in fact not a translator, this is just some rather clumsy tool that helps to understand the meaning of a page in an unfamiliar language. Translation into an unfamiliar language with its help is not recommended.

Recognition of the meaning of the picture (not the search for similar, but thematic tagging) is not done.

Machine vision with acceptable recognition of three-dimensional scenes is an unsolved problem.

Since all these tasks are not solved, they still retain some such romance, and it seems that in the future artificial intelligence, in a future robot that is a human assistant, such Jarvis or Friday, they will all be solved, and this is amazing. .

In fact, it is clear when all this will be done, but sooner or later it will be done and even probably the robot will imitate a person well, there will be no romance in this, of course. Everyone will look at it as something familiar, useful and uninteresting, like a spreadsheet or booking tickets via the Internet.

I think we will still distinguish a robot from a human being quite easily; in fact, in my opinion, even here in Skolkovo, a pair of domestic Turing tests will be made, and it will be possible to see how a robot is recognized, how a person differs from a talking robot (Ed.: Lexi also participates in it) .

Thus, we are talking about the imitation of human functions (perhaps completely by some other, inhuman methods), and not about the fact that the machine will start the mind. I personally do not believe it. This is, in fact, for filmmakers and journalists. You can discuss why, but at another time. (Ed .: Dear reader, we are waiting for your opinion on this issue in the comments)

And now I would like to say a few general words about robotics, and about home robots.

About home robots

There is such a researcher Geoffrey Moore, who has written some great books about the laws of the development of technological industries in general and technological startups, in particular. One of them is Jumping the Abyss . Maybe some of you read it. I highly recommend reading. And the second, his next book about the same thing - " Inside a tornado ." He introduces the concept of “technological tornado” in it, the explosive development of a new technological industry. Such a "tornado" before our eyes has arisen several times, when a huge surge of certain industry took place when a new breakthrough technology appeared. And there arose at once many users, money, manufacturers, and so on. Careers were made, careers collapsed, fortunes were made, the lifestyles of millions were changing, and so on.

We have seen such tornadoes with you several times . This, in particular, the creation of a personal computer , which had a hand in Jobs and Gates.

Then, this is a software tornado, that is, fashionable programming in the late 80s and early 1990s, when any girl thought it was an honor to walk with a programmer. Then in the mid-1990s, an Internet explosion occurred. Giant companies appeared from nothing again, from scratch, dizzying careers and fortunes were made, loud crashes and all that happened. Then we saw mobile phones, but they were there, just a little away from us, IT people. Well, then, here you are, smartphones and tablets.

In all these cases there were general processes and patterns. The most interesting thing for us is how the state of the future industry looks like “on the eve of a tornado”.

Condition "on the eve"

What happens on the eve of such a tornado? First of all, this is the absence of any market, and the most important is the fog in the brain. Everyone feels that something is being prepared, but nobody knows what. In particular, such a story was when a personal computer went on stage. Dmitry Grishin said that Visicalc allegedly pulled them all out and there was some kind of game, but this, of course, was not true. He, fortunately, is young and did not have time to participate in this tornado. I still managed to seize his end, because I started in the mid-80s. I then began to program the spelling project, which we then embedded in MS Word, about the 87th year, when the first personal computers had just appeared in the Soviet Union. Of course, it was not because of the explosion that a specific application was created. But why. When all this was done on the knee in different garages, all these people knew each other, they all met on the same get-togethers. It was relatively speaking there, 17 models of different, incompatible computers, some of which, for example, did not have a screen or had such a small green screen on glowing wires, some of them were programmed by turning pens, verniers (there was no keyboard, etc.) .

Method of use - the main condition of the tornado

What happened? Why did this tornado start? Because someone (and they were quite charismatic people, very energetic, and so on, like Jobs) suggested a way to use this device .

Everything decides, from my point of view, the method of use. What it is? I'll explain now. The way it is used is what Jobs did with smartphones. Smartphones were made several years before the iPhone appeared. Nokia has been making these smartphones in countless numbers. I bought them several times. They were quite painful, they were hard to use. Some of them were already with touchscreen. Jobs essentially invented nothing, all the elements have already been invented and implemented.

This is also one of the signs of the approaching tornado that the technology already has everything.
Maybe they are not standardized. There are no building blocks, as they said today. But, nevertheless, it is not necessary to invent technology.
We need to put them together and say - this thing, which is called this, and this is how it should be used.

To propose, in fact, a use case, a paradigm. I personally call it the "method of use."
We must give users this concept, give the manufacturer a direction of development. Then there is a market and users. And then a technological tornado comes and explodes overhead. It is clear that in robotics, in particular, in the home robot industry, with which I personally deal, this has not yet happened.



Another example is the automobile industry. She was exactly in the same condition "on the eve" until Ford built his conveyor and started producing the car we know. In fact, the car that he produced is quite a bit different from the modern car, according to the main components and functions.

Then it was, relatively speaking, 300 garages around the world where they did anything: two-wheeled, six-wheeled self-runners, with a windshield, without it, with doors, without doors, with a steering wheel, with a joystick, with handles of some kind, whatever , with a roof, without a roof, with one chair, two, five, and so on. What did Ford do? He proposed the concept that we know, that is, the car, most of the elements that are now in them.

And he managed to impose it. That is, in fact, in order to impose a method of use, you need not only a pleasant, understandable, obvious way of using a thing, which is a contagious enough mental virus, but you also need the founder’s charisma and, perhaps, a very dense crop of this virus. so that the density of infection is immediately large.

At Ford, it was just a conveyor, which produced a lot of cars at a relatively low price. Jobs also had such opportunities to release many iPhones at once and to advertise them very strongly.
And after that, here it is a cloud of the most diverse freaks ... I apologize that I did not draw all this, here all the reports with pictures. But it seems to me that smart people have enough words. All this is a huge variety of the most incredible models of the most incredible devices, all are compressed into a barrel of very similar products and business models, within which real competition begins, real business and so on.

Here's another example - the Internet to Jim Clark. Jim Clark, maybe none of you know or know all, but this is the man who actually made the Internet. He hired the author of the first non-commercial browser, made a real browser, namely Netscape Navigator, and then everything exploded. Not before. Yes, by itself, the Internet was already growing fast enough, like weeds, Mosaic browser was already, but nevertheless the Internet was what Jim Clark did (which he was a very unusual guy, he had done before that Silicon Graphics for all the modern special effects films that we know).
Accordingly, the Kalashnikov did assault rifles (assault rifles) what they are now, etc.

Probably, there are some real, technical, financial, pragmatic limitations in the way in which things can be used. Technically and financially, it is probably less profitable to make a six-wheel vehicle than to make four wheels. Nevertheless, the generally accepted way of using things is often strange (for example, Chinese sticks and hieroglyphs, ties or wrist watches), but it’s just that historically and everything, and we all use it without asking why.

Home robots on the eve of a tornado

I want to talk about home, personal robot, and here's why. It seems to me that things such as remote presence robots or fire robots, military robots or quadcopters, or even some kind of kinetic robots, toys — they are niche, that is, a real technological tornado in robotics will develop where millions will be involved, or rather hundreds of millions of users.

What it is? Most likely, it will be some kind of personal robot. We are fiction, Hollywood, as well as fantastic books have already shown what it is. This is some kind of robot that lives with people, like a butler, like Jeeves or Jarvis, and participates in everything in family life, in human life and so on. Small explosions will take place in many places, now there is an explosion in the area of ​​quadcopters, more precisely, copters, controlled / autonomous, and so on. But not everyone will have such a copter. But a personal robot, just like a personal smartphone, ultimately strives to be with everyone; personal computer too.

So, to participate in a tornado with robots, most likely, it will be a personal robot. Everyone will have such a personal home robot or will strive for it. That's where the next tornado happens. This is what in English is called “the next big thing”, the most important next thing.



What now? We see just such a classic state of “On the Eve”, that is, fog in the brain, no one understands what a home robot is, there are a lot of them, they are all unnecessary, all of them are overcome by the curse of the dusty corner.

Curse of the Dust Corner

This is how much I have not seen robots that my friends, relatives and I buy for children, for myself, the curse of a dusty corner lies on them all. Not later than in a week, two, this robot is discharged in a dusty corner and nobody else does it. Because it's boring, because it's too lazy to charge, because it's not needed.
For some reason, no one has yet overcome this barrier: to make a robot that would last at least a year.

Yes, there are vacuum cleaning robots, this is a very niche thing, and some people have it that they like order very much, and who are not too lazy to get this robot out of the wires, do it and so on. Robot vacuum cleaners live for a relatively long time, but for most of my acquaintances, they also fell into a dusty corner - just in a month or two, and not in two weeks. Only very boring and careful people have a working vacuum cleaner one year after purchase.

But there is an interesting hint that shows the general direction. One of my company partners told me:
“Here my sister also bought a vacuum cleaner; she says that he, of course, cleans up, but the most important thing is that the grandmother now has someone to talk to. ”



It is necessary to understand, at the same time, that the vacuum cleaner does not actually speak, it sometimes says in a nasty voice something like: “Error 502” when it is confused, or does not speak at all, but beeps when it is confused somewhere or in it something broke. . , , - . , , , , : «, , , , », , .

, , , . , , «, , .» , . . , , , , , , .

, , , , , , , , , , , .

?

.

— . , - , . , - , (, , , , , ), , , , - . , « », .

«» - , , , , , : . , , , . , , , , , , , , .

, 2-3 , , , . , ; — , , . , , , — , . , (, ), .

: , , , (, ).

: , . , , , . , « ». , , , , , , . , .



, , , , . , . : - — , , , .

. , -, , , 10 , , . , - - , . , , 10 .

. , , . – , . . , - . , , . , , . , . , .

, , ? , , , . , . , , -, , .

, . , .

?

, — .
, , , , . , , , , .

, . - , - .
, , . , , .

«» «».

–

. , . , — . , - . , , (SLA) , , , -.
— . Siri , . , . , , -, , 10: , , « », wi-fi , .

, . « ». . , . , , , .

, . , Siri . Siri , , , , , — - , . .

, . , , . , , , , , , . , , , , , – , , . , , – . — , , .

, . 2000- , , AIML — artificial intelligence markup language, Richard Wallace. Alice.

. , . . , . . .

, , 2002 , , : , , — . . , , : , , , , , .
, .

www.iii.ru , . - 2 . . . ICQ, ICQ , …

The average session was several hundred replicas, and the maximum was 1600 replicas. , 10-12 . «», , , , . , , — , , . — . , , . , .

. , . May be. , , .

, . , , , , , .
, , , . , . . — , . , , , 10 , , — . 100% — , . , , , , . .
, , ( ) ( -). : ?

, , . ( , -). , : , , - , . , , . « » , , , .



, , , , . : «» , , , - «» . , , , .. , , - , — ( ) .

What are the problems with the development? Speech recognition on board. Most of the talking projects, in fact, sit on the recognition of Nuance or Google, which should be addressed in the network. This is usually 3-4 seconds delay, this is an impossible time, no dialogue is obtained. A person at this time starts to get nervous, asks the next question, he disappears altogether unheard or the interlocutor gets off, begins to look for the answer to it, more precisely, to recognize it ...

In general, in fact, the situation with the recognition in the world is very bad: almost all captured Nuance, this is a monstrous patent troll that devours everyone, and then rolls out the huge prices for all its services. Well, now Yandex has released its API, we'll see. There are a couple of companies in Russia that Nuance just could not reach. I hope that our state will protect against it in the end.

Recognition of surround sound, few people do. In fact, one thing is when someone shouts Google directly to the smartphone, and another thing is to recognize you in a room with love and distance.
Because the essence of such an interlocutor is the seizure of the channel, so that you can wake up, and when you do not have a screen at hand, you could in the air ask what you need. What time is it? How is the weather there? Or something else. Do I have an email?

But, to be recognized, conditionally speaking, it is necessary that either there is an interlocutor in each room, or around the house in each room there are intelligent ears from this interlocutor, and he would stand in the kitchen. But, in any case, this surround sound is more or less such a home volume of 3-5 meters, it is necessary to recognize, and this is a rather serious problem. It is clear that it is necessary to recognize interlocutors, that is, it would be good to recognize, identify by voice, no matter how hoarse there is, to call him by name, and so on.
A home robot, it means - he is dealing with a family, as a rule, and not just there with a lonely geek who bought this thing to try.

It is clear that there should be a variety of communication, that is, games, different topics. He must intercept the initiative, that is, the linguistic intellect must be quite serious. We need a smart user model, that is, the interlocutor must be trained, he must learn more and more about his masters, he must memorize something about them. Building a user model is a pretty serious theory, which, in fact, very few people know how to do. We ourselves also strongly think about this, because how to memorize the right, but to forget the wrong, how not to retrain, is a pretty serious thing.

Self-learning and self-updating, I already said this, that is, of course, the interlocutor must download updates as well when he sees the network, but he, generally speaking, must work without the network — otherwise there is no point in it. Work with the Internet or everything on board? Everything should be on board from my point of view, and the Internet is needed to download updates.

The synthesis of speech and intonation, in fact, is also not yet very solved problems. While robotic voices are pretty lousy, monotonous. There is, by the way, a very unpleasant effect. If you have ever listened to reading the news, you might notice a very interesting effect on the human brain. I noticed it when we did the synthesis for dictionaries and for news as early as 95-97. You listen to the reading of the news in a synthesized voice, and at a certain point you notice that you do not understand anything that he reads. You hear, it seems, the rattle of the Russian language, and you cannot understand. The brain suddenly said: “Ah, I understand, you are deceiving me, this is not a person who is speaking, I refuse to understand this.” This is a real effect. The same with letters. We synthesized the text of the letter. We had a "Letter", which synthesized the letter. The same story: from some letters, the eyes jumped, slipped, could not sink. The brain refuses: it feels something unnatural, refuses to understand. Therefore, we need a very good intonation, effects that break up the dialogue into small pieces.

In short, the way to use all this remains to be shaped - what is a “home robot”. Perhaps there will be some kind of dynamic activity in it - driving around the house and so on. Perhaps there should be machine vision in order to at least recognize the hosts or find out that someone has entered the room and is silent in order to start a conversation with him. Maybe you need to recognize pets.

The functions of the butler should be - control equipment and all that. The governess - the simplest training is not difficult to integrate into infa. Arithmetic or something else, most likely, will be. Secretarial functions, for example, “Tell my wife that I will be later, but I will buy bread”. Analog notes on the refrigerator, of course, should be. An answering machine, an alarm clock, probably, should be done in any case. And the physical assistant in the house - it seems to me, it is nonsense. It will be so expensive that it is much easier to simply hire a grandmother.

Entertainment, feeding and walking of animals, most likely, will be niche business. These will be some special robots that do not have anthropomorphism and even intelligence in general. They should have intelligence about the level of the animal.



Sex is a very hot topic. I do not think it is in our niche and that it will spread widely. Probably, there will be some dolls in sex shops with artificial intelligence, as in the famous joke, here I will not risk to tell him. Most likely, it will be some kind of niche, and it certainly will not be inside this technological tornado.

Well, the anthropomorphism in me personally raises big questions. Perhaps, there will be the same effect of transfer of personality, but, unlike the text and even voices, the robots faked for human appearance still cause repulsion and shock. Just like a monkey is very much like a human, but it’s still not like that, and it seems ugly. If we tried to take her for a person, it would be a shock. If you saw a man very much like a monkey, it would cause a feeling of terrible deformity. It seems to me that the robots will be the same. In general, I am personally against anthropomorphism, it seems to me that this is a dead-end branch.

We are waiting or offering

That's all I wanted to say. We are waiting for the emergence of this method of use - someone must produce it, impose it on everyone. It must be either a very charismatic person or a person with huge advertising money. And then everyone builds up behind him a pig and will produce about the same thing and compete within this paradigm.

But the existing efforts of robotics will not disappear - they will be used as niche applications or in the mainstream of general development, in the central “trunk”. The qualifications and accumulated experience will not be lost either. As they say, "there is no extra work."