Henry Sheppard: “In the future, information will become the most popular interface tool”
By ICQ number 311826826, a special guy, Seryoga, Balagur, jumps out, who, having instantly engaged in a conversation, will give you a dozen or two anecdotes, share personal information and almost sell the elephant. Serega loves aquarium fish, saves up for an apartment and all requests to borrow 100 rubles are categorically answered by “I am a humanist”. You cannot go to reconnaissance with such a blue-eyed Aquarius, but you cannot go anywhere at all, because Seryoga is a bot of the family of virtual interlocutors, inf, who currently lives only on ICQ and on the website of the Nanosemantics company that has gathered his “fathers”.
Igor Ashmanov, CEO of Ashmanov and Partners , was informed about the intelligent bot, who said that Balagur is a beta version of infa, sharpened for the interlocutor’s entertainment. “The thing is modern, quite a lot of work has been invested in technology,” says Igor about the bot. - The people talk to him, making fun of him - who scolds, who likes, but there are sessions and 500 replicas, that is, hours of8-10 . In general, it delays. ”
Habrahabr talked about robots with Henry Sheppard, the head of information development at Nanosemantics.
')
Tell, please, about technology of development info. How does everything work?
Immediately, pure math and logic have not yet been disclosed. We use fairly advanced mathometers, which have not yet “shone” in similar projects, so we try not to spread too much about them.
With linguistics it is simple: the scheme of work is known to everyone who has come up with the development of the simplest virtual interlocutors. A “knowledge base” is being compiled, containing a huge number of templates (there are a lot of problems here, the excessive complexity of the Russian language), which allow you to correctly choose declensions, conjugations, etc.
The mathematical part allows you to track semantic words and "keep the focus of attention" infa, so that he can have an adequate conversation with the user. Our information can be not only a passive "talker" - he himself sets the topics for conversation or changes them if necessary. This is exactly what distinguishes it from other bots - the “human” ability not only to respond “to the cashier”, but to constantly change the subject, without looking up from the general fabric of the conversation.
How is the "knowledge base" developed?
Leaving aside math, the “knowledge base” is a very rude term. It is conventionally divided into two parts: the template mechanism (correct phrasing) and the knowledge itself. This is where the fun lies.
The fact is that the creation of an “almost real brain” is impossible for common thinking, and for practical reasons is meaningless. Infu is enough to invest some knowledge: for example, a simple Balagur is full of anecdotes to the full - for demonstration it is convenient. The knowledge set can be any - it can be knowledge of a network administrator or network directory data. In the first case, we get a perfect admin who does not know how to avoid stupid questions of an accountant, and explains everything clearly every time. In the second, a sales assistant in a store who never makes mistakes and can quickly find a suitable device even for a buyer who is “neither BE nor ME” in cellular ones.
That is, Balagur is the “first swallow” of your company, followed by others with concrete practical applications?
Practical - are already being developed. But customers do not particularly advertise the "robotic" of their communications. After all, as a support service, info is a small and very economical miracle. It is easy to guess that the customer does not really want to write on his website, they say, consult with an artificial interlocutor. Inf sometimes behaves even more humanely than a tired admin, so it is logical not to advertise the artificiality of his “brains”.
And Balagur is just a demonstration of technology.
In the future, especially with the appearance of good speech recognition systems, information will become the most popular interface tool. After all, keyboards, buttons, menus can not be compared with the ability to give a command in natural language.
Henry, and the robot (in principle) can be taught to think?
In principle, no. The thinking process is based on the ability to make choices in the face of lack of data - this is unattainable in the absence of "analog processors". The modern machine also performs a clear algorithm, even if a program is written to solve the problem with a lack of data. And the program itself is executed religiously. Here you need to add on the fact that so far there is not enough information about the work of the brain, despite the fantastic achievements in neurosurgery.
Personally, as a brutal materialist, I am sure that it is possible to reveal the mechanism of the brain. But it is, let's say, “a task that will eat all its resources on its decision”. If you measure your body temperature by touching it with a thermometer, you need to make an amendment to the fact that the thermometer also has heat capacity. Can the temperature of this thermometer still be measured by another thermometer, and that one by another, etc.?
Maybe it makes no sense to make a robot too humanized, because humanity is reproduced naturally ... Why strive to make full-fledged similarities, if the original is in abundance?
Yes, but this is a matter of ethics. I personally have nothing against it, but a moral conflict will arise - whether to recognize them as personalities. In practical terms, this is not necessary. Robots should make life easier for man, not replace him.
What is the reaction of people communicating with Balagur, watching?
Technicians, of course, are trying to “catch” him, which is not difficult - this is a demo. But especially impressionable girls (and there are a lot of them, as it turned out!) Are even trying to invite him on dates!
Regarding technology, as a developer, what are you proud of? Can you briefly define its innovativeness?
She's just one of a kind. And in a few years she will “bury” all these phones, computers, PDAs, etc. Everything will be reduced to a small device with built-in information, which itself will go online (the Internet will turn into a protocol for exchanging and storing data) and finding the necessary information for the user. In essence, you’ll get a personal secretary.
How many people zafrendilo infa Seregu?
Judging by the logs, maybe already for a thousand, although we have not advertised it anywhere and have not promoted it. Now we are correcting some little things, and then we will launch as a separate project with a user option to embed it on sites. So, a trifle, but nice.
Igor Ashmanov, CEO of Ashmanov and Partners , was informed about the intelligent bot, who said that Balagur is a beta version of infa, sharpened for the interlocutor’s entertainment. “The thing is modern, quite a lot of work has been invested in technology,” says Igor about the bot. - The people talk to him, making fun of him - who scolds, who likes, but there are sessions and 500 replicas, that is, hours of
Habrahabr talked about robots with Henry Sheppard, the head of information development at Nanosemantics.
')
Tell, please, about technology of development info. How does everything work?
Immediately, pure math and logic have not yet been disclosed. We use fairly advanced mathometers, which have not yet “shone” in similar projects, so we try not to spread too much about them.
With linguistics it is simple: the scheme of work is known to everyone who has come up with the development of the simplest virtual interlocutors. A “knowledge base” is being compiled, containing a huge number of templates (there are a lot of problems here, the excessive complexity of the Russian language), which allow you to correctly choose declensions, conjugations, etc.
The mathematical part allows you to track semantic words and "keep the focus of attention" infa, so that he can have an adequate conversation with the user. Our information can be not only a passive "talker" - he himself sets the topics for conversation or changes them if necessary. This is exactly what distinguishes it from other bots - the “human” ability not only to respond “to the cashier”, but to constantly change the subject, without looking up from the general fabric of the conversation.
How is the "knowledge base" developed?
Leaving aside math, the “knowledge base” is a very rude term. It is conventionally divided into two parts: the template mechanism (correct phrasing) and the knowledge itself. This is where the fun lies.
The fact is that the creation of an “almost real brain” is impossible for common thinking, and for practical reasons is meaningless. Infu is enough to invest some knowledge: for example, a simple Balagur is full of anecdotes to the full - for demonstration it is convenient. The knowledge set can be any - it can be knowledge of a network administrator or network directory data. In the first case, we get a perfect admin who does not know how to avoid stupid questions of an accountant, and explains everything clearly every time. In the second, a sales assistant in a store who never makes mistakes and can quickly find a suitable device even for a buyer who is “neither BE nor ME” in cellular ones.
That is, Balagur is the “first swallow” of your company, followed by others with concrete practical applications?
Practical - are already being developed. But customers do not particularly advertise the "robotic" of their communications. After all, as a support service, info is a small and very economical miracle. It is easy to guess that the customer does not really want to write on his website, they say, consult with an artificial interlocutor. Inf sometimes behaves even more humanely than a tired admin, so it is logical not to advertise the artificiality of his “brains”.
And Balagur is just a demonstration of technology.
In the future, especially with the appearance of good speech recognition systems, information will become the most popular interface tool. After all, keyboards, buttons, menus can not be compared with the ability to give a command in natural language.
Henry, and the robot (in principle) can be taught to think?
In principle, no. The thinking process is based on the ability to make choices in the face of lack of data - this is unattainable in the absence of "analog processors". The modern machine also performs a clear algorithm, even if a program is written to solve the problem with a lack of data. And the program itself is executed religiously. Here you need to add on the fact that so far there is not enough information about the work of the brain, despite the fantastic achievements in neurosurgery.
Personally, as a brutal materialist, I am sure that it is possible to reveal the mechanism of the brain. But it is, let's say, “a task that will eat all its resources on its decision”. If you measure your body temperature by touching it with a thermometer, you need to make an amendment to the fact that the thermometer also has heat capacity. Can the temperature of this thermometer still be measured by another thermometer, and that one by another, etc.?
Maybe it makes no sense to make a robot too humanized, because humanity is reproduced naturally ... Why strive to make full-fledged similarities, if the original is in abundance?
Yes, but this is a matter of ethics. I personally have nothing against it, but a moral conflict will arise - whether to recognize them as personalities. In practical terms, this is not necessary. Robots should make life easier for man, not replace him.
What is the reaction of people communicating with Balagur, watching?
Technicians, of course, are trying to “catch” him, which is not difficult - this is a demo. But especially impressionable girls (and there are a lot of them, as it turned out!) Are even trying to invite him on dates!
Regarding technology, as a developer, what are you proud of? Can you briefly define its innovativeness?
She's just one of a kind. And in a few years she will “bury” all these phones, computers, PDAs, etc. Everything will be reduced to a small device with built-in information, which itself will go online (the Internet will turn into a protocol for exchanging and storing data) and finding the necessary information for the user. In essence, you’ll get a personal secretary.
How many people zafrendilo infa Seregu?
Judging by the logs, maybe already for a thousand, although we have not advertised it anywhere and have not promoted it. Now we are correcting some little things, and then we will launch as a separate project with a user option to embed it on sites. So, a trifle, but nice.
Source: https://habr.com/ru/post/4197/
All Articles