Hybrid Model


1. Imaginative thinking, independent of natural language

Imaginative thinking is a method and form of mastering reality, characterized by the indivisible unity of sensory and semantic moments. In other words, it is the ability to understand the world and to reflect attitude towards it, own thoughts and experiences when using the system of images.

The core processes and stores information in its unique and independent of the natural language form. This makes the system flexible. We do not depend on any specifc linguistic representations or interpretations of a particular language. Now the system works in Russian and English.

[IMPORTANT] In order to add a new language of communication, it is enough to implement a module for a specifc language (1 - 2 month work) and all the knowledge and experience accumulated by the core in other languages will be automatically available in a new language.

What does this give?

a. Moving from topic to topic

In dialogue, people can change the subject frequently. This allows you to continue to communicate, not annoying each other, fixing on one topic. The intellectual assistant must also possess this ability.

If the topic is exhausted or close to exhaustion, it is necessary to change it. To do this, from the context of the conversation it is necessary to select the most close and logical continuation of the dialogue. This is impossible without a good picture of the world, which the AI is gradually building. This makes it possible to conduct a long, meaningful and interesting dialogue.


Human: Do you like Milan?

AI: Yes, I like.

AI: Have you been in Milan?

Human: Yes

AI: Do you like Italian food?

Human: Yes, I do

Human: What Italian food have you tried?

AI: Lasagna and pizza

b. Language mixing

While communicating in a non-native language, it is difficult for a person to find a rarely used word. And it happens that in the current language there is no word at all expressing something in another language. It is very useful if it is possible to build the sentence structure in one language and, if necessary, add words in other languages. This simplifies communication and adds the ability to more accurately convey thoughts.

This is an important and global feature that demonstrates the work of the figurative thinking of Artificial Intelligence.


Human: What is Байкал?

AI: Baikal is a lake.

Human: Где it находится?

AI: В России

Human: What is similar between тарелкой, кастрюлей и ложкой?

AI: They are dishes.

с. Unlimited questions in one message

In everyday communication, we rarely speak by the one sentence. More often it is mini-stories from several sentences. You can immediately ask a few questions, describe the situation, describe the situation and ask a question and etc.

Artificial Intelligence should be able to perceive such a set of sentences and take them all into account when constructing their phrase.


Human: Tell about Milan. What is a maximum depth of Baikal? Where is it located?

AI: Milan is a town. Area of Milan is 182 km². 1642 meters. Russia.

Human: What is Everest? В какой горной системе находится к6? Какая высота у к6? How many wheels does a bike have? And car? And engines?

AI: Everest is a mountain. Гималаи. 7 282 метров. 2 wheels. 4 wheels. 1 engine.


2. Weak hardware

The AI is able to work on very weak hardware (even Raspberry Pi) without any limitations. If the Internet is missing, then Alexa or Google Assistant do not understand you because gadget is just a transmitter. It's time to take a step forward.

This means that the AI will be able to accompany you wherever you need: on the phone, on the PC, at home, in the car - everywhere.


3. No internet

The AI can work completely isolated from the Internet. That is, we can create completely isolated systems that will be completely independent of external factors, only electricity is needed. This is an AI who cares about your personnal data.

It is also possible to combine: the basic knowledge and your personal data are stored offline and public big data on the 5th, 6th, 7th (common for all users) memory layers online.


4. Easy and fast learning

Training takes place once, there is no need to spend thousands of iterations of training. In addition, new knowledge does not destroy the system.

For example, it is possible to make a highly specialized consultant or car driver system, and then expand the scope of knowledge and get a full-fledged person out of a narrow system.

Special feature

a. Education in dialog

Artificial Intelligence must memorize during the various facts that were reported to it, or the conclusions that it draws on the basis of such facts. This knowledge should not be lost, and used in further communication. This behavior suggests that the interlocutor listens to you and is interested in communication.


Human: Where are date fruits grow?

AI: In Spain

Human: Remember that date fruits are grow in Italy.

AI: Okay

Human: Where are date fruits grow?

AI: In Spain, Italy.

Human: Are date fruits grow in Italy? What fruits are grow in Italy? In which country are date fruits grow?

AI: Yes. Date fruits. Italy.

Human: Forget that date fruits are grow in Italy. Where are date fruits grow?

AI: Okay. In Spain.


5. Transparency of thinking

We can always see and explain how the decision was obtained. Neural networks, for example, are black boxes. No one can fnd out and see what is happening inside them and why they make concrete decisions, often erroneous.

Our system, on the contrary, is completely transparent; we can trace the entire chain of judgments and conclusions. In the same way, the system itself will be able to explain its state, why a certain answer has been received, or, for example, why it is impossible to cope with the task


6. Different knowledges

The Hybrid Model of knowledge representation is able to store and process various types of knowledge. Sounds, facts, effects, photographs, images, and so on — everything is stored in one flexible structure.


7. Modularity

We can assemble a solution suitable for certain tasks from the core components, which will use all the power of the core and will not include models unnecessary for the task. This can save resources. The system is able to work on weak hardware.


8. Multilayer memory

The memory is a very important model that make AI able to keep up the dialogue. AI understands whether you are changing the topic of the dialogue or, for example, the property and correctly answers the question.

The memory consists of seven levels and includes:

  • operating;
  • operational;
  • permanent;
  • pending;
  • structured;
  • marked;
  • source.


Human: What is Baikal? What is its the maximum depth? Where is it located? And Lake Michigan?

AI: Baikal is a lake. 1642 meters. Russia. USA, Canada.

Human: How many wheel does a bike have?

AI: 2 wheels

Human: And car?

AI: 4 wheels

Human: And engines?

AI: 1 engine

Lets stay in touch. Subscribe to our newsletter.