We model the nature of human intelligence

According to AI researchers such as Melanie Mitchell, analogy is a basic function of human intelligence. She said: «Thoughts are impossible without concepts, and concepts are impossible without analogy». Key intelligence abilities such as reasoning by analogy, generalization, metaphor, comparison, syllogism and prediction are based on analogy. We have created a new way to extract analogy from texts through syntactic predicates, which allows us to create artificial intelligence based on it. Our project implements the possibilities of analogy to create a new generation of AI.

Unlike existing approaches, AI based on analogy has significant advantages:

• Learning in the process of use, solving the problem of "forgetting"

• Deep generalization based on several examples (induction)

• Accurate adherence to rules and examples by analogy (deduction)

• Memory for the facts of the current dialogue without the limitations of the hint window

• Checking and correcting your actions based on generalizations by analogy

• Learning based on "one textbook", not hundreds of gigabytes of texts

• The ability to ask questions in the presence of gaps instead of hallucinations

• Creativity. The ability to discover new interpretations of phenomena known to us

The research is being conducted by a group of developers headed by the author of the project, Alexander Khomyakov. We invite developers who are interested in testing a new approach to creating AI and becoming part of the team to collaborate.

Our principle is that intelligence higher than the intelligence of a person cannot belong to any group of people, countries or communities. This intelligence can only serve the common interests of all people. Therefore, the development will be carried out according to the principle of openness for everyone who is ready to cooperate with the project.

Important differences in the approach

• The new AI structure, built on analogy, allows us to overcome the main drawback of modern neural networks and transformers - the impossibility of learning during use.

• Analogy on predicates allows us to create AI systems with much lower costs both in terms of data and training. To train the analogy, only 200 fiction and popular science books were needed, and not the entire Internet, as in the case of transformers.

• In addition, AI by analogy, even during initial training, demonstrates common sense due to high-order generalization and generation of common sense predicates that are absent from the system's knowledge base.

• AI by analogy has a high degree of generalization that other AI systems cannot achieve. For example, the general concept of an obstacle, which allows you to not crash into it even if the object on the road has never been encountered in the training sample.

• The new approach is interpretable and controllable. The work of AI by analogy is not only understandable at every step, but can also be controlled by selecting predicates for analogy.