The story of Combinatorial AI

We are a team of AI developers with the background in physics and mathematics, specializing in cybernetics and electronics inspired by complexity and power of mind, the effective models of its nature proposed in the 20th century and the technological advances of the 21st that can bring the ideas to life.

Our interest in AI began in 2005 as we embarked on the quest to comprehend the essence of intelligence as a physical phenomenon. Our approach to intelligence was to construct a functional model that could elucidate the known facts of its operation that seemed to be in contradiction with the working models of AI. By the year 2012, we had formulated the physical principles underlying the intelligence, which were meticulously documented in a monograph.

The book “Physics of Intelligence: from the principles of self-organization to the model of thinking” was published in 2014 under review of Prof. B. Velichkovsky.

Subsequently, from 2014 onwards, our energies have been channeled towards crafting algorithms for the implementation of the proposed model of AI. We've designed the architecture and algorithms that set us apart from the pervasive trend of neural networks development. This distinction aroused from our understanding of the deep limitations inherent in the neural networks technology, particularly in the context of understanding the entirety of the potential of mind.

One of our initial points was to enable our AI to recognize patterns, combine them, and utilize these combinations to navigate problem-solving, all that while continuing to learn. We consider the ability to combine and reapply patterns in untrained situations as a pivotal facet of intelligent agent operation, which led us to term it combinatorial AI.

The proficiency of mind in tackling intricate problems hinges on the speed it acquires new knowledge and its aptitude for discerning structural regularities within complex problems, even at the abstract level. Through the process of agent development and functional testing, it became clear that speedy learning needs the use of a hierarchical training methodology. This mirrors the human training system, progressively advancing from simple to complex tasks, which we found to be a requisite for the cultivation of machine understanding. This approach is best described as intensive AI.

The framework of intensive AI hinges on stage-by-stage training, starting from memorizing relations and patterns and gradually transitioning to the mastery of abstract problem-solving methods.

The future

Our current goal is the incorporation of the intensive AI technology into practical applications across diverse technical domains.

We are actively seeking collaborative partners in projects focused on the creation of intelligent automation technologies for research and development.

If our approach piques your interest, we invite you to a detailed discussion of the methods and a pathway for integrating the intensive AI in your technologies to face the challenges you encounter.

© 2024 apparatus scientia d.d.o