Some considerations on how the human brain must be arranged in order to make its replication in a thinking machine possible

For the most of my life, I have earned my living as a computer vision professional busy with image processing tasks and problems. In the computer vision community there is a widespread belief that artificial vision systems faithfully replicate human vision abilities or at least very closely mimic them. It was a great surprise to me when one day I have realized that computer and human vision have next to nothing in common. The former is occupied with extensive data processing, carrying out massive pixel-based calculations, while the latter is busy with meaningful information processing, concerned with smart objects-based manipulations. And the gap between the two is insurmountable. To resolve this confusion, I had had to return and revaluate first the vision phenomenon itself, define more carefully what visual information is and how to treat it properly. In this work I have not been, as it is usually accepted, biologically inspired . On the contrary, I have drawn my inspirations from a pure mathematical theory, the Kolmogorov s complexity theory. The results of my work have been already published elsewhere. So the objective of this paper is to try and apply the insights gained in course of this my enterprise to a more general case of information processing in human brain and the challenging issue of human intelligence.

💡 Research Summary

The paper opens with a personal anecdote: the author, a veteran computer‑vision practitioner, has long believed—along with much of the community—that artificial vision systems closely emulate human visual perception. A sudden realization shattered this conviction: human vision and machine vision share almost nothing in common. Human vision, the author argues, is fundamentally a meaning‑oriented, object‑based process that extracts and manipulates high‑level concepts, whereas contemporary computer vision is a pixel‑level, massive‑calculation pipeline that treats images as raw data to be processed exhaustively.

To explain this gulf, the author abandons the usual biologically‑inspired narrative and instead draws on Kolmogorov complexity theory—the formal study of the shortest possible program that can generate a given data set. In this view, the brain’s primary function is to discover the minimal description (the “shortest program”) of sensory input, thereby compressing information to its essential algorithmic core. The compressed representation is then re‑expanded (or “decoded”) using stored priors and contextual knowledge, allowing rapid, goal‑directed behavior. This compression‑reconstruction loop is recursive: the brain constantly compares its reconstructed model with incoming sensory streams, correcting errors through feedback.

The author identifies three structural hallmarks of this process:

1. Hierarchical Compression – Low‑level sensory features (edges, colors, motion) are progressively abstracted into higher‑level symbols (objects, scenes, relationships).
2. Recursive Feedback – The compressed code is iteratively refined by matching it against fresh sensory evidence, minimizing prediction error.
3. Distributed Parallelism – Multiple cortical areas operate concurrently on different objects or relations, while integrative hubs (e.g., prefrontal cortex) synthesize a coherent global model.

From these observations the paper derives a set of design principles for a “thinking machine” capable of replicating human‑like intelligence:

• Compression‑First Architecture – Instead of feeding raw pixels into deep networks, an initial module should seek a Kolmogorov‑near‑optimal program that succinctly describes the visual scene. Techniques from algorithmic information theory, program synthesis, and minimum description length can be employed.
• Reconstruction‑Driven Reasoning – The compressed code, combined with a knowledge base (learned priors, semantic graphs), is used to reconstruct objects and their interrelations. This step replaces the conventional “feature‑to‑classification” cascade with a generative, model‑based inference.
• Decision Layer Built on Reconstructed Semantics – Action selection, language generation, or higher‑order cognition operate on the reconstructed symbolic representation rather than on high‑dimensional feature vectors.

The paper stresses that intelligence is not a function of raw computational throughput but of information quality—the ability to extract, compress, and reuse meaningful structure. Consequently, future AI research should shift focus from “more data, deeper nets” to “more efficient compression, richer re‑use.” The author proposes integrating meta‑learning, program synthesis, and minimum‑description‑length objectives into a unified learning framework that explicitly optimizes for algorithmic brevity.

In the concluding section, the author argues that any realistic attempt to replicate the human brain must abandon the current data‑centric paradigm. Only by embedding Kolmogorov‑complexity‑inspired compression‑reconstruction loops—mirroring the brain’s hierarchical, feedback‑rich, and parallel architecture—can we hope to build machines that think like humans. The paper thus calls for a radical re‑orientation of AI research toward algorithmic minimalism and semantic reconstruction as the core of machine cognition.