Уровень 0 · материалов: 11
Документы должны быть посвящены теоретическим подходам к созданию полноценного интеллекта и критике текущих нейросетевых методов, но не должны ограничиваться только проблемами распознавания речи.
Общие признаки: критика статистических моделей и LLM, поиск механизмов истинного понимания смысла, сравнение машинного и человеческого интеллекта, необходимость перехода к когнитивным или онтологическим архитектурам, проблема AGI (Strong AI)
Группа выше: Сильный ИИ: возможен ли он и каким путём
Смысл: The main idea is that computer 'understanding' of text is a functional capability achieved through statistics, ontologies, and machine learning, and while universal human-like understanding is not yet achieved, it is a matter of system integration rather than a fundamental impossibility.
The author argues that machines can 'understand' text by solving specific tasks using NLP and statistics, suggesting that the main hurdle is integrating these specialized tools rather than a lack of computational capability.
Смысл: The main idea is that true text understanding is an interactive process between a message and an interpreter's vast, non-algorithmic cultural and situational context, which cannot be replicated by computational devices due to the exponential complexity of human knowledge and the likely non-algorithmic nature of the brain.
Computers cannot truly understand text because meaning depends on complex, exponential cultural contexts and non-algorithmic cognitive processes that transcend mere computation.
Смысл: The main idea is that modern LLMs have transitioned from simple statistical token predictors to complex reasoning systems. The author argues that intelligence should be defined by functional utility and the ability to generate new, verifiable knowledge rather than by biological mimicry or outdated mathematical metrics.
Vladimir Krylov argues that viewing modern LLMs as mere token predictors is technically illiterate, as they have evolved into reasoning systems that can create original knowledge when paired with formal verification.
Смысл: The main idea is that true intelligence is based on the ability to extract abstract patterns (eidos) from minimal data using a categorical, multi-sensory approach, and that Strong AI can only be achieved by shifting from 'Big Data' architectures to a model of 'Computer Eidetics' based on human cognitive philosophy.
The author proposes 'Computer Eidetics' as a path to Strong AI, arguing that intelligence stems from extracting abstract qualitative patterns from small amounts of multi-sensory data rather than relying on the massive datasets used in Deep Learning.
Смысл: The main idea is to build a general AI by explicitly modeling the psychological structures and cognitive development of humans (top-down approach) rather than relying on statistical neural networks (bottom-up approach), enabling true 'understanding' and natural interaction.
The author presents a top-down AI development method that replicates human psychological structures and a semantic network based on child development to achieve human-like cognitive understanding.
Смысл: The text argues that while humans have successfully amplified their physical senses through technology, they have failed to create a truly thinking artificial intelligence capable of understanding meaning and knowledge in natural language.
Humanity has mastered physical technical amplification but remains unable to create artificial intelligence that can truly understand and operate with knowledge and meaning.
Смысл: The main idea is that true artificial intelligence cannot be achieved through statistical probability (LLMs) but only through hybrid architectures that combine formal logic, symbolic meaning, and cognitive dynamics, all while being grounded in a genuine Theory of Mind and ethical accountability.
True AI requires a hybrid of logic, symbols, and cognitive processes, whereas LLMs are just sophisticated statistical tools lacking actual understanding and agency.
Смысл: The text proposes that the path to creating Strong AI (AGI) lies in shifting from 'definition-based' models (like current neural networks) to a 'context-meaning' model. The author argues that human understanding is based on applying contexts that transform data into meaningful interpretations, a theory he claims is hidden in ancient philosophical and religious allegories.
The author argues that Strong AI requires a 'context-meaning' model rather than a 'definition-based' one, claiming this insight is mirrored in both brain architecture and ancient religious texts.
Смысл: The main idea is that current Artificial Intelligence is not truly 'intelligent' because it mimics reflexive behavioral patterns (Psychic Technology) rather than the analytical and ontological processes of the human mind (Reason Technology), and only the science of brainetics can bridge this gap by introducing knowledge-based world-modeling.
Modern AI lacks true intelligence because it relies on simple conditional programming instead of the complex, knowledge-driven reasoning and world-modeling characteristic of human consciousness.
Смысл: The main idea is that both the human brain and modern AI operate as generative, predictive engines that construct models of the world to minimize uncertainty, though they differ fundamentally in energy, embodiment, and consciousness.
Human perception is a controlled hallucination based on predictive coding, a principle that mirrors the architecture of modern generative AI like LLMs and diffusion models.
Смысл: The main idea is that current neural network architectures are an evolutionary dead end due to their opacity and lack of genuine reasoning. The author suggests replacing this 'black box' approach with a formal, ontological framework that emphasizes stability, measurability, and agentic decision-making.
The author posits that current AI is a fragile 'black box' of pattern matching and proposes a new ontological architectural framework to achieve true agency and stability.