The Foundation World Model (FWM?): True, Real Universal AI (TRUAI)
Our mission is to model and organize the world’s data/information/knowledge/intelligence and make it universally accessible and useful for machines and humans.
Some assumptions:
Acausal science and technology is a pseudoscience and pseudotechnology
All statistical models are wrong falling short of reality and causality.
All statistical AI is a pseudo-AI technology falling short of reality and causality.
Conclusion. The Tech Stack for AI requests the world foundation models for True, Real Universal AI and TRUAI Society and Economy
Advancements in computer science and technology, computational power, big data, algorithms, and foundation models are the key factors for intensive research, development and investment into generalized AI, like as the 500B Stargate Project.
We advance the Generalist AI Foundation World Model (GAI/FWM) as the man-machine hyperintelligence platform for all valuable knowledge sources and intelligences, human and artificial, including narrow AI models, computer vision, ML, DL, NLP, large language, predictive, generative, symbolic or hybrid models, making just statistical predictions or decisions in special domains using some specialized algorithmic logic by which that AI models operate.
The hyperintelligence platform has the power to simulate the world, understand its contents and data and act intelligently, being able to learn or infer, predict or explain, discover or make decisions and interact autonomously and intelligently via its AI stack:
It is noted that machine intelligence and learning must be the world-reality-universe-truth-science-grounded construct, to be true, real and universal AI (TRUAI).
Developed as the GAI/FWM and trademarked as Ontological AI?, which universe of intelligence is all possible realities, it is driven by the universal computing ontology of the world of data (Omniverse Ontology Platform?).
The OOP? is made up of several core components to structure the world (W), its all possible realities, R, and define all possible interrelationships or interactions (I) between all possible things (T) as entities (E), states (S), changes (C), allowing TRUAI systems to understand, organize, interpret, reason, decide, and interact with the world intelligently and effectively.
FWM? = OOP? = W = <R; T; E, S, C, I>
In the TRUAI's modelling the world, everything and anything is a thing in the world, whether it is entities, substances or objects, states, quantities or qualities, changes, events, or actions, relationships, associations, links or connections, you name it.
Thus the OOP? could virtually simulate all and everything, including the physical universe, the totality of physical things, with all fundamental interactions, space and time, and their contents, where hypernodes could be physical objects, states, events, or processes, and the hyperedges - the?fundamental interactions?or?fundamental forces, including four fundamental interactions known to exist: gravity; electromagnetism; weak interaction; strong interaction.
From Foundation AI/ML/DL Models to the General AI Foundation World Model
What is a Foundation World Model?
World models come from the mental models of the world, internal representations of external reality within one's mind that humans develop naturally. Such mental representations or cognitive simulations or worldviews, as cognitive maps,?mental maps,?scripts,?schemata, and?frame of reference, play decisive roles in cognition, reasoning , decision-making, and action. Mental models are based on a small set of fundamental assumptions, principles, axioms, fundamental truths, facts or beliefs, condensed and compressed by fundamental ontology and sciences,
The Foundation World Model, or the world simulator, is to organize and simulate the world of possible realities, physical, mental or digital, so that the whole aggregate reality works as the unit totality with each thing having a proper function, where all things put into their proper places in relation to each other.
It underpins all domain models, including scientific models to?understand,?define,?quantify,?visualize, or?simulate the facet of the world and machine world models, as generative AI models understanding the dynamics of the real world, including physics and spatial properties, using input data, including text, image, video and movement, to generate new content.
The GAI/FWM enables the AI foundation models, as language models, to really understand the things or concepts behind tokens, text, images, video, audio, code, to reason about the consequences of actions, and to rationally and effectively interact with the world.
The GAI/FWM has the ability to create virtual, interactive worlds generating 3D worlds on demand for planning, gaming, and more without data training and running world models requiring massive compute power, hundreds of thousands of GPUs to train and run, even compared to the generative models.
The GAI/FWM endows robots with an awareness of the world around and inside them.
The GAI/FWM is Generalized AI multiple hypergraph ontological networks underlying all possible neural network architectures and topologies, generating domain world models for downstream applications, special tasks or specific domains.
The GAI/FWM is conceptually or technically the universal computing ontology (UCO) of all reality,
The GAI/FWM is fundamental for all domain-specific foundation models, a foundation model or large X model, FM (LxM), large language models (LLMs), large concept models (LCMs), large world models (LWMs)..., all using established machine learning techniques like?deep neural networks,?transfer learning, and?self-supervised learning.?
What is a Foundation AI/ML/DL Model?
Foundation models are AI neural networks trained on massive unlabeled datasets to accomplish a broad range of tasks,
It is defined as "any model that is trained on broad data (generally using self-supervision at scale) that can be adapted (e.g., fine-tuned) to a wide range of downstream tasks"; [Introducing the Center for Research on Foundation Models (CRFM)]
Or, "an AI model that is trained on broad data; generally uses self-supervising; contains at least tens of billions of parameters; is applicable across a wide range of contexts".
FMs are a type of AI technology that are trained on vast amounts of data that can be adapted to a wide range of tasks and operations. [AI Foundation Models: Initial Report].
As an update of FMs, its LLMs, there are Large Concept Models, World Foundation Models, and Large World Models.
"LLMs have revolutionized the field of artificial intelligence and have emerged as the de-facto tool for many tasks. The current established technology of LLMs is to process input and generate output at the token level. This is in sharp contrast to humans who operate at multiple levels of abstraction, well beyond single words, to analyze information and to generate creative content. We present an attempt at an architecture which operates on an explicit higher-level semantic representation, which we name a “concept”. Concepts are language- and modality-agnostic and represent a higher level idea or action in a flow. Hence, we build a“Large Concept Model”".
World foundation models are neural networks that simulate real-world environments and predict accurate outcomes based on text, image, or video input.?Physical AI?systems like?robots?and?autonomous vehicles?(AVs) use world foundation models to accelerate training and testing.
There is Large World Model (LWM), a general-purpose large-context multimodal autoregressive model. It is trained on a large dataset of diverse long videos and books using RingAttention, and can perform language, image, and video understanding and generation.
World Labs is developing such-like “large world models” that will be used by professionals such as artists, designers, developers, and engineers to build AI models that understand and interact with the 3D world.
Examples of FMs are generative AI applications like Large Language Models,?OpenAI's GPT?series and?Google's?BERT, DALL-E?and Flamingo?for images, MusicGen?for music, and RT-for robotic control, as well as the sciences, astronomy,?radiology,?genomics,?music,?coding, times-series?forecasting,?mathematics,?and chemistry.
Again, learning from millions or billions or trillions of tokens of audio, code, video and language sequences poses challenges due to memory constraints, computational complexity, and limited datasets.
LLMs can be structured into open models such as Llama, Mistral, Bloom or Falcon, on the one hand, and closed models such as Gemini, GPT or Claude, on the other. All these models are based on the same underlying FM architecture: a transformer-based, decoder-only language model, which is pretrained to probabilistically predict the next token, given a long context of preceding tokens relying on statistical correlation patterns.
Meanwhile, "all statistical models are wrong" falling short of the complexities of reality.
To address these challenges, statistical AI/ML/DL/LL models are in need of the FWM capabilities.
The FWM as True, Real and Universal AI: the Omniverse Superintelligence Platform?
We advance, define, design and develop the FWM as Ontological AI (Omniverse AI), True, Real and Universal Intelligence (TRUAI), a pro-human, sustainable and rational framework for AI, ML, DL, LLMs, Agentic AI, Physical AI, AGI, or ASI.
The TRUAI's universe of discourse or intelligence (the domain of discourse, universal set or universe) is the whole world as the OMNIVERSE of realities, physical, mental, social, or virtual, with its fundamental things, entities, states, changes and interrelationships.
The method to build TRUAI as an onto-mathematical supercomputing information technology documented as "Engineering Superintelligence by 2025: AI for Everything and Everyone (AI4EE)" in:
WHITE PAPER | January 17, 2021 | EIS ENCYCLOPEDIC INTELLIGENCE SYSTEMS LTD
Besides, the innovation of True, Real, Universal or Ontological AI was published in 2021 in the Russian Ontological Design Journal (in English) titled as in:
The onto-mathematical information supercomputing systems has the necessary and sufficient means for general intelligence/superintelligence/hyperintelligence and rational interactions with the world. It embraces the physical symbol hypothesis: "A physical symbol system has the necessary and sufficient means for general intelligent action." and the OECD AI model:
"An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments".
The mathematical foundations of AI are essential for developing intelligent systems, but linear algebra, statistics, probability theory, optimization, calculus, or formal logic added with digital language data could hardly enable AI to really learn, predict, make decisions, etc.
The Omniverse Ontology Platform? (OOP?) of Hyperintelligent Information Systems
The TRUAI's universe of intelligence is the Omniverse Ontology Platform (OOP) encoded as the universal multi-hypergraph ontological networks of hypernode-things and hyperedge-interrelationships.
The OOP as universal computing ontology to?enable AI to model, understand,?define,?qualify, quantify,?or?simulate the world of realities as the omniverse of physical, biological, mental, social, digital or virtual realities, as multiverses and metaverses.
The OOP? is made up of several core components to structure the world (W), its all possible realities, R, and define all possible interrelationships or interactions (I) between all possible things (T) as entities (E), states (S), changes (C), allowing TRUAI systems to understand, organize, interpret, reason, decide, and interact with the world intelligently and effectively.
OOP? = W = <R; T; E, S, C, I> (1)
In the TRUAI's modelling the world, everything and anything is a thing in the world, whether it is entities, substances or objects, states, quantities or qualities, changes, events, or actions, relationships, associations, links or connections, you name it.
Thus the OOP? could simulate all and everything, including the physical universe, the totality of physical things, with all interactions, space and time, and their contents, where hypernodes could be physical objects, states, events, or processes, and the hyperedges - the?fundamental interactions?or?fundamental forces, including four fundamental interactions known to exist:
gravity; electromagnetism; weak interaction; strong interaction.
Or, it could be the world of biological things of living organisms, where hypernode-cells interact with themselves, other cells, and the environment, via cell signaling, a fundamental property of all?cellular life?in?prokaryotes?and?eukaryotes. Cell signals could be chemical in nature as ligands, including?ions?(e.g. Na+, K+, Ca++, etc.), lipids (e.g. steroid, prostaglandin), peptides (e.g. insulin, ACTH), carbohydrates, glycosylated proteins (proteoglycans), nucleic acids, etc., or physical quantities such as?pressure,?voltage,?temperature, or light. [Each cell is programmed to respond to specific extracellular signal molecules, and is the basis of?development,?tissue repair,?immunity, and?homeostasis. Errors in signaling interactions may cause diseases such as?cancer,?autoimmunity, and?diabetes].
The OOP embraces all complex structures and networks:
all major mathematical structures, with all possible mathematical structures and their combinations, measures,?algebraic structures?(groups,?fields, etc.),?topologies,?metric structures?(geometries),?orders,?graphs,?events,?equivalence relations,?differential structures, categories or the number systems, statistical models, probabilistic structures, etc..
ontologies, ?to?understand,?define,?quantify,?visualize, or?simulate the domains of knowledge
scientific models, to?understand,?define,?quantify,?visualize, or?simulate the empirical world, its objects, phenomena, processes, and laws of nature
knowledge graphs, taxonomies, schemas,
AI/ML/DL/genAI models, LLMs, etc.
It underlies all the complex networks,
physical networks,
biological networks,
brain networks,
social networks,
technological networks,
computer networks,
including artificial neural networks of different topologies.
Ontological AI? (Omniverse AI?) as the General World Model of Realities
Universal Formal Ontology (UFO) as the universal world model of reality, organized into integrative levels like physical, biological, mental, social or digital, for knowledge systems was originated in the books:
Artificial Superintelligence (USA, 1999)
UFO underpins different upper and domain silos ontologies created in the fields of computer science, artificial intelligence, machine learning, knowledge graphs, the Semantic Web, the systems engineering, the software engineering, etc., all to limit complexity and organize information and for problem solving.
The TRUAI's universe of intelligence is encoded as the universal multi-hypergraph ontological networks of hypernode-entities and hyperedge-interrelationships, covering all major mathematical structures, ontologies, knowledge graphs, scientific models, taxonomies, schemas, LLMs, and complex networks, physical, biological, brain, social, technological, computer, including artificial neural networks.
TRUAI as the Supreme Intelligence Platform
Despite seeming advancements in AI, the universal computing ontology (UCO) remains a foundation for building real, true and universal intelligent systems that require trust, transparency, responsibility, interoperability, and a structured way to represent data, information, knowledge, concepts, relationships, etc., in the universe of machine intelligence.
Following the OAI TRUAI paradigm, today's Artificial Intelligence, Machine Learning, Large Language Models, or GenAI platforms, DeepSeek-V3, Google’s Gemini Flash 2.0 model, Grok 3, are in need of general pre-trained machine ontology (fundamental world modeling framework) to be true and real AI.
TRUAI =
Universal World Models +
Symbolic AI +
Predictive AI +
Generative AI +
Large Language Models (xAI's Grok3, DeepSeek, GPT-4, OpenAI's ChatGPT, Google's Gemini, Claude) +
Agentic AI +
Causal AI +
Physical AI + ...