Intellectualizing AI & ML & DL & LLMs: Reality Models + AI Models
[Real is no more real, false is no more false, all is blended by generative AI and LLMs, right or wrong, truth and falsity, as false positive or false negative].
All today's AI is the case of pathetic fallacy, "attribution of human feelings and responses to inanimate things or animals, especially in art and literature", and now in digital technology.
It is all the pathetic fallacy or personification, "...giving human feelings or qualities to objects, nature, or animals, for example, by referring to the "artificial intelligence", "machine learning", "deep learning", "artificial neural networks", "training data", etc.
One can not train SOMETHING to to do something. One can not "teach" some statistical algorithms or mathematical models. The whole idea of "training data", or "training set", "training dataset" or "learning set" is simply a commercial fraud, like as a neural network, "a computational model that mimics the way nerve cells work in the human brain".
The same refers to "machine learning" as "the sub-category of artificial intelligence (AI) that builds algorithmic models to identify patterns and relationships in data", with all the consequences.
One can train only a reality/world model like as a human mind with its mental world model
We argue that real machine intelligence and learning (MIL) is an AI world model with learning, inferential and interactive powers.
No AI models can understand and simulate the real world, without having as encoded, programmed and trained its reality models.
To intellectualize AI, we have to bridge the gap between its data models and the world with the categories of knowledge and reality, as the onto-scientific classes and general rules of real-world entities and interactions.
Reality, Model, Data, Intelligence
By its definition, a model is a simulation or approximation or theory of reality or the world, the pieces of reality, including their properties and interactions, all represented by data models at various levels of abstractions and levels or scales of measurement:
Nominal
Ordinal
Interval
Ratio
Numeric or Cardinal
There are ontological, conceptual, mathematical, scientific, physical, mental, language, social, digital, virtual or technological models or abstractions of reality.
For example, a mathematical model describes reality, its systems, functions and processes, numerically and quantitatively, by a set of real-world variables and a set of equations that establish relationships between the variables, as exponential decay, exponential growth, quadratic functions, and linear functions.
"Mathematical models of reality are the vastly more important type of representation", widely applied for the physical or biological world.
Now, what are AI models? AI models or artificial intelligence models are programs that detect specific patterns using a collection of data sets, receiving data inputs and draw conclusions or conduct actions depending on those conclusions. Once trained, an AI model can be used to make future predictions, produce decisions or act on data that was not previously observed. AI models can be used for a variety of activities, from image and video recognition?to natural language processing (NLP), anomaly detection, recommender systems, predictive modeling and forecasting, and robotics and control systems...or generating human-like language content.
It is realty modeling, what all AI models are missing to become truly intelligent:
True AI = Reality Models + Training data/LLM- based AI/ML/DL models [employing learning algorithms like decision trees, random forests, gradient boosting, and linear and logistic regression or deep neural networks to learn from unstructured data, with TensorFlow, PyTorch, and Caffe-like tools)
In all, reality models provide real-world context to data, information, knowledge and intelligent behavior, being a basic natural tool to understand and describe the world in all its complexity to make meaningful discoveries, explanations and predictions.
The intelligence/mind naturally constructs "small-scale models" of reality, as a schema, picture or image of the world around it.
AI's Reality Models are like mental models or mental representations or mental simulation generally which are internal representations of external and internal reality, playing a major role in intelligence, cognition, learning, reasoning, and decision-making, anticipating events or predicting outcomes, shaping behavior, solving problems and doing various tasks.
Incorporating the reality modelling engine, the concept of REAL AI is transcending or generalizing and intellectualizing its numerous human-mimicking applications, including natural language processing, computer vision, robotics, decision-making systems, or LLMS/ChatGPT, as a range of statistical techniques, tools and methods that replicate human intelligence in machines.
Again, without world modeling, you can not teach machines to learn from data and make predictions or judgments based on that data.
AI model can imitate realistic and imaginative scenes from text instructions, but without any intelligence or understanding.
For example, video generation models as world simulators , like Sora , to understand the real world, a milestone capability for achieving Real AI, must have reality model learning and inference and interaction engine.
Or, Real and True AI is a world model with learning, inferential and interactive powers.
Scaling statistical AI/ML models across GPU, compute, people, and data requires a combination of technology, infrastructure, and expertise.
Scaling real AI/ML models across GPU, compute, people, and data requires more comprehensive world models, more sciences and world knowledge.
Presently, no AI/ML/DL models, predictive or generative AI, self-driving cars, humanoid robots or large language models, employing the world models about reality or pieces of reality, with their entities, properties and interactions.
Reality is Everything and Everything is Reality
Why could the AI foundation models lose its value? Why are OpenAI’s GPT4 or Alibaba's Qwen-VL or Google's Gemini losing their status as the most powerful Large Language Models?
Or, Will AI Models Matter Anymore?
AI is a simulation or approximation of human body/brain/brains/behavior/business or human intelligence, cognition or the mind.
Human intelligence is about knowing and interacting with the world by means of its mental capacities, perception, thinking, understanding, communication, reasoning, learning and memory formation, intention, action planning, problem solving, decision-making and agency.
ALL REAL AI MODELS MUST HAVE THE MOST ESSENTIAL ELEMENT OF ANY POWERFUL INTELLIGENCE, a world model, A MODEL OF REALITY, A SIMULATION OF THE WORLD.
领英推荐
The only commonality with all the narrow AI/ML/DL/LL models, they are lacking real intelligence and learning due to missing reality modeling encoding the information of how the world is structured, operates and changes at all possible levels.
AI World Models can hardly be replaced with statistical language models or ML training data examples, supervised or unsupervised, however large the training datasets could be. If one aims to build a general-purpose humanoid robot or universal physics, chemistry or biology AI model, from Optimus to Bioptimus , first create a machine ontology of your piece of reality.
Reality Modeling and Simulation Engine automatically categorizes, analyzes and classifies the things in the world, tagging, labeling, or annotating the data to the criteria of the world model, finding objective patterns (causal links, similarities and differences) in the data to give explanations, discover regularities, make all possible inferences and conclusions and perform effective interactions.
The Reality Engine (RE) acts as a real AI technology platform that knows or learns, understands and draws inferences, predicts and interacts based on comprehensive scientific world modeling, not merely statistics and correlations, of data input.
Such a Machine World Model [Learning, Inference, Interaction] Engine is superior to a traditional machine learning AI platform, generative AI and causality engine.
What is reality? Why nobody understands the world's true nature
It’s the ultimate quest – to understand everything that there is, was and will be. But the more science we get, the further away it seems. Can we ever get to grasp with the true nature of reality?
Both humans and AI machines have a lot of problem with reality. Humans experience it all the time, while struggling to define it, let alone understand it.
We don’t know when reality or universe of the world began, how big it is, where it came from, how it works and where it is going, and we certainly have no idea why it all exists, including its special realities, like human beings.
Still, the inherent desire to understand reality in all its complexity seems part of human nature, and we have come a long way, from mythologies and divine creation to science, engineering and technology.
Regardless, the mystery of reality has only deepened. We are now at a trifurcation point where it is equally credible to claim that reality is entirely dependent on subjective experience, or entirely independent of it, or it is all a simulation mental or virtual. Reality has never been so unreal, and its definitions as the meaning of everything have never bein so confused.
WHAT IS NOT REALITY
Reality is the sum or aggregate of all that is real or existent within the universe, as opposed to that which is only imaginary, nonexistent or nonactual. The term is also used to refer to the ontological status of things, indicating their existence. In physical terms, reality is the totality of a system, known and unknown.
WHAT IS REALITY
Reality is the sum or aggregate of all that is real or existent, imaginary, nonexistent or nonactual, all what is real or causal or actual or interactive.
Reality is the totality of all possible worlds, or everything, known and unknown.
Reality is the totality of all things, structures (actual and conceptual, digital or virtual), changes/events (past and present and future) and phenomena, whether observable or latent, noumenal or phenomenal.
It is what a world model/view/theory ultimately attempts to describe or map or explain.
Reality is the absolute universal, of which all particular things have in common, or everything in the world has in common the universal quality/characteristics/property of being real.
Again, everything in the world is real, including multiverses, virtual and physical, as parallel universes, "alternative universes", "quantum universes", "interpenetrating dimensions", "parallel dimensions", "parallel worlds", "alternative realities", "alternative timelines", and "dimensional planes", among others.
Reality can be defined in a way that links it to worldviews (conceptual frameworks) or data (data models), while having various manifestations:
Reality is causally related with data and intelligence; for there is no intelligent entities, agents or systems or technologies, without inherent/programmed/embedded/trained models of reality or its pieces and domains.
Or, Reality Representation and Learning, Inference and Interaction Engine is the General Mechanism of General Intelligence, existing as intelligent beings or AI machines [General AI and ML = Trans-AI = Unified World Model Machine + Intelligent Neural Networks ]
All Reality Model + Narrow AI/ML/DL/LL Models
AI as ML, DL or ANNs, LLMs, AGI or ASI is all about mathematical techniques and statistical algorithms, probability theory and computational manipulation.
It is a mystification or bad science to anthropomorphize it all as that which “enables computers and machines to mimic the perception, learning, problem-solving, and decision-making capabilities of the human mind”.
Or, as human-like machines which would feature highly advanced reasoning, decision-making, and problem-solving capabilities far beyond the creative or logical capabilities of any human being.
Making sense of the world, inferring cause-and-effect relationships between variables, understanding and analyzing the causal relationships between events, or learning to reason about causality, causation or interaction, is a hallmark of intelligence systems, natural or artificial.
If your AI/ML/DL/LLM is unable to reason about the world or the environment around you, it is absolutely unintelligent.
If your software applications, as LLMs or GPT-x or ChatGPT, are unable to solve causal reasoning problems , they are absolutely unintelligent.
LLMs are trained on the internet datasets to extrapolate or interpolate, "predict" sequences of words generating human-like text/code/image/audio on demand. LLMs are nonintelligent due to the very design, wanting an encoded world learning, inference and interaction model, or the power of making sense of the world of complex cause-effect relationships.
Real Intelligent AI is about reality and truths, models and facts, of reality and by reality, with the real intelligence system architecture:
Real AI Technology (RAIT) = the world or reality (the real world, the environments, the internet, virtual reality, etc.) + reality modeling and simulation and interaction engine [perception (sensors, the internet of things, robotics) + knowing/conception/classification + inference/reasoning/decision making (GOFAI) + learning (AI & ML & DL & ANNs) + actuation (actuators, robotics)] + the environment (the real world, the internet, virtual reality, etc.)
Resources
Project Commonssense, ULB Holistic Capital Management, ULB Institute
9 个月I believe the Community(accessinter)Link model is the missing link .... the open data ecology (CODE) model for reality ..real time ..utility ... combined with the universal (ULB) multi level integrated knowledge model. Personal and community data ownership is the fundamental prerequisite to step into this new world ...your reality, universal ontology and knowledge systems. Where to start? ULBH RTO?