Real AI Technology (RAIT): general world models and/or AI/ML/LL models
"There is real AI (real and true, scientific and objective) and unreal AI (false and fake, imitating and subjective) AI. To build true intelligent machines, teach them how to model. simulate and effectively interact with the world". [A New Man-Machine Real AI World]
We advance Real AI Technology (RAIT) driven by the world modeling and reality simulation engine, integrating emerging and digital technologies,
AI, Expert Systems,
ML, Generative AI,
Large Language Foundation Models,
Interactive Intelligent Agents, Chatbots,
Semantic Networks/Knowledge Graphs, etc.
A real world intelligence implies an effective interaction and exchange of information with the environment by any optimal means and ways, neural or cellular, electronic, mechanical, electric, etc.
Real AI machines model and simulate or understand the world with the encoded world knowledge, embedded semantics and programmed world ontology, representing reality, its categories of things as foundational data categories and interactions between entities as foundation data interrelationships.
The Reality-Universal Ontology-Knowledge Systems book demonstrated an intellectual need for an integrated world model as a formal comprehensive and consistent, complete and coherent world model providing cross-machines semantic interoperability and unifying communication between humans and machines.
Discovering meanings in concepts and actions, signals, signs and symbols, data and information or communication, the RAI technology succeeds logical AI, Expert Systems, ML, Generative AI, Large Language Models, Interactive Intelligent Agents, Chatbots, Semantic Networks/Knowledge Graphs, and the Semantic Web.
The generative AI systems and its biggest large language models (LLMs) need learn the world models to understand the data they’re processing to reach the level of real-world AI technology learning ontological, causal, semantic, and statistical structures and regularities in the world and its representations.
Understanding Faking AI
Common AI refers to the simulation or approximation or faking of human intelligence in machines by software-coded?heuristics or statistical learning algorithms yo recognize patterns and make predictions. It is a method of making a computer, a computer-controlled robot, or a software or agent think and act intelligently like humans, by studying the patterns of the human brain and by analyzing the cognitive functions.
Such a human-faking AI seems more powerful than ever, with generative AI systems and LLMs and chatbots like Gemini, Bard and ChatGPT capable of producing human-like content, code, audio, video, images, or text.
The Gen AI technology promises future disruptions in economy and politics, industry and manufacturing, education and health, communication and social media, entertainment and creativity, etc.
LMMs are predicted to contribute from $2.6 trillion to $4.4 trillion annually into the global economy.
But for all their effects, these applications still leave many wondering: Do such models actually understand what they are doing, saying, painting, etc.?
The spectrum of belief is ranging from the blind trust to believing they are just stochastic parrots. As it was suggested in the a 2021 paper: LLMs with modern chatbots generate content/text only by combining information they have already seen “without any reference to meaning,” which makes an LLM “a stochastic parrot.”
These models power many of today’s biggest and best chatbots, and as Hinton argued that it’s time to determine the extent of what they understand, stressing two critical things:
(i) It's important that AI scientists reach consensus on risks-similar to climate scientists to shape good policy.
(ii) Do AI models understand the world?
As it is noted by Yann LeCun, "We have mental models of the world in our minds that allows us to simulate what will happen. That's what gives us common sense. LLMs don't have that".
How AI Models Could Understand the World
The necessary condition for machine intelligence, learning or understanding is a general world model (GWO) constructed by universal machine ontology, science, mathematics and computer science.
A world model is a Real-World AI system that builds an encoded representation of the world and its various environments to model and simulate entities, changes and processes within a wide range of environments. Research in world models has so far been focused on very limited and controlled settings, either in generative neural network models of popular reinforcement learning environments, toy simulated worlds (video games) or narrow contexts (such as developing GAIA - world models for driving) or mental world models for self-supervising learning, JEPA.
The aim of general world models will be to represent and simulate a wide range of situations and interactions, like those encountered in the real world or virtual world.
The GWM generates consistent maps of the environments, with the ability to adjust and adapt, navigate and interact in those environments, tracking the dynamics of the world, the dynamics of its agents, including realistic models of human behavior.
Machine' s world models are essentially different that mental models or small-scale models of reality, mental representations or mental simulation generally, Such biologically limited human's world models are guiding our cognition, values and inferences, decisions and actions:
“The image of the world around us, which we carry in our head, is just a model. Nobody in his head imagines all the world, government or country. He has only selected concepts, and relationships between them, and uses those to represent the real system.” (Forrester, 1971)
It is thought that mental models are so basic to understanding the world that people are unconscious of them.
A computer program capable of acting intelligently in the world must have a general representation of the world in terms of which its inputs are interpreted. Designing such a program requires commitments about what knowledge is and how it is obtained. …More specifically, we want a computer program that decides what to do by inferring in a formal language that a certain strategy will achieve its assigned goal. This requires formalizing concepts of causality, ability, and knowledge. [McCarthy, J. and Hayes, P.?J. [1969]. Some philosophical problems from the standpoint of artificial intelligence]
The Integrated World Model is embracing integrated data/information/knowledge models as its applications.
Real machine intelligence is marked with its power to qualify and quantify, to conceptualize/generalize/abstract things. measuring or computing their properties and features.
Machine Intellectual Needs
Humans have several basic needs to function properly: physical or psychical needs, like Maslow's hierarchy of needs, from physiological to self-transcendence spiritual needs, where cognitive needs imply meaning, information, comprehension and curiosity – a will to learn and attain knowledge,
And "transcendence refers to the very highest and most inclusive or holistic levels of human consciousness, behaving and relating, as ends rather than means, to oneself, to significant others, to human beings in general, to other species, to nature, and to the cosmos."
Such a desire to reach the infinite is modelled with the Integrated World Model, a universal hierarchical categorical classification integrating all scientific, industrial and statistical classifications, taxonomies or typologies, physical, chemical, biological, social, etc..
For RAI machines, the human needs could be translated as Energy/Physiological Needs; Intellectual/Cognitive Needs; Information/Knowledge Needs; Technological Needs.
The intellectual needs of real intelligent systems are to include 5C needs:
(1) need for conception, generalization, conceptualization or structure, the need to organize the data collected or knowledge learned into a systematic logical structure, the need to analogize structures, problems, and solutions to problems by transfer learning
(2) need for certainty, facts and truth, statements and proof schemes, inductive or deductive or abductive or analogical
(3) need for causality, why something true, to explain, to determine a cause of a phenomenon, [we suppose ourselves to possess unqualified scientific knowledge of a thing, as opposed to knowing it in the accidental way in which the sophist knows, when we think that we know the cause on which the fact depends as the cause of the fact and of no other. (Aristotle, Posterior Analytic, p. 4)
(4) need for computation, transforming categorical or ordinal data into interval, ratio or numerical data, or quantifying, as the act of transforming a sensation (i.e., a perceptual action scheme—visual, auditory, tactile, etc.) into a quantity—a measurable sensation. For example, the sensation fastness is transformed into speed; heaviness into weight; extent into length, area, or volume; pushing or pulling into force; rotational twist into torque; hotness into heat (i.e., thermal energy), etc.
(5) need for communication, as the act of conveying and exchanging ideas by interactions (by means of a spoken language and gestures), which are defining features of intelligent agents. It involves formulating and formalizing. Formulating is the act of transforming strings of spoken language into encoded (algebraic or digital) expressions amenable to computation. Formalization is the act of externalizing the exact intended meaning of an idea or a concept or the logical basis underlying an argument. In the formal ontology and modern mathematics, the acts of formulation and formalizations are reflexive, formalizing an idea necessitates its formulation, and, conversely, as the agent formulates something it need to formalize it.
领英推荐
By providing a formal representation of reality, its entities and their interactions and relationships, the IWM creates a world model engine that allows machines to learn features, interpret data, process information, inference conclusions and interact effectively.
RAI Models of Reality to Understand the World
As to Lecun, "there are three main challenges that AI research must address today:
1. How can machines learn to represent the world, learn to predict, and learn to act largely by observation? Interactions in the real world are expensive and dangerous, intelligent agents should learn as much as they can about the world without interaction (by observation) so as to minimize the number of expensive and dangerous trials necessary to learn a particular task.
2. How can machine reason and plan in ways that are compatible with gradient-based learning? Our best approaches to learning rely on estimating and using the gradient of a loss, which can only be performed with differentiable architectures and is difficult to reconcile with logic-based symbolic reasoning.
3. How can machines learn to represent percepts and action plans in a hierarchical manner, at multiple levels of abstraction, and multiple time scales? Humans and many animals are able to conceive multilevel abstractions with which long-term predictions and long-term planning can be performed by decomposing complex actions into sequences of lower-level ones".
An architecture for autonomous real-world intelligent agents with possible solutions to all three challenges and beyond is formulated and formalized as
Real AI: The World [Learning and Inference and Interaction] Model
The World = Reality > the World's Data/Information/Knowledge/Meaning >
Real-World AI Technology [AI, Expert Systems, Generative AI, Large Language Models, Interactive Intelligent Agents, Chatbots, Semantic Networks/Knowledge Graphs/Ontologies, Semantic Web] >
Real Semantics [Meaning = Definition + Understanding] =[Reference + Representation = Extension + Intension = Denotation +Connotation = Signification + Sense] >
World Learning [Understanding = Knowledge =Science = Facts, Data, Observations + Postulates, Laws, Rules, Principles, Theories, Models] : :
Ontology [ Reality = Being = Existence = Universe = World = Entity/Thing/Variables + Interaction/Relationship/Networking] >
Physics [Substance/Matter/Material + Interaction/Process/Force/Energy] >
Mathematics [Quantities + Relationships] + Statistics [Data/Variables + Inferences/Classification Algorithms] > Probability [Events + Probabilities/Rules/Laws] >Computation [Data/Information + Computing] >
Brain [Neurons + Connections] + Mind [Perception/Conception/Idea/Thought + Cognition/Thinking/Reasoning/Inference] >
Logics [Symbols + Calculus/Inference Rules] >
AI/ML [Training Data + Algorithms. Models] >
ANNs [Nodes/Units/Neurons + Topologies, Structures, Networking] >
NLP [ Language Data + Grammar, Syntax, Semantics, Pragmatics] >
the Internet [Computers + Communication Protocols] > the World Wide Web [Web Data + Hypertext, HTML, URI and HTTP] >
Global Communication & Information Networks [World's Information + the IoT] >
Technology [Technologies/Machines/Devices + Integration/Interoperability] >
Human-Machine Interaction [Man, Machines + Interfaces] >
Real-World Man-Machine Intelligence and Learning
Conclusion
True and real intelligence, human or machine, natural or artificial, is dealing with reality in terms of the general world models and data/information/knowledge representations for effective and sustainable and intelligent interactions with a wide range of complex environments.
RECOMMENDATIONS
STOP SELF-DELUDING BELIEVING THAT AI COULD MIMICK HUMAN INTELLIGENCE, OUR NATURAL PERCEPTION, THINKING AND ACTION.
MACHINE INTELLIGENCE IS ABOUT BUILDING THE WORLD MODELING AI MACHINES TRANSCENDING BUT COMPLETING THE HUMAN MIND.
WE DON'T NEED THE HUMAN-COMPETE LARGE LANGUAGE AI/ML MODELS, BUT RATHER THE HUMAN-COMPLETE LARGE WORLD MODELING AUTONOMOUS INTELLIGENT SYSTEMS.
Resources
Abstract
How could machines learn as efficiently as humans and animals? How could machines learn to reason and plan? How could machines learn representations of percepts and action plans at multiple levels of abstraction, enabling them to reason, predict, and plan at multiple time horizons? This position paper proposes an architecture and training paradigms with which to construct autonomous intelligent agents. It combines concepts such as configurable predictive world model, behavior driven through intrinsic motivation, and hierarchical joint embedding architectures trained with self-supervised learning.
SUPPLEMENT
Simple cells, not just highly specialized neurons, can exhibit basic cognitive abilities such as memory, learning, and problem-solving. Researchers suspect that?body cells are able to use weak electric fields to store information.?
Plants, slime molds, and single-celled organisms also demonstrate surprising abilities to sense and respond to their environment, challenging the idea that intelligence is limited to creatures with brains. Weak fields of bioelectricity could be how cells communicate with each other and transmit information throughout the body.