The world models and patterns rules as the essence of intelligence, natural and artificial
https://www.forbhttps://www.forbes.com/sites/cognitiveworld/2019/09/17/the-seven-pattees.com/sites/cognitiveworld/2019/09/17/the-seven-patterns-of-ai/

The world models and patterns rules as the essence of intelligence, natural and artificial

"Nobody can forecast the future of human world. It is a riddle wrapped in an enigma inside a mystery: but there is a key. That key is man-machine superintelligence. And whoever believes that humanity will do business as usual lost in his fiction reality made of fictional entities".

There is artificial intelligence and Artificial Intelligence, as Fake AI (FAI) vs. Real AI (RAI).

One is imitating, replicating, modeling or simulating human intelligence to effectively substitute it, which is Fake and False AI, promoted and capitalized by big tech, setting the global AI agenda.

The other one is imitating, replicating, modeling or simulating reality as the world to effectively complete human intelligence, which is REAL and TRUE AI.

We advance the Iron Law of Intelligence (ILI/AA) evidenced by all intellectual activities and meaningful human practice:

"World/Reality Modeling/Simulating defines Intelligence and Learning and Inference and Interaction, and vice versa"

No Reality Modeling, No Real Intelligence, and vice versa…

Real intelligence, human intelligence or machine intelligence or man-machine hyperintelligence, consists in its power to effectively interact with the world, as a dynamic causal world model, to be INTERACTIVE as an AGENT INTELLECT (intellectus agens; active intelligence, active reason, or productive intellect)...

The reason for positing a single Agent Intellect is that all (intelligent) entities possess or have access to a fixed and stable set of world's categories, a unified world knowledge of reality (the universe). The only way that all rational agents could possess the same correct knowledge is if they all had access to some central knowledge store, as terminals might have access to a mainframe computer, or task-specific ML models to the cloud AI platforms, or the Internet of Things to the Future Internet.

[The world models and patterns rules as the essence of intelligence, natural and artificial]

One of the consequences of the ILI/AA is that the complexity, scope and scale, accuracy and precision, of reality modeling determines the complexity, scope and scale, accuracy and precision, of intelligence, human intelligence or machine intelligence or man-machine hyperintelligence,

This rule refers to all AI, including gen AI, foundation models, LLMs and ChatGPT, to be transformed to Multimodal Large World or Reality Models, following the ILI/AA.

We extend the post, General AI and ML = Trans-AI = Unified World Model Machine + Intelligent Neural Networks, exposing what is a world, world model and worldview, with some examples from artificial intelligence, machine learning, large language models, and human intelligence (Supplement).

What is a world?

Our modern silos sciences are not putting much research and studies on the most critical concepts in the world, the World itself, as reality and all existence or the total environment, the sum total of all things and interactions, with all its key constructs, as world models, worldviews, world patterns, common sense knowledge, etc.

The idea of world has been conceptualized as the sensible world or the intelligible world of forms and ideas (Plato) or the temporal sensory world vs. the spiritual world or a plurality of possible worlds or "the world is everything that is the case" (Wittgenstein) or "a world of matter governed by the law of cause and effect" Sankhya), etc.

The idea of world is decisive for the theoretical intellectual activities, as philosophy and sciences, and technological innovations as artificial intelligence and machine learning.

One may study the types of world models in science and technology, as representational or perceptual, descriptive, spatial, mathematical or computer models, or to research if Large Language Models learn world models or just surface statistics.

Here is the Britannica dictionary, where the term "world" has all the sense but its major meanings.

It all depends on the scale and scope of study, its totality and generality or specificity and subjectivity, namely:

the world as a whole, "the totality of entities, the whole of reality, or everything that is, has been, and will be"

the physical universe, cosmos, "[t]he totality of all space and time, all forms of matter and energy, including planets, stars and galaxies"

the earth, together with all of its countries and peoples

the inhabitants of the earth: the human race, human society, world history, world politics

a particular region or group of countries

the real world, the world where everyone lives, works, and deals with everyday problems

a particular kind of interest, activity, or social situation, or the people who are involved in it

  • the art/music/fashion world
  • the business and financial worlds
  • the world of the rich and famous

a group of things of a particular type

  • the animal/plant/insect world

a particular environment

  • the natural world
  • exploring the underwater world
  • Technology is forever changing our world

a particular part of human life and experience

  • the physical/material/spiritual world

the life and experiences of a particular person

  • His (whole/entire) world fell apart when his wife left him.

Worlds are usually defined as all-inclusive totalities, as the universe or the world of music, the world of business, the world of sports, the world of experience or the Western world.

What is a World Model?

Now, "models are schematic representations of reality or of one's view of a possible world, constructed to improve one's understanding about the world and/or to make predictions".

Models imply data and facts confirming or falsifying theories and models, and theories which are often expressed by models, thus being the patterns of data.

Then a world model is modeling or simulating, mapping or replicating the world at large and in detail, its nature and constituents, structures and functions, states and behavior, at all its levels and scales.

Generally, a world model is an abstract representation of the whole reality. Specifically, it is an abstract representation of spatial or temporal dimensions of our world.

The world model is a key to any intelligent systems, their learning and knowledge, inference and behavior and interactions with any environment.

A typology of world modeling consist in the following hierarchy:

Metaphysics/Ontological world models

Mathematical world models

Scientific world models

Physical world models

Informational world models

Mental world models

Perceptual world models. We humans use a world model as a simulator in our brain. The model is obtained by learning from large amounts of sensorimotor data through interaction in the environment.

Computational world models, World Models in Machine Learning, Deep learning, reinforcement learning, and world models

ML models, supervised learning, unsupervised learning, reinforcement learning

Language world models, as the World Models in LLMs, Neural World Models, LANGUAGE MODELS REPRESENT SPACE AND TIME

World model architecture for AGI

The idea that humans, animals, and intelligent systems use world models goes back many decades in psychology and in fields of engineering such as control and robotics.

Meta AI’s Chief AI Scientist Yann LeCun proposes that one of the most important challenges in AI today is devising learning paradigms and architectures that would allow machines to learn world models in a self-supervised fashion and then use those models to predict, reason, and plan.

LeCun, who is after a human-like general-purpose AI, thinks that animal brains run a kind of simulation of the world, which he calls a world model. He proposes that the ability to learn “world models” — internal models of how the world works — may be the key.

A system architecture for autonomous intelligence. The configurator gets inputs from other modules, but we have omitted those arrows in order to simplify the diagram

The world model module constitutes the most complex piece of the architecture. Its role is twofold: (1) to estimate missing information about the state of the world not provided by perception, and (2) to predict plausible future states of the world.

The world model may predict natural evolutions of the world or predict future world states resulting from a sequence of actions proposed by the actor module. The world model is a kind of simulator of the part of the world relevant to the task at hand. Since the world is full of uncertainty, the model must be able to represent multiple possible predictions.

In all cases, the agent makes a prediction about how it would like the world to be, and then act in order to make that prediction come true.

Agents continuously endeavor to predict and understand their surroundings, generating world models, to minimize discrepancies between goals or beliefs and environmental [sensory] evidence.

What is a Worldview?

The world model implies a worldview as a comprehensive representation of the world and our place in it. It is a subjective perspective of the world and thereby different from the world it represents.

A worldview can be unique to one individual but worldviews are usually shared by many people within a certain culture or religion.

It implies common sense knowledge and reasoning, as "a collection of models of the world that can guide on what is likely, what is plausible, and what is impossible".

Patterns in the World

Any intelligent agents seek out patterns to make sense of of the world of information. Otherwise, failing to find a pattern that makes sense, chaos and confusion, uncertainty and randomness abound.

A pattern is a regularity in the world, and the elements of a pattern repeat in a predictable manner.

The world is a world of all possible patterns, created by its entities and their infinite interactions.

All science and mathematics can be seen as the search for regularities, where theories and models explain and predict regularities in the world, as the patterns of world's data.

Considering that in AI the world knowledge is major patterns, a typology of world's patterns consists in its world knowledge hierarchy:

metaphysical patterns, substances and objects, states and quantities and qualities, changes, events and actions, interactions, processes and relationships

natural patterns

physical patterns

chemical patterns

biological patterns

mental patterns

behavior patterns

social patterns

economic patterns

political patterns

environmental patterns

engineering patterns, as software design patterns, form UI design patterns to pattern language of programming to interaction design patterns...

The world's specific patterns are abstracted by scientific knowledge as a variety of data patterns, from data structures and data types to ordered information and knowledge.

Scientific knowledge refers to a generalized body of laws and theories to explain a phenomenon or behavior of interest that are acquired using the scientific method. Laws are observed patterns of phenomena or behaviors, while theories are systematic explanations of the underlying phenomenon or behavior.

The world's patterns are generalized by all sciences, including mathematical sciences, statistics, probability theory and data sciences.

Some generic/abstract patterns that are used to structure and order world's entities and information could be:

  • Metaphysical patterns and Ontological patterns, Classification patterns, from universal categorization to domain classifications, distributing into a class or category according to characteristics; grouping and identifying things, objects, ideas or data, into predetermined categories, classes, kinds, types or groups according to features, similarities or dissimilarities.
  • Typological patterns, from topical patterns to biological taxonomies to as Biologists classify living things at 10 levels: life, domain, kingdom, phylum, class, order, family, genus, species, and individuals, by characteristics, such as relationships, appearance, reproduction, movement, etc.
  • Cause-Effect Patterns.
  • Interaction Patterns.
  • Temporal/Chronological Patterns.
  • Spatial Patterns.
  • Sequential Patterns.

Generalized AI as autonomous pattern-recognizing world modeling machines

Autonomous pattern-discovering WMMs are endowed with machine intelligence and learning capacities to know the external and internal world, predict the future behavior of its elements, and plan for how to deal with those changes.?

Such autonomous AI machines imply modeling and simulating, identifying and classifying, computing and interpreting the world's data patterns, in all their varieties and complexities, as entity embeddings patterns, causal data patterns, number patterns, image patterns, logic patterns, NL patterns, and so on.

So, real machine intelligence and learning consists in its knowing, encoding, and programming, inferencing or processing the world's data in terms of world models and patterns rules.

Being inspirited by the world's models and data patterns, embracing autonomous vehicles, predictive analytics applications, facial recognition, chatbots, virtual assistants, cognitive automation, fraud detection, and large language models, real AI is interacting with the world via the patterns, combining a few common patterns:

individual hyperpersonalization,?in personal life and in industries such as finance, healthcare, or personalized fitness and wellness applications, like creating personalized recommendations based off of browsing patterns and searches,

autonomous?systems patterns, physical and virtual software and hardware systems that are able to accomplish a task, reach a goal, interact with their surroundings, autonomous machines and vehicles of all sorts includes cars, boats, trains, airplanes, and more, autonomous documentation and knowledge generation, autonomous business processes, and cognitive autonomation,

predictive analytics and decision support, some examples of this pattern include assisted search and retrieval, predicting some future value for data, predicting behavior, predicting failure, assisted problem resolution, identifying and selecting best fit, identifying matches in data, optimization activities, giving advice, and intelligent navigation,

conversational machine-human interactions,?chatbots, voice assistants, content generation, sentiment / mood / intent analysis, and machine translation,

patterns and anomalies,?using ML and other cognitive approaches to learn patterns in the data and learn higher order connections between data points to see if it fits an existing pattern or if it is an outlier or anomaly,

recognition systems, image, video, audio, and object recognition, classification, and identification,

large language models patterns,

goal-driven systems patterns.?

Conclusion

In making predictions about the world, any intelligent agents, animals or humans, machines or aliens, form beliefs or guesses, hypotheses, theories or patterns, about the world, constantly doing some types of world modeling, with causal cycles of trials, errors and corrections, thus making sense of the world while interacting with the environment.

World models are schematic representations of reality or of one's view of a possible world, constructed to improve one's understanding about the world and/or to make predictions".

Models imply data and facts confirming or falsifying theories and models, and theories which are often expressed by models, thus being the patterns of data.

A world model is modeling or simulating, mapping or replicating the world at large and in detail, its nature and constituents, structures and functions, states and behavior, at all its levels and scales.

Generally, a world model is an abstract representation of the whole reality, a totality of world's patterns. Specifically, it is an abstract representation of spatial or temporal patterns of our world.

The world model is a key to any powerful intelligent systems, their learning and knowledge, inference and behavior and interactions with any complex environment.

Resources

Interactive AI (IAI) or Agent Intellect: ML > DL > ANI > GenAI > AGI > ASI > Active Intelligence = Trans-AI = Meta-AI

General AI and ML = Trans-AI = Unified World Model Machine + Intelligent Neural Networks

TRANS-AI: HOW TO BUILD TRUE AI OR REAL MACHINE INTELLIGENCE AND LEARNING

Real AI vs. Hype AI: Generative AI Implications

Yann LeCun on a vision to make AI systems learn and reason like animals and humans

A Path Towards Autonomous Machine Intelligence Version 0.9.2, 2022-06-27

SUPPLEMENT: World Model: Mental Models: Artificial Causation: Internal Simulation

Mental world models can occur in various forms; e.g., (Craik, 1943,?Evans, 2006,?Furlough and Gillan, 2018,?Gentner and Stevens, 1983,?Halford, 1993,?Johnson-Laird, 1983,?Treur and van Ments, 2022).?

It is worthwhile to mention the hypothesis of artificial causation by Craik:

If the organism carries a ‘small-scale model’ of external reality and of its own possible actions within its head, it is able to try out various alternatives, conclude which is the best?of them, react to future situations before they arise, utilize the knowledge of past events in?dealing with the present and future, and in every way to react in a much fuller, safer, and?more competent manner to the emergencies which face it. (p. 61)

Such internal models work in a way similar to how the real world works.

K.J.W.?Craik,?The nature of explanation, University Press,?Cambridge, MA?(1943)

"all mental models use the same causal [brain-mind] mechanisms with the same brain structures and processes, neural circuits and pathways and networks, in different brain?modules, regions and areas".

The causal pathways or internal association mechanisms?are the ‘simulation of behaviour and perception’ or?simulated perception-behaviour cycles/chains.

One universal mechanism in the brain that handles all mental models.

Mental models involving emotions and feeling states associated to some considered action or belief, will use parts and pathways of the brain that are not the same as mental models that do not involve such emotions and feeling states, It is, as?simulating the mental processes of other persons or?perceiving the own or someone else’s body states?or?perceiving states of the physical world.

They are a kind of structures or processes in the mind that reflect structures or processes in the world or in other persons.

The idea of internal simulation is that in a certain context (which may cover sensed aspects of the external world, but also internal aspects such as the own goals), preparation states for actions or bodily changes are activated, which, by prediction links, in turn activate certain sensory representation states. The latter states represent the (predicted) effects of the prepared actions or bodily changes, and can be activated from the preparation states by internal connections without actually having executed these actions or bodily changes in the external world or in the body. The notion of internal simulation has been put forward, among others, for:

As another example, by religious humans a mental God-model is simulated for influencing their?behaviour?as also addressed in (Van Ments et al., 2018,?Van Ments et al., 2022). This mental God-model refers to the personal God of the individual.

Fig. 5.?A mental model for context?C?realised in the brain and its correspondence to the world.

Here for a faithful realisation,?all?relations in the mental model network have to correspond to similar relations in the brain and for a faithful representation of the world?all?relations in the mental model network have to correspond to similar relations in the world. As a result, the corresponding network in the brain will faithfully simulate the world processes.

Mental models in the brain: On context-dependent neural correlates of mental models


Dhruv Tyagi

CERN - Summer Student | GSoC'23 - RedHenLab | X : DevRel - MinusZero, Siemens Intern & Scholar

9 个月

Great Article!

回复

要查看或添加评论,请登录

Azamat Abdoullaev的更多文章

社区洞察

其他会员也浏览了