AGI Bible: Intelligent Causal Machines: Overwriting AI/ML/DL/LLM/AGI
There is only one constant, one universal. It is the only real truth. Causality. Action, reaction. Cause and effect. Merovingian, The Matrix Reloaded.
There is only one constant, one universal. It is the only real truth. Causality. Action, reaction. Cause and effect. Merovingian, The Matrix Reloaded.
The basic principles, rules, laws as the groundwork of truly intelligent machines
REALITY IS ALL ABOUT CAUSE AND EFFECT, ACTIONS, REACTIONS, and INTERACTIONS...
REAL IS CAUSAL, CAUSAL IS REAL... CAUSAL IS RATIONAL, RATIONAL IS CAUSAL...RATIONAL IS INTELLIGENT, INTELLIGENT IS RATIONAL
Reality, the Universe, Evolution, History, Science or Technology, Artificial Intelligence and Machine Learning, all about cause and effect...
To build truly intelligent machines, encode causal models as world vector embeddings.
We proceed with the AGI Bible series highlighting the three key points:
1. Causality is the General Science of Causal Interrelationships, Interactions, Associations, or Correlations
2. The Essence of Intelligence, Natural or Artificial, is Causal Intelligence or Interactive Intellect
3. True or Real AI, Machine Intelligence and Learning, is Causal AGI
The Causal AGI runs or operates <the Causal World Multi-Hypergraph Network Model W> where the world is described in terms of hyperlinked world's variables; taking on various domain variables and values, categorical, ordinal, interval, ratio or numerical, cardinal (see the Supplement).
A History of Intelligent Machines
AI or AGI as a fiction concept had first emerged in the ancient mythologies and religions, as superintelligent beings, deities or god, followed by Sci-Fi literature and movies. H.G. Wells and I. Asimov introduced the idea of intelligent machines in their works, such as 'The War of the Worlds' and 'I, Robot.' These novels laid the foundation for the anthropomorphic portrayal of AI as human-like robots and their potential to either benefit or harm humanity.
In the emerging computing science and technology, the portrayal of artificial intelligent beings in Sci-Fi was extended as human-like thinking machines (A. Turing, Computing Machinery and Intelligence, 1950 ) and "Artificial Intelligence" defined by John McCarthy as “the science and engineering of making intelligent machines” (1955).
We update McCarthy's classical definition of Artificial Intelligence (AI) as "the science and engineering of making REAL intelligent machines”, rejecting the anthropomorphic AI as an inadequate and unacceptable, faulty, fake and false, nonscientific and amoral technology.
Or, today's human-mimicking AI, with its many subfields, ML, Predictive AI or GenAI, DL, NNs, robotics, expert systems, NLP, large language models (LLMs), chatbots, and AGI, is NOT real AI, true machine intelligence and learning, being "unintelligent stochastic machines" unable of real intelligence, learning, reasoning or interactions.
True AI and Machine Learning is to automatically and autonomously identify, understand, explain, infer, discover or predict possible causal variables, relationships, mechanisms, patterns, rules and laws, systems and networks from the universe of data to interact with the world in the most effective, efficient, rational and sustainable ways.
Real AI Agents are defined as causally intelligent entities which can perceive their world, make inferences and decisions and take actions based on true symmetrical or interactive cause and effect: X causes Y if and only if Y causes X
It is aligned with a probabilistic model of causal reciprocity, known as Bayes' theorem or formula: the joint distribution P(Cause, Effect) = P(Cause)*P(Effect | Cause) = P(Effect)*P(Cause | Effect)
True AI as a Causal AGI has a deep understanding about Why (Cause) and How (Process), as well What (Entity), When (Time), Where (Space).
The scale and scope and variety of causative questions might be as different as "why and how the universe emerged to "why and how general anesthesia works in the brain".
And the Why guides all the rest, the What, the How, the When and the Where, relying on the Causal Multi-Hypergraph Network , generalizing the causal structures, mathematical structures, theoretical structures and statistical learning models about reality, its domains or contents.
Causal AGI machines operate all possible types of networking topologies and graph architectures, directed or undirected, and causal diagrams, with all possible causal relationships as special cases:
1) unidirectional cause-effect relationships, positive or negative linear causal links, as the input-output, cause-effect directed graph mapping a set of causes to a set of effects (decision tables, trees, or conditionals, the if–then-else construct in programming languages, or ML input-output datasets);
2) inverse or reversed cause-effect relationships;
3) bidirectional or symmetrical or mutually dependent, reciprocal causal relationships;
4) common-cause relationships ("fork"), explaining confounding and spurious correlations;
5) common-effect relationships ("collider");
6) causal chains, where each link in the chain represents something in the real world, with the origin of the chain as the root cause, as the origin of the universe;
7) causal feedback loops or circuits or cycles, as chemical or nuclear reactions, forming a stable homeostatic cycle or reinforcing mechanism, a vicious or virtuous circle,
Plus all possible types of reasoning, deduction, induction, abduction or analogy, deterministic or probabilistic, the "bottom up" inductive causality with the "top down" deductive causality.
It includes modern scientific models rooted in the bottom-up, reductionist view of cause and effect, where biology is determined by chemistry, which is governed by the underlying physics of four fundamental interactions.
As a general multi-hypergraph network of world's variable, the Causal AGI Engine covers two notions of causality, type causality or general causality) and token causation or actual causality and specific causality, like a space-type physical causality.
The Causal AGI can generate general rules or common sense statements, such as “smoking causes lung cancer”, "gravity attracts bodies", "sex brings pleasure", "war brings destruction", or “printing money causes inflation”. By way of contrast, token causality focuses on particular events, specific contexts or space-time settings, like “if a central bank prints $X billion, a its country is likely to see X% inflation”.
Thus, AGI Causality is about modeling how things causally interact C =I in the world W:
W = <E, C, R; D> (1)
where the world variable W consists of its fundamental ontological categories as the Entity variable E, the Change variable C, the Relationship variable R, and its Data representation D, x - is the Cartesian product as the universal class/set of all ordered pairs of the categorical variables. It could be visualized as a world matrix constructed from a set of rows and a set of columns, where the cells of the world matrix-table contain ordered pairs of the form (row world's variable/values, column world's variable/values):
I = <W x W> =<E x E; C x C, R x R, D x D; E x D, C x D, R x D...> (2)
Unlike the Causal AGI and its general interactive intelligence, today's statistics-driven AI and ML and LLMs all focus on the training datasets, D x D, "learning form data", missing all reality and causality, C x C, semantics or meanings and senses, E x D, C x D, R x D, mapping or representing the classes of entities, changes or relationships.
The reason of such dramatic inconsistencies is ignoring reality and causality, and misrepresenting the fundamental principle of reality as the linear science of linear cause and effect, with all the consequences.
World Hypergraph Embeddings as General Knowledge and Causal Intelligence
Most ML algorithms can only take numerical data as inputs. Therefore, it is necessary to convert the data D into a numerical format using embedding as a vector (list) of floating point numbers. It is supposed to represent data types objects like words, text, images, audio, graph as numerical points in a continuous vector space where the locations of those points in space are "semantically meaningful to" the embedding models used for similarity search, recommendation, clustering, etc.,
World embeddings , used in various domains and applications, are to transform high-dimensional and categorical data into continuous vector representations, capturing meaningful patterns, relationships and semantics.
World's multi-hypergraph embedding is essential for digitizing general knowledge, causal learning and general intelligence. These embeddings find applications in any fields or domains, including social network analysis, NLP, recommendation systems, biological network analysis, where data can be represented as multi-hypergraphs.
"Popular word embedding models include Word2Vec, GloVe (Global Vectors for Word Representation), FastText and embeddings derived from transformer-based models like BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-trained Transformer)". Having been trained on massive amounts of pre-trained embedding corpora, such as Wikipedia and Google News, they may include word embeddings, text embeddings, image embeddings, audio embeddings, or graph embeddings.
Causal AGI involves causal world hypergraph embedding models as Global Vectors for World Representations, including entity embeddings, state embeddings, change embeddings, relation embeddings, causality embeddings, etc.
Causality as the General Science of Interaction
True Causality is the Science of Interactions, Causal Interrelationships, Associations or Correlations, the study of how things interact, causing, changing, affecting or influencing each other, or how causes lead to effects via interactions.
It is a naive, conventional or classical view of causality as the linear science of cause and effect, where effect follows causes in a linear, predictable, deterministically or probabilistically, as it is preached in “The Book of Why: The New Science of Cause and Effect. ”
In our dynamic nonlinear complex world, big effects could grow from little causes due to nonlinear feedback interactions, as all nonlinear effects or the quantum entanglement n the quantum world.
Again, the language of causality is symmetric, as all physical laws and mathematical rules, structures and equations: If X tells us about Y, then Y tells us about X:
If X causes Y that means that Y causes or could cause X.
So, when we claim that "To Build Truly Intelligent Machines, Teach Them Cause and Effect ", we have to program the causal world model, encoded as the Causal Multi-Hypergraph Network ...
The Iron Law of Intelligence, natural or artificial
To know the world is to know its causes and effects, causal variables, relationships, interactions, mechanisms, patterns, laws, transformations, systems and networks.
Or, any analysis, problem, solution, decision or action needs to be traceable to causes (upstream input) and effects (downstream outputs).
This Iron Rule, reasoning or method applies to all rational things: new experiments, value chains, contingency planning, operational sequences, algorithmic processing, data organization, new discoveries, etc.
Philosophy, mathematics, science, engineering, technology, history, literature and daily life, all define cause and effect and study or practice it in its generic or specific contexts, conditions, settings or environments.
Causal world modeling is the most rational way of understanding the world and how it works.?
Cause and effect is everywhere, except of today's Artificial Intelligence (AI), wit its many subfields, Machine Learning (ML), Predictive AI or Generative AI (GenAI), Deep Learning (DL), Neural Networks, Robotics, Expert Systems, Computer Vision, Natural Language Processing (NLP), large language models (LLMs), chatbots, or causal AI, all relying on statistics and probabilities and correlations or asymmetric causal inference,
Causal Intelligent Machines are to automatically and autonomously identify, learn, understand, explain, infer, discover or predict all possible causal variables, mechanisms, patterns, rules and laws of systems and networks to interact with the world in effective and efficient, rational and sustainable ways.
Causal Intelligent Machines overwrite human-like AI models, LLMs, generative AI, which are limited to recognizing and analyzing correlations across data, and Causal AI "designed to identify and understand the cause and effect of relationships across data".
How Can We Learn AI?
What is AI really about?
AI is the science and engineering of intelligence or intellect, its nature, models, theories, algorithms and architectures.
It is about three interrelated universes: pure intelligence/intellect/mind/understanding; reality/world/environment and its data universe, how it all is represented, mapped, coded and processed by computing machinery of any complexity, from smart phones to the internet and beyond.
AI has philosophical, logical, epistemological, ethical, semantic, cognitive, social and techno-scientific or computing foundations.
All in all, there are two polar kinds of AI: like Boolean data type: True and False. The false one is a sci-fi AI.
Or, Anthropic AI [AAI] vs Real AI [RAI], and RAI in Russian means Heaven.
AAI is like cancer cells, leading to the Black Mirror/Black Museum dystopia, a future techno-world or totalitarian or post-apocalyptic society in which people lead wretched, dehumanized, fearful lives.
The whole idea of AAI as humanoid robots, with human-like AI algorithms, is not just extremely stupid, but existentially risky for all human species. It hardly could be justified by being “useful for future dangerous and/or distant space exploration missions, without having the need to turn back around again and return to Earth once the mission is completed”.
Robotics as developing “machines that can substitute for humans and replicate human actions” is similar to artificial human cloning, which is mostly prohibited.
It was good , but as mechanical or electronic toys, or miniature human-like toy automatons. In 1700s "androides", as elaborate mechanical devices resembling humans and performing human activities, were displayed in exhibit halls.
Male androids or female androids, or "gynoids", are often seen in science fiction, and can be viewed as attempting to create the stereotypical "perfect man/woman"
Androids featured in most science fiction movies, as I, Robot, mentally and physically equal or superior to humans—moving, thinking and speaking as humans.
Android heroes seek, like Pinocchio, to become human, as Bicentennial Man, or Data in Star Trek: The Next Generation, rebelling against abuse by humans, as in the film Westworld, .
It all might end up as with android hunter Deckard in Do Androids Dream of Electric Sheep? and its film adaptation Blade Runner, who discovered that his targets appear to be, in some ways, more "human" than humans.
The true and genuine AI is a non-anthropomorphic scientific AI (SAI) imitating/replicating/simulating/mapping/representing/modeling/computing reality/causality/mentality:
Causal Machine Intelligence and Learning
Real AI/Actual AI
Transdisciplinary AI or Meta-Human Computer Intelligence
The False AI is a Human AI (HAI), human-like and human-level AI imitating/replicating/simulating human cognition/learning/intelligence/reasoning/brains/behavior:
Statistical Learning
领英推荐
Machine Learning
Deep Learning
LLMs
Symbolic AI
Neuro-symbolic AI
AGI, Strong AI
ASI…
“When I read about AI, 80% of the time it’s just flat out wrong information,” says Prof. Stewart Russell of the University of California at Berkeley. In reality, “When I hear or read about AI, 99.9999% of the time it’s just flat wrong information”. This misunderstanding of what AI really is has largely enabled the emergence of fake AI, a subjective, non-scientific human-like, human-level AI.
Fake AI is an aggregation of statistical techniques, mathematical methods, unethical marketing strategies and immature tech solutions. Fake AI does not offer a true competitive advantage, and the market is becoming increasingly aware of what real AI can achieve, fake AI products and novelties are unlikely to stand the test of time.
What is Applied AI or AI Technology?
Applied AI or AI Technology includes Data Science and Engineering (data analytics, statistics, machine learning, deep learning, artificial neural network); simulation, human brains/intelligence, computing, hardware, software and programming, machine vision, NLU, robotics, automation and cybernetics.
Each AI Technology has its own application areas, with its specific algorithms, techniques and methods.
ML technology is applied to big data automation tasks, with its algorithms such as Linear Regression, Logistic Regression, K Nearest Neighbor, Support Vector Machines to make the machine mathematically process the data.
Each NLU, NLP, or NLP Technology has its own application areas, as Chatbots for Business, Text Generation, Sentiment analysis, etc., with its specific algorithms that deal with NLP, as Na?ve Bayes or Recursive Neural Networks.
Now, Computer Vision is a specialised form of Machine Vision to enable a machine to “see” the world around them the way humans see it. The idea is to capture the world through hardware (cameras etc.) and convert the captured images into a form that then the computer can understand, process and use them to come up with meaningful results.
The Computer Vision Technology involves Deep Learning methods as the Convolutional Neural Networks. It includes Medical Image Analysis, Facial Recognition, Authentication of Signature or Self Driving cars.
What are AI [Technology] applications
AI Applications (AIAs) are how AI systems and tools and algorithms and techniques are solving tasks in all parts of human life, government, management, industry, agriculture, manufacturing, finance, engineering, administration, health, education, security, military, etc; the technological, economic, social, political, cultural and environmental impacts of AI.
What is NOT AI?
While studying, it is most important to avoid tautological “oily oil” or senseless popular definitions, as
“AI is intelligence demonstrated by machines, unlike the natural intelligence displayed by humans and animals”
“AI describes machines (or computers) that mimic cognitive functions that humans associate with the human mind, such as learning and problem solving"
“AI simulates human intelligence to solve problems and complex human tasks”
"AI?is the broader concept of enabling a machine or system to sense, reason, act, or adapt like a human"?
"ML?is an application of AI that allows machines to extract knowledge from data and learn from it autonomously"
"Artificial intelligence is a broad field, which refers to the use of technologies to build machines and computers that have the ability to mimic cognitive functions associated with human intelligence, such as being able to see, understand, and respond to spoken or written language, analyze data, make recommendations, and more"
"Machine learning is a subset of artificial intelligence that automatically enables a machine or system to learn and improve from experience,,, using algorithms to analyze large amounts of data, learn from the insights, and then make informed decisions", etc.
Last, not least, one has to see the principal difference between today's AI and ML.?
The former is about a machine that can mimic human intelligence, while the latter aims to train/teach a machine, its algorithms, to perform a specific task, produce predictive models and provide accurate results by identifying patterns.
Again, the patterns must be causal, objective, real and true, sensible and insightful, not meaningless statistical models, correlations and probabilistic associations. ?
True AI/ML systems rely on causal models to learn, infer, self-correct and interact effectively with the world, humans and other machines.
To Sum Up
Today's statistical AI, with its many subfields, ML, Predictive AI or GenAI, DL, NNs, robotics, expert systems, NLP, large language models (LLMs), chatbots, etc., is NOT real AI, true machine intelligence and learning, being "dumb stochastic machines" unable of real intelligence, learning, reasoning or interactions.
Artificial Intelligence is the science and technology of developing computers and robots as intelligent causal machines.
Who will win the global AI race? It is who first to build true AI as Intelligent Causal Machines.
And the group of big technology giants, Microsoft, Apple, Nvidia, Amazon, Alphabet (Google), Meta (Facebook), and Tesla, has nothing to do with that all.
Resources
SUPPLEMENT: the Industrial Applications for Causal Intelligent Machines
As introduced, the Causal AGI machines run or operate <the causal world multi-hypergraph network model> where the world is described in terms of hyperlinked causal variables; taking on various values, categorical. ordinal, interval, ration or numerical.
That means that industrial domains could be mapped as multi-hypergraph projections, like it is listed by the LONDON INSTITUTE OF SKILLS DEVELOPMENT:
Education Industry
Agricultural Industry
Mining Industry
Gas Industry
Energy Industry
Chemical Industry
Pharmaceutical Industry
Hospitality Industry
Farming Industry
Creative Industry
Textile Industry
Manufacturing Industry
Financial services Industry
Fishing Industry
Electric power Industry
Automotive Industry
Music Industry
Green Industry
Engineering Industry
Computer Industry
Real estate Industry
Information Technology Industry
Fashion Industry
Cultural Industry
Transport Industry
Media Industry
Aerospace Industry
Merchandising Industry
Food Industry
Electronics Industry
Basic metal Industry
Oil and gas Industry
Health services Industry
Entertainment Industry
Construction Industry
Retail Industry Leisure Industry
Film Industry
Defense Industry
Wholesale Industry
Communications Theorist ,AI Technology, AI Ethics , Researcher, Author
6 个月Your article offers a thought-provoking exploration of the fundamental role of causality in developing truly intelligent machines. By emphasizing that real AI must go beyond current models, which you critique as limited and stochastic, and instead focus on causal relationships and interactions, you present a compelling vision for the future of AI. This perspective challenges the status quo and pushes for a more profound understanding of intelligence, whether natural or artificial. A key question to ponder: How can we practically integrate causal models into existing AI systems to bridge the gap between current capabilities and the true intelligence you envision?
Web Developer @Altq | CS Grad at Concordia | Former Nokia Engineer | AI & Blockchain-Enthused
6 个月Focusing on causal intelligence is key to advancing AI. Besides computational power and resources, what are the main challenges in developing causal intelligent machines today?