Real AI vs. Unreal AI = Causal Intelligence Interactive Machines (CIIM): Converging Symbolic, Predictive, Generative, Industrial, and Causal AI
"There are real and true, scientific and objective AI or unreal and fake, imitating and subjective AI. To build true intelligent machines, teach them how to interact with the world, transforming deep neural networks into causal world graph networks". ASHA
Key Message
True causality or real causation or interaction modelled as full causal world graph networks is how machines or humans understand, explain, and make decisions, adjust, navigate, manipulate or interact with the world.
Intellectualize artificial intelligence and machine learning with causal world modelling to get an explainable and interactive, true and real AI.
Power causality, its mechanisms and laws, analysis and modeling, reasoning and inference, with AI models and deep machine learning algorithms and data science and engineering techniques to create a general-purpose, trustworthy AI (“valid and reliable, safe, fair, and?bias?is managed, secure and resilient, accountable and transparent, explainable and interpretable, and privacy-enhanced”).
All current AI platforms, ML applications and deep learning systems, as LLMs, conversational AI or chatbots or recommender engines or search engines or industrial AI belong to a special class of unreal, fake or false AI.
The key beneficiaries are the largest high-tech corporations capitalizing on the fake AI technology, Apple and Amazon, Alphabet and Microsoft, Meta Platform or Tencent, NVIDIA or Intel, Tesla or TSM, Baidu or Huawei, IBM or Intel.
The big tech oligopolies, from Apple to Facebook, Microsoft to Google, Alibaba to Tencent, have managed to exponentially enhance their global destructive influence due to deepfake ML/AI/LLMs applications and other social engineering digital technologies...There is a freshly published book Technology and Oligopoly Capitalism studying how technology oligopolies are shaping America’s social, economic, and political reality.
We have to mention here the fake AI startups, like OpenAI, "with a mission of creating artificial general intelligence (AGI) that benefits all of humanity. No less. no more. The top 100 is listed below:
In all, there are currently?approximately?58,000?AI companies worldwide; 1 in 4?AI companies are based in the US; there are?over 115 million?companies currently using AI; almost half?(42%)?of companies are exploring the use of AI. According to the Global AI Index (June 2023), the US has the highest AI capacity in terms of AI talent, infrastructure, research, development, government strategy, and commercial viability.
Real AI's 10 commandments:
Real AI = CIIM = Interactive AI (IAI) = Reality AI Engine = Trustworthy AI= Correlation AI [Generative AI, ML, DL, ANNs, NLP] + Causal AI + World Model Engine
As noted, there are real and true, scientific and objective AI or unreal and fake, imitating and subjective AI.
Real AI is causal machine intelligence and learning augmenting human intelligence, cognitive capabilities, decision-making processes, and productivity.? It is the science and engineering focusing on building and managing an integrative, general-purpose technology that can causally learn, infer, make decisions to effectively and autonomously interact with the world.
Unreal AI is "a branch of computer science that focuses on building and managing technology that can learn to autonomously make decisions and carry out actions?on behalf of a human being".
Real AI is a general-purpose computing technology that includes the most advanced software and hardware supporting narrow and weak AI models, artificial neural networks, machine and deep learning, symbolic AI, rules-based expert systems,?discriminative and generative AI, cloud and edge AI, NLP and NLG, the internet of things and robotics, and all sorts of industrial AI systems.
Unreal AI includes three classes:
So, if Narrow AI is to do one task and solve one given problem, as a chatbot, General Artificial Intelligence (GAI) is a human-like and human-level AI applied in any domain, solving any problem, while operating in a changing environment.
Applications of ANI are unable reasoning for themselves but simulating some cognitive functions, or replicating some human behavior based on a set of rules, parameters, and contexts that they are trained with. Some of the most common NAI techniques are?machine learning, natural language processing, or?computer vision, as search engines, recommender systems, LLMs or object/face/speech recognition software.
All NAI and machine learning models fail at generalizing beyond their narrow domains and training data.
“Machine learning often disregards information that animals use heavily: interventions in the world, domain shifts, temporal structure,... the majority of current successes of machine learning boil down to large scale pattern recognition on suitably collected?independent and identically distributed (i.i.d.)?data.”
“Generalizing well outside the i.i.d. setting requires learning not mere statistical associations between variables, but an underlying causal model,” the AI researchers write.
The cost of Narrow AI one could see from its overhyped generative AI and its "so-called large language models like ChatGPT which have scraped vast parts of the internet to assemble data that inform how the chatbot responds to various inquiries. The data-mining is conducted without permission. Whether hoovering up this massive repository is legal remains an open question.
If OpenAI is found to have violated any copyrights in this process, federal law allows for the infringing articles to be destroyed at the end of the case.
In other words, if a federal judge finds that OpenAI illegally copied the?Times' articles to train its AI model, the court could order the company to destroy ChatGPT's dataset, forcing the company to recreate it using only work that it is authorized to use.
Federal copyright law also carries stiff financial penalties, with violators facing fines up to $150,000 for each infringement "committed willfully."...
Getty Images is suing Stability AI for allegedly training an AI model on more than 12 million Getty Images photos without authorization."
Correlation-based AI and causal AI have essential differences, including the following:
[Correlation-based AI relies on statistics to provide assumptions about what’s happening.
Causal AI can clearly trace and explain exactly what’s happening at every step based on specific contextual data.
Correlation-based AI is probabilistic and requires humans to verify the accuracy of results.
Causal AI is fact-based and thus can do automated analyses.
Correlation-based AI can make only predictions with limited ability to explain an event.
Causal AI, on the other hand, provides details on how it arrived at a conclusion.
Correlation-based AI needs to be checked for bias due to the limitations of various data, algorithms, or sampling.
Causal AI, however, relies on actual data and not training data and is therefore not prone to bias issues.
Correlation-based AI may be completely off base in novel situations.
Causal AI can adapt to new situations and find unknown unknowns].
ML/AI needs a substantial paradigm shift in the worldview and concepts, models and algorithms, techniques and and practices of how it works and accomplished.
A paradigm shift is a major change in how people think and get things done that upends and replaces a prior paradigm.
A paradigm shift can result after the accumulation of anomalies or evidence that challenges the status quo, or due to some revolutionary innovation or discovery.
There is often resistance to a new paradigm coming from incumbents, as the mainstream AI stakeholders.
As a matter of fact, any statistical, narrow and weak AI systems, applications, networks or devices should integrate causal intelligence and learning, at least, to be real intelligent. Otherwise, it is all a quasi-AI or fake AI, imitation ML or counterfeit DL or false ANNs, a simulated AI, including all its forms and uses, as pictured above and below.
Causal AI vs. Acausal AI
In Artificial Intelligence, a paradigm transition from a knowledge-based to a data-driven correlative AI paradigm has begun in 2010.
Such an Acausal statistical and probabilistic AI (AAI) as implemented by machine learning models, deep learning algorithms and artificial neural networks has nothing common with human intelligence or human consciousness.
AAI/ML/DL is mostly statistics and big data analytics, probabilities and applied mathematics, software programming and hardware computation, searching for some optimal solutions in a multidimensional data space using?hyperdimensional computation.
It’s all about making predictions using data hierarchies of feature detectors up to hundreds of layers, and "learning" via adjusting the model's parameters, gradient descent or weights by?backpropagation/error-correction.
Today's ML AI does not take reality and its causality into account, and instead is merely concerned with prediction based on statistical associations.?
Here come a market demand for AI which is able to handle cause and effect, with all its complexities, scales and scopes and levels, dubbed as a Causal AI:
Over the last few decades, computer scientists and statisticians have developed a framework for a linear causal inference, a 3-level hierarchy of causal reasoning, answering three kinds of questions.
The headline successes of AI over the past decade — such as winning against people at various competitive games, identifying the content of images and, in the past few years, generating text and pictures in response to written prompts — have been powered by deep learning. By studying reams of data, such systems learn how one thing correlates with another. These learnt associations can then be put to use. But this is just the first rung on the ladder of linear probabilistic causation.
Causal inference has long been used by economists and epidemiologists to test their ideas about causation. The 2021 Nobel prize in economic sciences went to three researchers who used causal inference to ask questions such as whether a higher minimum wage leads to lower employment, or what effect an extra year of schooling has on future income. Now, there is a growing number of computer scientists who are working to embed causality with AI to give machines the ability to tackle such questions, helping them to make better decisions, learn more efficiently and adapt to change.
Causal AI as Causal Machine Learning as "improving machine learning with causal reasoning, and automating causal reasoning with machine learning" is viewed by some big tech, as Microsoft or Meta, as a prospective replacement of Fake AI.?
There is the whole book recently published, Causal AI, with the following points
"Enhance machine learning with causal reasoning to get more robust and explainable outcomes. Power causal inference with machine learning to create next gen AI. There has never been a better time to get into building causal AI".
Causal machine learning is a major milestone in machine learning, allowing AI models to make accurate predictions based on causes rather than just correlations. Causal techniques help you make models that are more robust, explainable, and fair, and have a wide range of applications, from improving recommendation engines to perfecting self-driving cars.
Causal machine learning gives you the insight you need to make predictions and control outcomes based on causal relationships instead of pure correlation, so you can make precise and timely interventions.
Interactive and True AI vs. Causal and Fake AI
The nature of causality and causation is more deep than causal inference modelling, and true Causal AI is much more insightful.
Namely, "causality is fundamentally symmetric", “correlation or association does imply causation”, and the formalisms, mechanisms, and techniques of causal inference are in need of paradigm shift.?
Real causation is interaction, a reciprocal action or influence or productive correlations and causative associations, or X causes Y if and only if Y causes Y.
X affects, produces or influences or changes Y by a transfer of matter, energy, force or information as much as Y affects, produces or influences or changes X by a transfer of matter, energy, force or information.
Its solid evidence comes from the hard science of physics where the fundamental forces are interactions, a way in which matter, fields, and atomic and?subatomic?particles affect one another, e.g. through?gravitation and electromagnetism, strong and weak nuclear forces.
In statistics, "an?interaction?may arise when considering the relationship among three or more variables, and describes a situation in which the effect of one causal variable on an outcome depends on the state of a second causal variable (that is, when effects of the two causes are not?additive)".
In fact, statistical interaction means?the effect of one independent variable(s) on the dependent variable depends on the other independent variable(s) as well as the dependent variable itself. If n causal variables of interest interact, the relationship between each of the interacting variables and a selected "dependent variable" depends on the values of all interacting variables, independent and dependent.
Then the statistical and ML workhorse, regression analysis,?is no more valid as "a set of statistical processes for?estimating?the relationships between a?dependent variable?(a "response variable", "regressand", "criterion", "predicted variable", "measured variable", "explained variable", "experimental variable", "responding variable", "outcome variable", "output variable", "target" or "label", endogenous variable)
and one or more?independent variables?( a "predictor variable", "regressor", "covariate/control", "manipulated variable", "explanatory variable", "exposure variable", "risk factor", "feature" or "input variable", an exogenous variable).
It is necessary to account for interaction variables/factors/determinants and for the fact that a change in an exogenous variable is reciprocally interrelated with a change in an endogenous variable and that the antonym pairs distinction as independent-dependent is arbitrary and conditional and extraneous confounding variables, as subject, experimental, situational or environmental, are of critical importance:
Z = a + bX +cY + d(X x Y) +.e(X x Z) + f (Y x Z) + g (X x Y x Z)..
Including into the causative regression analysis all the possible interaction variables, we avoid all sort of biases, confounding or omitted-variables and errors, as the "residual", "side effect", "unexplained share", "residual variable", "disturbance", or "tolerance", containing the variability of the dependent variable not explained by the independent variable.
Basing on the holistic conception of causation or interaction, we have developed a comprehensive realistic world model enabling the Interactive AI
to have a causal model of its environment,
to do formal reasoning about cause and effect,
to identify the causal relationship between things,
to learn and model non-linearities and complex relationships,?
to learn causal structure from data,
to infer?causal?relations from [observational and experimental] data,
to discover a causal graph from big data, structured or unstructured,
to infer causality from correlations,
to estimate a joint probability distribution over all the variables of the universe of interest,
to make realistic predictions and decisions to effectively interact with the world.
Types of causal representation learning, inferences and algorithms
Several types of causal AI models are developed as a result of observing causal relationships of all possible types and sorts, causing, allowing, preventing, producing, creating, growing, etc.: linear causality, common-cause relationships and common-effect relationships, causal chains, reverse causality and causal cycles
A data input variable is a single cause resulting in several effects (as food intoxication (a pathogen) - fever, headache and nausea).
A warfare is an example of one effect with several causes.
Examples, digital illiteracy is leading to disinformation and miscommunication, which leads to mental confusion. Sanding causes dust?and?Dust causes sneezing, Sanding causes sneezing.
In epidemiology, it's when the exposure-disease process is reversed, the exposure causes the risk factor.
The emergency of flight effect is explained by causal reinforcing interactions of feathers, hollow bones, high metabolic rate and flight reinforce each other in birds.
Causation and Interaction as the essence of human and machine realities
How causality can be unambiguously determined, defined and demonstrated?
In science and engineering, causality is demonstrated by experiments and computer simulations.
In social sciences, causality is demonstrated by observational studies or A/B test..
In clinical medical research, causality is demonstrated by randomized controlled trials (RCTs).
All in all, causality is present when 5 conditions of sufficiency, necessity, reversibility, symmetry and probability are satisfied:
1) A is a sufficient cause of B if B always happens when A has happened; B always follows A—in which case, A is called a sufficient cause of B
2) A is a necessary cause of B if B only happens when A has happened; if A does not occur, then B does not occur—in which case, A is called a necessary cause of B
3) A produces, leads to, creates, influences B if and only if B causes changes in A
4) cause and effect are a symmetrical relationship, causative association, interactive correlation; i.e. correlation implies causation
5) sufficient and necessary causes are the deterministic cases of a symmetrical probabilistic or stochastic causation Bayes' law/rule/theorem P (A/B) P (B) = P (B/A) P (A) in which P (B | A) = 1 and P (B | not A) = 0, respectively.
The Causal World Model Engine AI as the Undirected Cyclical Causal Graph Networks
One of the key applications of the causal theory of interaction is the real-world machine intelligence and learning, or real and true AI, which is driven by the causal world [learning, inference and interaction] model engine AI.
Full causation as interaction is formally encoded by full causal undirected cyclic graph networks embracing all known and unknown artificial neural networks architectures.
Full causal graph networks allow the Reality AI Engine to compute/infer/reason about the flow of mass, energy or information.
Besides, it could be formalized as a joint probability distribution over all the interacting random causal variables of the universe of interest and visualized as intersecting Venn diagrams. The joint probability function is expressed in terms of chain of conditional probabilities (the fundamental chain rule for probability calculus):
P(A1,A2,A3...,An) = P(A1|A2,A3,..,An)P(A2|A3, ...,An) ... P(An)
Thus, an expression of P(digital illiteracy, nationality, gender, education) could describe the probability of a person being digitally illiterate depending on his nationality, gender and education.
As an example, the number of free parameters in a raw joint probability distribution over a million binary variables would be 2exp [1,000,000] -1.
See the interaction table showing the number of terms for each number of predictors and maximum order of interaction.
All in all, all ANI, as Generative AI with ML/DL Algorithms, is to be transformed into CAUSAL AI which is to be transformed into INTERACTIVE AI...
By its nature, the interactive AI is a causal system or cyber-physical system or anticipative system, where the output depends on past and current and future inputs.
y(t) = f(x(t - a), x(t), x(t + a)), where x(t - a) is the previous state (s), x(t) - the present state (s), x(t + a) is the future state.
As such, a?full causal system?(as a real physical system or anticipative system) is a?system?where the output (outputs and internal states) depends on past and current and future inputs (values). It makes invalid the conceptual misassumptions as causal, acausal and anti-causal systems due to a defective linear causation.
Real-world AI as the Causal Intelligence Interactive AI has all the necessary capabilities of powerful intelligence, as
descriptive,
exploratory,
explanatory,
deductive,
inductive,
abductive,
retrodictive computing capacities.
CIIM AI and LLMs
Today's AI is after building more powerful large language models (LLMs) to make them larger to train them on more data, to become an "infinitely big" model, trained on "infinite data."
Most of today’s leading LLMs were trained on data corpuses of about 300 billion tokens, including OpenAI’s GPT-3 (175 billion parameters in size), AI21 Labs’ Jurassic (178 billion parameters in size), and Microsoft/Nvidia’s Megatron-Turing (570 billion parameters in size).
But how much more language data is there in the world for training??
The world’s total stock of high-quality text data is between 4.6 trillion and 17.2 trillion tokens. This includes all the world’s books, all scientific papers, all news articles, all of Wikipedia, all publicly available code, and much of the rest of the internet, filtered for quality (e.g., webpages, blogs, social media). [Will we run out of data? An analysis of the limits of scaling datasets in Machine Learning]?
Another?recent estimate?puts the total figure at 3.2 trillion tokens.
Say, DeepMind’s Chinchilla model was trained on 1.4 trillion tokens, while multimodal GPT-4 could be trained on a dataset at least an order of magnitude larger than this, as large as 10 trillion tokens..
In other words, they are close of exhausting the world’s entire supply of useful language training data, a material impediment to brute-force progress in language AI.
From Deep Learning and Reasoning Systems Towards CIIM
We outline the agent-environment (the real world, virtual reality, etc.) interaction system architecture converging Symbolic, Predictive, Discriminative, Generative, and Causal AI:
CIIM = Real AI system = Reality AI Engine = Causal Input/Flow of Matter, Energy and Information (stimulation and signals, multi-modal machine perception, vision, speech, sound,...) + Machine Reasoning (causal inference, reasoning/decision making (GOFAI) + Machine Learning (causal learning algorithms, deep learning (ML & DL) + Causal Output/Flow of Matter, Energy and Information (behavior, actions, actuation, communication) + Causal Backpropagation (adjusting the model's parameters, adapting to the environment).
AI as a learning rational system has been defined by High-Level Expert Group on Artificial Intelligence, an independent expert group that was set up by the European Commission in June 2018.
Figure 1: A schematic depiction of an AI system.
Both machine learning and reasoning include many other techniques, and robotics includes techniques that are outside AI. The whole of AI falls within the computer science discipline.
Sensors and perception. In Figure 1 the system’s sensors are depicted as a wifi symbol. In practice they could be cameras, microphones, a keyboard, a website, or other input devices, as well as sensors of physical quantities (e.g. temperature, pressure, distance, force/torque, tactile sensors). In general, we need to provide the AI system with sensors that are adequate to perceive the data present in the environment that are relevant to the goal given to the AI system by its human designer. For example, if we want to build an AI system that automatically cleans the floor of a room when it is dirty, the sensors could include cameras to take a picture of the floor.
As to what regards the collected data, it is often useful to distinguish between structured and unstructured data. Structured data is data that is organized according to pre-defined models (such as in a relational database), while unstructured data does not have a known organization (such as in an image or a piece of text).
Reasoning and Decision Making. This group of techniques includes knowledge representation and reasoning, planning, scheduling, search, and optimization. These techniques allow to perform the reasoning on the data coming from the sensors. To be able to do this, one needs to transform data to knowledge, so one area of AI has to do with how best to model such knowledge (knowledge representation). Once knowledge has been modelled, the next step is to reason with it (knowledge reasoning), which includes making inferences through symbolic rules, planning and scheduling activities, searching through a large solution set, and optimizing among all possible solutions to a problem. The final step is to decide what action to take. The reasoning/decision making part of an AI system is usually very complex and requires a combination of several of the above mentioned techniques.
Learning. This group of techniques includes machine learning, neural networks, deep learning, decision trees, and many other learning techniques. These techniques allow an AI system to learn how to solve problems that cannot be precisely specified, or whose solution method cannot be described by symbolic reasoning rules. Examples of such problems are those that have to do with perception capabilities such as speech and language understanding, as well as computer vision or behaviour prediction. Notice that these problems are apparently easy, because they are indeed usually easy for humans. However, they are not that easy for AI systems, since they cannot rely on common sense reasoning (at least not yet), and are especially difficult when the system needs to interpret unstructured data. This is where techniques following the machine learning approach come in handy. However, machine learning techniques can be used for many more tasks than only perception. Machine learning techniques produce a numeric model (that is, a mathematical formula) used to compute the decision from the data.
Actuation. Once the action has been decided, the AI system is ready to perform it through the actuators available to it. In the cartoon above, the actuators are depicted as articulated arms, but they don’t need to be physical. Actuators could be software as well. In our cleaning example, the AI system could produce a signal that activates a vacuum cleaner if the action is to clean the floor. As another example, a conversational system (that is, a chatbot) acts by generating texts to respond to user’s utterances.
The action performed is going to possibly modify the environment, so the next time the system needs to use its sensors again to perceive possibly different information from the modified environment.
Rational AI systems do not always choose the best action for their goal, thus achieving only bounded rationality, due to limitations in resources such as time or computational power.
Rational AI systems are a very basic version of AI systems. They modify the environment but they do not adapt their behaviour over time to better achieve their goal.
A learning rational system is a rational system that, after taking an action, evaluates the new state of the environment (through perception) to determine how successful its action was, and then adapts its reasoning rules and decision making methods.
To the Real-World AI
The next generations of AI are Causal Learning Rational Systems to be replaced by Causal Intelligence Interaction AI Technology, Real and True AI, or Interaction AI.
The Interaction AI is the whole new game unknown to the big tech fake/false AI/ML/DL playing with spurious correlations and statistical associations as data regularities and patterns of various input types [labeled or unlabeled].
Using statistical [machine] learning algorithms to identify patterns, it blindly classifies data based on statistical information gained from patterns and their representation.
As mentioned, all current AIs are weak and narrow AI, being focused on performing a very narrow field of tasks, following the ML life cycle.
It is missing the core element or module to be real intelligent understanding machines, intelligence mechanism?per se, as the encoded/programmed Real?AI World Model (Intelligence/Learning/Inference/Interaction) Engine.
Basing on the universal ontology/data framework, causal world model and science at large, its special models and laws, and causal master algorithms, the World Model Engine is transforming the causal input of matter, energy or information, as raw data, structured and unstructured, into the causal output og matter, energy and information, as decisions and predictions, knowledge and wisdom and rational interactions, performing designed tasks and solving any complex problems.
Instead, the big tech companies usually have created huge internal [embarrassingly parallelizable, perfectly parallel, delightfully parallel or pleasingly parallel] systems which are capable of crawling, ingesting, and processing the biased data of Wikipedia, Quora, Reddit. Without large companies publishing their LLM models to the open source world, startups wouldn’t have a chance to create their applications, be it chatbots, virtual assistants, voice assistants, or other automated online software.
If you think that AI is a statistical ML, you are badly and sadly mistaken. Even the most advanced deep-neural network ML systems can’t yet do simple tasks such as abstraction, generalization and explanation. To build real AI, we will need narrow AI, ML, DL, symbolic AI or knowledge graphs to work together.
All the ML in the world will not help you if you have no causal world model strategy to represent the world’s data/information/knowledge and perform reasoning, abstraction, planning, decision, prediction and action on this data/information/knowledge.
The processes and components in Real AI/ML/DL models computing the causal real-world or digital data are as follows:
RAI = World’s Data/Information/Knowledge + World Model (Intelligence/Learning/Inference/Interaction) Engine+ Statistical Natural language processing (NLP) + NLU + NLG + ML/DL algorithms processes + Industrial AI + Data Computing Infrastructure (Software/Hardware)
Conclusion
The IAI?as the Reality AI Engine could be the industry’s first hyperdimensional AI (HyperAI) or transdisciplinary AI (Trans-AI), converging logical, fact-based, statistical-, predictive-, generative- and causal-AI models and algorithms, techniques and capabilities.?
SUPPLEMENT: If Real AI is just Coding?
Real AI is a Data/Information/Knowledge Processing Causal System or Intelligent Entity using Computing, Programming, Algorithms and Coding, as advanced hardware and software.
RAI collects, processes and generates coded “representations”, data, numbers, symbols or signals, as programmable cognitive functions, sensation, perception, learning, reasoning, self-knowing, understanding, decision making and acting.
Coding is inputting a collection of data as instructions into a programming language to convert into low level instructions that machines understand, as machine language/code.
Algorithms as automated data instructions can be simple or complex, depending on how many layers deep the initial algorithm goes.
Machine learning and artificial intelligence are both sets of algorithms, but differ not depending on whether the data is structured and unstructured, as widely used to think.
It is rather universal ontology data models, schemata or global ontology with the causal/mental models of reality, world views, common sense knowledge, science, domain ontologies, theories of mind, etc., what makes all the difference.
Again, the true RAI is an informational intelligent entity, a cyber-physical system of causal mechanisms, control and computation, information and data processing by the intelligent hardware and software of causal neural networks and causal learning and inference master algorithms that can self-learn and self-improve and self-code and self-organize.