Real Science and Technology, or how to build Real/Interactive/General AI vs. Spurious/Statistical/Narrow AI
What is Real Science and Technology (RST)?
Real science is all about the causes of things and laws of nature, while technology is about applying the [causal] knowledge about the world,?to the practical goals of human life,?to the change and manipulation of the human environment. As simple as that.
Today's science and technology is lost its basic subjects and goals, investigating, learning, knowing and applying the causal patterns of the world, the fundamental causes and forces and laws of its objects and facts, appearances, occurrences and phenomena, with all possible interactions, structures, systems and networks,
Phenomenology as the multi-fragmented sciences of observations overruled Ontology as the integral causal science of deep causes and forces. What is visible or directly observable and measurable is overruled what is really exists as its lurking causal variables and confounding causes.
Here is a relevant passage from the Royal Society publishers' article, Fake science and the knowledge crisis: ignorance can be fatal": Ignorance of the truth, or knowledge that is not acted upon, can be fatal. This basic principle applies at levels from personal to planetary. Fakery affects science as well as everyday social information and, since the two have become highly interactive globally, a vicious cycle is now operating on an increasing scale. The fake news/fake science cycle undermines the credibility of science and the capacity of individuals and society to make evidence-informed choices in their best interests".
As for now,?illiteracy of all sorts and types is intolerable. Here is the shocking case of?AI literacy?in America (16%).
The lost RST is the root cause of all the illiteracy causally associated with big noisy data, fake news, information overload, cognitive fogs, deepfakes, false accounts, spamming bots, generative algorithms, manipulating software, synthetic data, redundant memes, social media commercial newsfeeds, etc.
By the end of 2022, there will be?97 zettabytes?of data in the world, which is mostly the data size of the internet, the amount of information created, captured, copied, and consumed on the web.
Up to 2025, global data creation is projected to exceed 180 zettabytes. A zetabyte is about a trillion gigabytes.
Where is the Life we have lost in living?
Where is the wisdom we have lost in knowledge?
Where is the knowledge we have lost in information?
T.S. Eliot, “The Rock”
Returning to the Roots of Science and Technology
The study of Causal Science extends from ancient philosophy to contemporary science. Causal learning has its roots in metaphysics since Aristotle's four causes, four explanations, Four Whys or Four Becauses:
The material cause of a change or movement (That from which..., what something is made of, "that out of which" it is made),
The formal cause of a change or movement (That into which...i.e., structural, the essence of the object, how something is made, its structure, design and form),
The efficient/moving cause of a change or movement (That by which, moving; the source of the objects principle of change or stability, necessary for the effect’s existence, a chicken is the cause of an egg),
The final cause of a change or movement (That for the sake of which...,i.e., functional, the purpose, the end/goal of the object, or what the object is good for, an egg is the cause of a chicken).
At the beginning of the second book of his “Posterior Analytics”, Aristotle claims that there are four questions for investigating the nature of things and their properties, to obtain scientific knowledge.
With four fundamental types of answer to the question "why?", added with what/nature, whether/existence, who/agents, where/place, when/time, and how/method, an intelligent agent could have causal understanding and deep learning, knowing things or objects or phenomena or processes or relationships of any scales and scopes, simplicity and complexity.
For Aristotle, each science consists in the causal investigation of a specific department of reality, resulting in causal knowledge; that is, knowledge of the relevant or appropriate causes. For knowing what a cause is, and how many kinds of causes there are, is essential for a successful investigation of the world around us.
For those laymen and sophisticated researchers, academics, businessmen and politicians, who never heard of Aristotle, is a short reminding. Aristotle innovated many subjects of knowledge: physics,?biology,?zoology,?metaphysics,?logic,?ethics,?aesthetics,?poetry,?theatre,?music,?rhetoric,?psychology,?linguistics,?economics,?politics,?meteorology,?geology, and?government. It is from his writings and teachings that?the West?inherited its intellectual?language and lexicon, as well as problems and methods of inquiry.
So, True Science is about a systematic causal learning of reality, from mathematics, as the formal modeling of reality, to philosophical, mathematical, natural, cognitive, social and technological sciences, as iy is pictured in the diagram.
In Real and True Causal Science and Technology, a scientific theory is to explain how and why things change, interact, evolve, indicating the driving forces and mechanisms and predicting all possible effects.
The orthodox science is generally defined as "a systematic endeavor that builds and organizes knowledge in the form of testable explanations and predictions about the universe".
It must be "a systematic endeavor that builds and organizes CAUSAL knowledge in the form of testable explanations and CAUSAL predictions about the universe".
Or, Science is?the pursuit and application of knowledge and understanding of the natural and social world following a systematic methodology based on evidence.
It must be "the pursuit and application of CAUSAL knowledge and understanding of the natural and social and digital world following a systematic methodology based on CAUSAL evidence".
Causal Science and Technology is the only form of science and technology, true and real science and technology. Every observation is CAUSALLY predictable and repeatable. All the CAUSAL variables can be completely controlled. Everything is straight-forward. We can look at the facts and say that changing this CAUSAL factor will have that effect - every time.
Lacking the CS&T, relying on the orthodox acausal science, results in all sorts of scientific anomalies, as non-science, pseudo-science, anti-science, creation science, fiddling science, speculative science, fraud science.?
As the non-science goes philosophical, mathematical, mythological, religious and metaphysical formulations. Meantime the mathematical sciences make all the orthodox science, as statistics and data science, theoretical astronomy,?theoretical physics,?theoretical?and?applied mechanics,?continuum mechanics,?mathematical chemistry,?actuarial science,?computer?and?computational science,?quantitative biology,?operations research,?control theory,?econometrics,?geophysics?and?mathematical geosciences.?
The orthodox science has been reacted the mass belief in ancient astronauts,?climate change denial,?dowsing,?evolution denial,?astrology,?alchemy,?alternative medicine,?occultism,?ufology, and?creationism.
The pseudo-scientific projects as "artificial intelligence", "machine learning", "deep learning", "cold fusion", "hot fusion" are driving all the megaprofits of the big tech corporations and their "deep technology" startups.
CAUSAL SCIENCE AND TECHNOLOGY: Causality, Spontaneity and Reversibility
To know is to know all the causes and their effects,?We start understanding causes after 8 months due to the spontaneous?inference capacity.
There are two types of processes in reality: spontaneous and non-spontaneous, falling under spontaneous causation or non-spontaneous?causation.
Most processes in the world look as?spontaneous, automatic, voluntary or unforced, happening naturally under certain conditions, without external forces or energy input or influences.
Examples are innumerable, as from nuclear physics to thermodynamics to chemistry to human's?behavior and beyond:
water flows downhill vs pumping uphill
bodies fall down vs uplifting
water melts vs, freezing
iron rusts/oxidize vs. deoxidize
radioisotopes decay/fission vs. fusion
Nuclear decay or disintegration or radioactivity is the interaction effect of three causal forces, the fundamental laws of nature, the strong nuclear force (alpha decay, the weak force (beta-decay and electromagnetism (gamma-decay).
As I said on many occasions, all processes reverse in nature, which is the fundamental causal principle of reversibility.?
A?spontaneous?process in one direction [under a particular set of conditions] is nonspontaneous in the reverse direction,?as nuclear fission vs. fusion.
It is real/causal science and no magic.
Note that the first who described?spontaneous causation was Aristotle.
It is Spinoza's Causa Sui which is relevant to the point.
He viewed God as Nature and Nature as God, a substance which is absolutely self-caused, infinite and eternal.
The metaphysical question: "What is?" answered as
All is "Substance, attributes, modes"
A polar position is Everything is "Process, attributes, modes"
As combined, All is " Interaction, substances + processes".?
All is "Interaction and Reciprocal". Everything?is integral and interacts with everything else.?
Paraphrasing Lord Rutherford, that there are two kinds of science, "Physics" and "stamp collecting", there are two kinds of science and technology, Causal Science and Technology and the rest as the "fake science and technology".
To make our point, Real Science & Technology vs. Fake Science & Technology, we briefly evaluate some harmful effects of today's AI as the leading science and technology.
Causal MIL vs. Spurious AI/ML/DL
All AI's industrial applications could be classified as artificial narrow intelligence systems (narrow and weak AI). Its most advanced deep reinforced learning models are aiming to reach a human-level and human-like AI to simulate the breadth and depth of the human intellect, rather than focusing on more specific or narrower types of tasks in some specific domain, training on billions of labeled examples with no understanding and knowledge transfer.
The orthodox [pseudo-science] program of creating "artificial general intelligence" (AGI) as a human-like and human-level AI deserves a special attention. AGI pictured as the holy grail of all modern science and technology, and stubbornly pursued by the big tech corporations, as the Meta Platforms/Facebook, Alphabet/Google, Microsoft/OpenAI.
Here are the expectations of national security expert General Robert Spalding and Michael Hochberg, president of Luminous Computing: "The stakes could not be higher in training artificial general intelligence systems. AI is the first tool that convincingly replicates the unique capabilities of the human mind. It has the ability to create a unique, targeted user experience for every single citizen. This can potentially be the ultimate propaganda tool, a weapon of deception and persuasion the likes of which has not existed in history".??
Meantime the causal science and technology enables creating the best ever human invention, "the first ultra-intelligent machine, the last invention that man need ever make", as True Real AI or Causal Machine Intelligence and Learning, Causal AI, or?Trans-AI.
Reality =the World > Causality = Interaction > Real Science and Technology > Causal/Interactive AI = Real Machine Intelligence and Learning >Man-Machine Superintelligence: Trans-AI augmenting the human world
Real/Causal AI Science and Technology disproves the potentially apocalyptic perspective of AI, pictured, narrated and described by many techno-dystopian movies, sci-fi literature and books as?Our Final Invention: Artificial Intelligence and the End of the Human Era.?
Reinforcement Learning as the Causal Learning of the agent-environment interactions
If you search 'causal learning", one could find out how poorly studied it. Here is the first batch of search results.
Human causal learning consists of?the different methods by which people learn the causal structure of the world. There are at least two different types of causal learning: causal perception and causal inference. Causal perception is, as the name suggests, the relatively direct perception of causal relations (e.g., one object “pushing” another). In contrast, causal inference involves learning causal relations from multiple cases or instances, when we cannot directly determine causal relations from any particular case (e.g., learning whether aspirin relieves headaches). Causal learning and reasoning play a critical role in people’s ability to make appropriate decisions, perform suitable actions, and understand the structure of their world.
Causal learning is?the process by which people and animals gradually learn to predict the most probable effect for a given cause and to attribute the most probable cause for the events in their environment. Learning causal relationships between the events in our environment and between our own behavior and those events is critical for survival. From learning what causes fire (so that we could either produce or prevent the occurrence of fire at will) to learning what causes rain, what causes cancer, or what caused that particular silly accident that we had with the car a few days ago, both the history of humankind and our individual history are full of examples in which causal learning is crucial.
Reinforcement Learning (RL) is the science of decision making. It is about learning the optimal behavior in an environment to obtain maximum reward. This optimal behavior is learned through interactions with the environment and observations of how it responds, similar to children exploring the world around them and learning the actions that help them achieve a goal.
领英推荐
In the absence of a supervisor, the learner must independently discover the sequence of actions that maximize the reward. This discovery process is akin to a trial-and-error search.?
The main elements of an RL system, where the agent interacts with the environment to sample trajectories of states and rewards, are:
Any real-world problem where an agent must interact with an uncertain dynamic environment to meet a specific goal is a potential application of RL, from robotics to autonomous driving.
There could be developed a general framework called interactive/causal reinforcement learning (ICRL), which?combines structural invariances of causality with reinforcement learning.
Causal Learning + Reinforcement Learning = Interactive/Causal Reinforcement Learning
In the behavioral sciences, reinforcement refers to the reward system and associative learning process, when a person or animal or plant or machine learns a causal association between two stimuli or events or actions or information.
A causal reinforcement (by reward) or a punishment is given after a given behavior, changing the frequency and/or form of that behavior.
"positive reinforcement" refers to the addition of a pleasant factor, reinforcer
"negative reinforcement" refers to the removal or withholding of an unpleasant factor
"positive punishment" refers to the addition of an unpleasant factor
"negative punishment" refers to the removal or withholding of a pleasant factor.
Causal reinforcement learning in the brain to predict rewards based on environmental cues is considered essential for animal/human survival. The mesolimbic dopamine system signaling the reward prediction errors, the outcome deviation from expectations, acting as a key controller of learning. Animals can infer predictions by associative learning the retrospective and prospective cause of rewards, or retrospective and prospective causal learning.
In cybernetic and control theory, it is applied as positive or negative feedback control causal loops, where the RSP-CPV error can be used to return a system to its norm.
This controller monitors an essential variable, or the controlled process variable (CPV) of a control system, and compares it with the reference set point (RSP). The difference between actual and desired value of the process variable, called the?error?signal, or RSP-CPV error. It is applied as an error-control regulation feedback to generate a control action to bring the CPV to the same value as the RSP.
Causal RL in Machine Learning
The same model could be applied in the ML/DL/ANNs pipeline, with input data, transfer/system/network function, learnt algorithms, output data and error-correcting backprop algorithms.
Again, reinforcement learning is a machine learning training method based on rewarding desired behaviors and/or punishing undesired ones, where developers devise a method of rewarding desired behaviors and punishing negative behaviors.
Such a reinforcement learning agent is able to perceive and interpret its environment, take actions and learn through trial and error.
But it is difficult to consistently take the best actions in a real-world environment, say, to navigate a complex physical environment, because of constant environment changes.
There are reinforcement learning algorithms varying due to their strategies for exploring their environments.
A neural network is a set of algorithms designed to recognize causal data patterns.
Use cases of CRL include, but are not limited to, the following:
All in all, CRL could play a key role in operations research, information theory, game theory, control theory, simulation-based optimization, multiagent systems, swarm intelligence, statistics and genetic algorithms.
RS & T: Causal Intelligence, Knowledge and Learning
The RST is largely about what is largely ignored by the Orthodox Science and Technology. It is the real-world of causal interactions in all its dynamic complexities, with its causal modeling, intelligence, knowledge, learning and applications, physical and digital, machines and mechanisms, hardware or software.
If one will search the keywords of "causal intelligence", "causal knowledge", and "causal learning", one could see how big a gap of understanding the nature of causality, causation and interaction.
They identify cause-and-effect relationships not as the complex interactions of multiple variables, as both causes and effects, which are various phenomena, as changes and processes. It is rather as a linear, one-way temporal relationship between two events or two variables, random or deterministic, with divided regularity, probabilistic, counterfactual, mechanistic, and manipulationist views. Such poor understanding of the cause-and-effect relationships is a common thing for the all orthodox science and technology. Causality has been inferred using A/B tests and look-alike analysis between test/treatment and control groups, failing with the greater number or complexity of interventions and interactions.?
It is still a linear causal modeling, reasoning and inference. It is the same common frameworks for causal inference, like as the?causal pie model?(component-cause),?Pearl's structural causal model?(causal diagram?+?do-calculus),?structural equation modeling, and?Rubin causal model?(potential-outcome), misapplied in social sciences and epidemiology. As the separate types of such causal models go: common-cause relationships, common-effect relationships, causal chains and causal?homeostasis (when causal relationships form a stable cycle or self-reinforcing mechanism).
The best what it has is one of the most popular Causal Graphical models, a Bayesian Network of connections that reveal the effect of each variable on others, as how various factors impact sales.
And it is stilled widely confused with spurious correlations, “the kryptonite” of Wall Street’s AI rush.
Without Causal Intelligence, Causal Knowledge and Causal Learning modeling the real-world interactions of all forms and kinds, no Science and Technology is real, true and genuine.
Again, there is no real AI models or ML algorithms without causal intelligence and deep learning and interactive understanding.
Causal intelligence and learning is the key in building real AI/ML models, in AI/ML solutions to automate processes and decision-making, in boosting AI/ML performance in business.
More and more businesses declare about introducing the conceptually defective causal inference into AI-based systems for decision-making, predicting things or human's behavior, with no causal effects. Among them are the big tech players, as Microsoft, Meta, Amazon, Google, as well as start-ups, as CausaLens, Causalis, Oogway.
A public health agency from one of the world’s biggest economies (causaLens cannot disclose publicly which one) used its causal AI engine to determine why certain adults have been holding back from getting COVID-19 vaccinations, so that the agency could devise better strategies to get them on board.
It is self-proclaimed as the pioneer of?Causal AI?— "a new category of intelligent machines that reason about the world the way humans do, through cause-and-effect relationships and with imagination".
Believe it or not, but this is the level of leveraging Causal AI in medicine, healthcare, marketing, retail, e-commerce, financial services, banking, manufacturing, transportation, logistics, etc.
There is a rare non-orthodox AI book in line with the RST views:
Here is a passage: "Ever since Alan Turing, AI enthusiasts have equated artificial intelligence with human intelligence. This is a profound mistake. AI works on inductive reasoning, crunching data sets to predict outcomes. But humans don’t correlate data sets: we make conjectures informed by context and experience. Human intelligence is a web of best guesses, given what we know about the world. We haven’t a clue how to program this kind of intuitive reasoning, known as abduction. Yet it is the heart of common sense".
My philosophy?of machine intelligence and learning is plain and clear.
It is all the world of causal interactions. with its machine ontology, real science and technology, mathematics and statistics and probability, programming and computing.
That's it. No human-like perceptions, concepts, ideas, and thoughts, cognition, reasoning and understanding, intellect, intelligence, sentience and consciousness, decision-making, willing, action and interaction.
It is all numbers, variables and values, quantities and data or statistics, measurements scales (nominal, ordinal, interval, ratio and cardinal), digital/binary numbers of zeroes and ones, with all its possible types and causal patterns and operations.
It tokenizes and digitizes and processes all data [as to ASCII characters], thus creating [non-human] intelligent computing systems, from self-driving cars to large natural language models.
The faster we comprehend that machine intelligence and learning is fundamentally complementary to human intelligence, the better for all of us.
Resources
SUPPLEMENT: RST as the product of Transdisciplinary R & D
Transdisciplinary research (TDR) first emerged at the OECD International conference on Interdisciplinary Research and Education in 1970 to integrate both academic researchers from unrelated disciplines – including natural sciences and social science and humanities - and non-academic participants to achieve a common goal, involving the creation of new knowledge and theory.
I have underlined a critical role of transdisciplinary R & TD in all science and emerging technologies, like as machine intelligence and learning (MIL), widely commercialized as artificial human-like intelligence (AI).
The key message is that only transdisciplinary science and technology is real S & T. The criteria of real S & T are defined as reality and causality, interaction and interactivity, knowledge and discovery, universality and locality, methodology and validity, understanding and application.
Transdisciplinarity is largely ignored by the established science and engineering institutions regardless that the fusion of Science and Engineering is crucial for Complex Problem Solutions in Science and Technology, Society and Economy, Politics and Culture, Environment and Nature.
There is Transdisciplinary Journal of Engineering & Science, the Division of Transdisciplinary Sciences, Collaborative educational courses of "Kanazawa University" and "Japan Advanced Institute of Science and Technology, The International Center for Transdisciplinary Research (CIRET), and International Centre for Transdisciplinary Studies and Research.
Transdisciplinary research (TDR), which involves the integration of knowledge from different science disciplines and (non-academic) stakeholder communities, is required to help address complex societal challenges. However, and despite increasing interest at the policy level, there are significant barriers to conducting rigorous TDR. [OECD Global Science Forum (GSF): Addressing societal challenges using transdisciplinary research]?
I advance the necessity of Global Center of Transdisciplinary Studies, Research, and Development in Science, Arts, and Technology, Society and Economy, Politics and Culture, Environment and Nature.
The idea is to see in each country the National Center of Transdisciplinarity, as the key nodes of the global network of transdisciplinarity, like as the European Center of Transdisciplinarity or the Russian Institute of Transdisciplinary Science and Technology. All is supported by the Global Trans-AI Platform operating the Global Knowledge Platform, as the transdisciplinary unification of philosophy, science, arts and technology
It is decisive in solving complex national problems and obtaining the UN Sustainable Development Goals.