On the Global AI Race...

Artificial intelligence (AI) promises to have the disruptive effects on the human world and its global politics and economy much exceeding electricity, ICT and the internet and control systems (robotics and automation) taken together. This has led governments and big corporations to compete in the global AI arms race, forever revolutionizing our world.

Because leadership in AI offers advantages both in economic competitiveness and military prowess, great powers are racing to develop advanced AI systems.

Been underway for years, the global AI race has been exacerbated by the AI Cold War between the US and China and the big tech generative AI overhyping, as LLMs, ChatGPT and Google's Gemini.

So, a final race to human extinction has begun. For “AGI is inevitable, so the United States, China, Russia or Europe should be first.”

"And, if the world cannot manage the current race to superhuman artificial intelligence between great powers, everything may die much sooner than expected". [A Race to Extinction: How Great Power Competition Is Making Artificial Intelligence Existentially Dangerous]

Now, what is the real state of affairs with AI, and what is real and true AI/AGI/ASI, including its features or characteristics and development architecture?

Real AI vs. Fake AI

The goal of mainstream AI is to implement, simulate, replicate human intelligence in machines i.e., invent AI systems that "understand, think, learn, and behave" like humans. [Why Machines, AI Agents, Neural Nets and LLMs can not Think or Reason or Learn like Humans]

Today's Narrow/Weak AI is an imitation, simulated AI, being really automation software, advanced predictive analytics, and statistic inductive machines.

Such an AI is increasingly found as a big intellectual confusion, as its founders intentionally ignored a key ingredient, Wiener's Cybernetics, 1948.

From the very beginning, it must be a real, cyber-physical AI, as in:

Cyber-Physical AI = Techno-Intelligence = Cyber-physical Intelligence = Cybernetic Intellect = Data-Knowledge-Intelligence Machines = Cybernetics (Web/Internet) AI + Symbolic AI + Machine learning AI + Generative AI + LLMs + Human Intelligence +...

https://futurium.ec.europa.eu/en/european-ai-alliance/posts/if-any-real-artificial-intelligence-reality-and-who-over-profits-ai-mythology

Now the Cyber-AI is fast changing the world: from audio/speech/face recognition to fighting fraud to voice recognition to online recommendations to emerging cyborgs of all sorts and kinds, as autonomous transportation systems or enhanced humans.

Here are some examples of cyber-physical AI systems. Software you can chat with. Cars that drive themselves. Phones that recognize human faces. Advanced supercomputing technologies which is powered by Intel? Xeon? Scalable processors and NVIDIA Tesla GPU accelerators or the big tech generative AI, as LLMs, ChatGPT and Google's Gemini.

What is Real AI with its Benchmark Characteristics??

Five main characteristics of?AI/AGI/ASI/TAI are as follows.?

• World Knowledge and Intelligence:?It?will have access to massive amounts of world knowledge/data/information with the background knowledge on every domain, subject, theme and topic, problem and tasks.
• Common Sense Knowledge and Reasoning: to develop common sense, from common sense physics to common sense psychology, to help AGI make intelligent decisions accordingly.?
• Transfer learning:?It?will be able to transfer the knowledge and skills of one task to other similar tasks.
• Abstract Thinking/Conceptual Reasoning/Categorical Inference: It is able to understand and break down all the major ontological categories, abstract concepts, ideas, and thoughts.
• Communicating in Computing Language or Natural Language, including NLP, NLG, NLU algorithms, as LLMs
• Interaction/Cause and Effect:?It?is able to understand interactions of any scales, levels and complexities using cause and effect relationships to effectively interact with the world and its changing environments, situations and circumstances.

Compare it with the OpenAI's ChatGPT or Gemini, Google's largest and most capable AI model, "built from the ground up for multimodality — reasoning seamlessly across text, images, video, audio, and code". It is promoted as "the first model to outperform human experts on MMLU (Massive Multitask Language Understanding), one of the most popular methods to test the knowledge and problem solving abilities of AI models".

Human AI is artificially divided into: narrow, general and superhuman AI, where

Narrow AI is programmed to perform a specific task, be it generative AI, machine learning algorithms or deep neural networks, applied as self-driving cars, robotics, or voice assistants, such as Siri, Alexa, and Google Assistant or LLMs, as ChatGPT or Gemini.

General AI, which is reaching the level of human intellect, as AGI systems, would be able to reason and think like a human, having common sense and understanding any intellectual task, think abstractly, learn from its experiences, and use that knowledge to solve new problems.

Super AI, a supercomputing system which human-mimicking intelligence surpasses all forms of human intelligence, in all aspects, outperforming humans in every function, thus disrupting humankind.

In reality, AI's true goal is not human intelligence, but reality, its truths and facts, models and simulations, all to effectively and sustainably interact with the world, learning and inferring, predicting and explaining all the predictable and unpredictable, as known and unknown outcomes, anticipated and unanticipated consequences, expected or unexpected behaviors.

As it was initially defined, AI is “the science and engineering of making intelligent machines” [John McCarthy, 1955].

Or, AI is the intelligence of?machines?or?software, as opposed to the intelligence of humans or animals. It is a?field of study?in?computer science?which develops and studies intelligent machines. Such machines may be called AIs. No more, no less.

Real AI Integration System Architecture

Any development and commercialization of digital and emergent technologies is liable to the Law of Unintended or Unanticipated Consequences.

It leads to unexpected outcomes that may be positive, negative, or perverse, with historical evidence suggesting that these consequences are quite often disadvantageous, counterproductive, detrimental, fraudulent, and often dangerous to individuals and communities.

Real AI Technology is a general-purpose, system integration technology integrating discrete AI models and disparate ML systems, including the latest techno-scientific breakthroughs and most advanced emerging technologies:

Cyber-Physical AI = World Knowledge/Intelligence Models [World Model+ Foundation/Ontological Model + Scientific Models + Mathematical Models + Data Models + Language Models + Information Architecture + Global Knowledge Graph] +

RAI [Learning. Inference, Interaction] Engine +

Web Data + Web Intelligence +

Extended Reality [Augmented Reality (AR), Virtual Reality (VR), Mixed Reality (MR), and everything in between] + the Metaverse technology+

Unreal AI + ML + DL +

LLMs + Generative AI +

Causal/Explainable/Responsible/Trustworthy AI/ML/DL +

Interactive AI/ML/DL +

Cognitive Robotics + Intelligent Automation + Quantum Supercomputing +

Intelligent Cyber-Physical Systems + Internet of Things +

Emergent Technologies + Digital Intelligence Technologies + Climate Change Technologies + Space Technologies +

Intelligent Communities, Cities and Countries +

Hyper-Intelligent Hyper-Automation = the Intelligent Internet of Everything >

Global Digital Intelligence + Global Community Knowledge and Intelligence =

Global Man-Machine Hyperintelligence = Global Brain/Mind/Intellect Man-Machine Networks >

5I-World (Intelligent, Innovative, Interconnected, Instrumented, Inclusive)

Who is to Win Real AI Race?

Many nations are racing to achieve a global advantage in the AI foundational technology to secure their future, progress, growth and competitiveness, productivity, national security, solving societal challenges.

They could be compared in terms of their relative standing in the AI economy by examining some key metrics and scores—talent, research, development, adoption, data, and hardware.

Or, it is ranking of countries based on; investment, innovation and implementation; talent, infrastructure, operating environment, R & D, government strategy and commercial.

Pillars

The Global AI Index is underpinned by 111 indicators, collected from 28 different public and private data sources, and 62 governments. These are split across seven sub-pillars: Talent, Infrastructure, Operating Environment, Research, Development, Government Strategy and Commercial.

Implementation

Talent?focuses on the availability of skilled practitioners in artificial intelligence solutions. Infrastructure?assesses the reliability and scale of access infrastructure, from electricity and internet to supercomputing capabilities.

Operating Environment focuses on the regulatory context and public opinion on artificial intelligence.

Innovation?

Research looks at the extent of specialist research and researchers, including numbers of publications and citations in credible academic journals.

Development focuses on the development of fundamental platforms and algorithms upon which innovative artificial intelligence projects rely.

Investment

Government Strategy gauges the depth of commitment from national governments to artificial intelligence; investigating spending commitments and national strategies.

Commercial focuses on the level of startup activity, investment and business initiatives based on artificial intelligence.

It is not the numbers of human AI start-ups or supercomputers or researchers or papers or R & D funding, what decides the global AI race, where only few winners and many losers.

It is about the whole new subject of AI dealing with reality, reality and facts, instead of the subjective human brain and brains, statistics and probabilities, deepfakes and disinformation, biases and prejudices, training data sets and hugely unsustainable computing with a catastrophic impact on the environment.

All stakeholders and organizational parties have to strategically consider how they will eliminate or minimize negative outcomes emerging from the global human AI race.?

"AI may be the technology that shapes this century. While AI capabilities are advancing rapidly, progress in safety and governance is lagging behind. To steer AI toward positive outcomes and away from catastrophe, we need to reorient. There is a responsible path, if we have the wisdom to take it".

Resources

AI Superintelligence/AI/AGI/ASI/Trans-AI

Trans-AI: How to Build True AI or Real Machine Intelligence and Learning

Real AI Project Confidential Report: How to Engineer Man-Machine Superintelligence 2025: AI for Everything and Everyone (AI4EE); 179 pages, EIS LTD, EU, Russia, 2021 (see Supplement).

World Knowledge and Intelligence Platform = Interactive Human-AI Hyperintelligence = AI/ML/DL/AGI/ASI + LLMs + DTs + Robots + 4IR Networks + SCs

Future AI: Real AI & ML & DL: Machine Ontology + Science + AI + ML + LLMs +...

Exposing the great unknown unknowns: the World of Reality. Part I.

Exposing the great unknown unknowns: the World of Causality. Part II.

[Mind, Machine Intelligence, AI, Artificial, Automated, Autonomous, Automatic Intelligence, Machine Learning, Deep Learning]: exposing the great unknown unknowns. Part III.

Real AI vs. Unreal AI: Intelligence is Power, Real Intelligence is Superpower

The Post-Information Age of Real AI Technology

Real AI vs. Gen AI and LLMs: post-information technologies vs. post-modern technologies

SuperIntelligence [Mind, Machine Intelligence, AI, Artificial, Automated, Autonomous, Automatic Intelligence, Machine Learning, Deep Learning]: exposing the great unknown unknowns. Part III.

Large Language Model Landscape

GAIA: A Benchmark for General AI Assistants

MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI

Gemini: A Family of Highly Capable Multimodal Models

SUPPLEMENT 1: LLM/ML as ML Rote Learning: Memorization vs Generalization vs Generative

Large Language Models (LLMs) are mere memorization devices, their extensive training on vast datasets leaves little room for genuine creativity.

We’ll explore the sophisticated mathematical frameworks that differentiate LLMs from traditional memorization models, diving into the intricate equations that drive their advanced functionalities.

Memorization in machine learning refers to a model’s ability to precisely recall specific data points or patterns it has seen during training. It’s akin to storing and retrieving exact instances from the training data.

Generalization is the ability of a model to apply the knowledge gained during training to new, unseen data. It’s a measure of how well the model has learned the underlying patterns or principles in the data, rather than just memorizing specific instances.

“Generative” refers to the ability of a model to generate new data instances that resemble the training data. This doesn’t just mean replicating or slightly modifying existing instances, but creating entirely new samples that are consistent with the learned data distribution.

A larger corpus alone isn’t enough for effective LLM training; what’s crucial is the variety in linguistic features it contains, from Lexicon, Syntax and Semantics to Genre and Format Varieties.?

• L?represents a vector of linguistic features for a given data point (e.g., a sentence, a paragraph, or a document). Each element in this vector could represent different aspects of language, such as semantic features, syntactic structures, or lexical choices.
• μ?is the mean vector, representing the average feature vector across the entire dataset. This can be thought of as the “central” linguistic pattern in the data.
• Σ is the covariance matrix of linguistic features, representing how these features vary together across the dataset.
• M?is the number of dimensions in the feature space, equivalent to the number of linguistic features being considered.
• ∣Σ∣ is the determinant of the covariance matrix Σ, which in this context reflects the diversity and complexity of linguistic patterns.
• The term (L?μ)^T?Σ^?1 (L?μ) measures the Mahalanobis distance from the average linguistic pattern, capturing how typical or atypical a specific set of linguistic features is within the context of the dataset.

A larger corpus alone isn’t enough for effective LLM training; what’s crucial is the variety in linguistic features it contains. While a vast dataset contributes to better model regularization and generalization, the diversity of language structures and content within it is key. This ensures that LLMs are exposed to a broad range of linguistic phenomena, which is essential for their ability to generalize effectively.

Following are the diversity in linguistic features that are needed:

Lexical Features:

• Vocabulary Range: Incorporation of a broad array of words, including common, rare, archaic, and neologisms.
• Synonyms and Antonyms: Variations in word choice expressing similar or opposite meanings.
• Collocations: Commonly paired words or phrases (e.g., “heavy rain”).
• Idioms and Fixed Expressions: Phrases with meanings not deducible from individual words (e.g., “kick the bucket”).

Syntactic Features:

• Sentence Structures: Simple, compound, complex, and compound-complex sentences.
• Grammar Variations: Diverse grammatical constructions, including active/passive voice, different tense aspects, and moods.
• Syntactic Ambiguity: Sentences with multiple interpretations based on structure (e.g., “I saw the man with the telescope”).
• Clause Usage: Different types and arrangements of dependent and independent clauses.

Semantic Features:

• Contextual Meaning: Words or phrases with meanings that change based on context.
• Polysemy and Homonymy: Words with multiple meanings and words that sound alike but have different meanings.
• Figurative Language: Metaphors, similes, and allegories.
• Pragmatic Markers: Words or phrases that serve conversational functions (e.g., “well,” “you know,” “actually”).

Morphological Features:

• Word Formation: Use of prefixes, suffixes, inflections, and compounding.
• Irregular Forms: Irregular verbs, nouns, and adjectives.
• Morphological Variations: Different forms of a word (e.g., singular/plural, gender forms).

Phonological and Phonemic Features:

• Phoneme Variations: Different sounds and their linguistic functions.
• Stress and Intonation Patterns: Variations in word and sentence stress.
• Rhythmic Structures: Patterns of sound in language.

Discourse Features:

• Coherence and Cohesion: Logical flow of ideas and use of cohesive devices (e.g., conjunctions, reference words).
• Narrative Structures: Various storytelling elements and structures.
• Argumentation and Persuasion Techniques: Structures used in persuasive and argumentative language.
• Dialogue and Conversational Structures: Features of spoken language and dialogue.

Sociolinguistic Features:

• Dialectal Variations: Different dialects and sociolects of a language.
• Socio-Cultural References: Language reflecting cultural and societal norms.
• Register and Style: Formal, informal, academic, colloquial, and domain-specific language styles.
• Language Varieties: Different languages or language families for multilingual models.

Pragmatic Features:

• Speech Acts: Language functions like requesting, apologizing, instructing.
• Implicature and Inference: Understanding implied meanings and making inferences.
• Politeness and Social Conventions: Language reflecting social norms and politeness.

Genre and Format Varieties:

• Literary and Non-Literary Texts: Fiction, poetry, essays, and non-fiction materials.
• Technical and Scientific Language: Jargon and specialized language of various fields.
• Media and Journalism: Language styles found in news articles, broadcasts, and social media.
• Legal and Administrative Language: Language specific to legal documents, contracts, and official communications.

Incorporating these diverse linguistic features into the training corpus of an LLM ensures a comprehensive understanding of language nuances and complexities, significantly aiding in the model’s ability to generalize effectively across different contexts and use cases.

SUPPLEMENT 2: Managing AI Risks in an Era of Rapid Progress

Abstract

In this short consensus paper, we outline risks from upcoming, advanced AI systems. We examine large-scale social harms and malicious uses, as well as an irreversible loss of human control over autonomous AI systems. In light of rapid and continuing AI progress, we propose urgent priorities for AI R&D and governance.

SUPPLEMENT 3: Artificial Intelligence Act: deal on comprehensive rules for trustworthy AI

Safeguards agreed on general purpose artificial intelligence

Limitation for the of use biometric identification systems by law enforcement

Bans on social scoring and AI used to manipulate or exploit user vulnerabilities

Right of consumers to launch complaints and receive meaningful explanations

Fines ranging from 35 million euro or 7% of global turnover to 7.5 million or 1.5% of turnover

MEPs reached a political deal with the Council on a bill to ensure AI in Europe is safe, respects fundamental rights and democracy, while businesses can thrive and expand.

On Friday, Parliament and Council negotiators reached a provisional agreement on the Artificial Intelligence Act. This regulation aims to ensure that fundamental rights, democracy, the rule of law and environmental sustainability are protected from high risk AI, while boosting innovation and making Europe a leader in the field. The rules establish obligations for AI based on its potential risks and level of impact.

Banned applications

Recognising the potential threat to citizens’ rights and democracy posed by certain applications of AI, the co-legislators agreed to prohibit:

• biometric categorisation systems that use sensitive characteristics (e.g. political, religious, philosophical beliefs, sexual orientation, race);
• untargeted scraping of facial images from the internet or CCTV footage to create facial recognition databases;
• emotion recognition in the workplace and educational institutions;
• social scoring based on social behaviour or personal characteristics;
• AI systems that manipulate human behaviour to circumvent their free will;
• AI used to exploit the vulnerabilities of people (due to their age, disability, social or economic situation).

.......................................................................................................................................

Guardrails for general artificial intelligence systems

To account for the wide range of tasks AI systems can accomplish and the quick expansion of its capabilities, it was agreed that general-purpose AI (GPAI) systems, and the GPAI models they are based on, will have to adhere to transparency requirements as initially proposed by Parliament. These include drawing up technical documentation, complying with EU copyright law and disseminating detailed summaries about the content used for training.

For high-impact GPAI models with systemic risk, Parliament negotiators managed to secure more stringent obligations. If these models meet certain criteria they will have to conduct model evaluations, assess and mitigate systemic risks, conduct adversarial testing, report to the Commission on serious incidents, ensure cybersecurity and report on their energy efficiency. MEPs also insisted that, until harmonised EU standards are published, GPAIs with systemic risk may rely on codes of practice to comply with the regulation.

https://www.europarl.europa.eu/news/en/press-room/20231206IPR15699/artificial-intelligence-act-deal-on-comprehensive-rules-for-trustworthy-ai

A small thing left, developing a generalist AI.