Developing, Engineering and Deploying Real AI Superintelligence Platform
"Humans, who are limited by slow biological evolution, couldn't compete and would be superseded". Stephen Hawking
“As AI gets probably much smarter than humans, the relative intelligence ratio is probably similar to that between a person and a cat, maybe bigger.” Elon Musk
The big tech ML/DL/AI labs, as Microsoft’s OpenAI or Meta’s AI or Tesla, are striving for a HLAI (human-level AI), like as the prospective Tesla Optimus humanoid AI robot. We show why these projects are unsustainable and how to engineer the first GAI robotic platform with real superintelligence (RSI).
Artificial Intelligence will take a couple of years, not decades, to reach the general artificial intelligence stage and real superintelligence stage, meeting the requirements of?general AI:
Introduction
In less than a decade, computers had "trained" how to identify objects, diagnose diseases, translate languages, drive cars, or transcribe speech, outplaying humans at complicated strategy games, create photorealistic images, etc.
Yet despite these impressive achievements, AI has shocking weaknesses, being easily fooled, duped or confounded by situations they haven’t seen before, to be trained to carry out only one task, suffering from “catastrophic forgetting.”
These shortcomings have something in common: they exist because the mainstream AI/ML/DL systems don’t understand the world of realities, physical or social, digital or virtual, its various interactions and causation.
Today's AI/machine learning systems are constrained in their capacities, driven by big data processing and pattern recognition and statistical correlations, instead of processing the world's data, inferring causation and learning from small examples/samples, as the child could do.
AI won’t be any human smart or hyperintelligent if computers don’t grasp interactions, their causes and effects and relational entities. That’s something even most humans have trouble with.
So, such an AI has zero chances to be developed as AGI or ASI due its nonscientific subjective assumptions to mimic or replicate human brains/mind/cognition/intelligence/behavior.
All in all, all five ways to achieve AI Superintelligence has zero prospects, namely:
Artificial General Intelligence (AGI) mimicking human intelligence,
Brain implants, ‘updating’ human brains with Neuralink chips,
Whole brain emulation or mind uploading,
Biological enhancements (genetic technologies),
Human-machine superminds (creating a superintelligence interlinking together human intelligences by neuro-implants).
The Tesla Optimus project, blindly following one of the ways, AGI, should properly correct its conception, model and business strategy as the Tesla Optimus Prime relying on the RSI (Real Superintelligence) model, as described in
Tesla Optimus with Human-Like Intelligence or Tesla Optimus Prime with Real Superintelligence
Elon Musk gave a timeline to production for the first time for the Tesla Optimus project, a humanoid robot capable of doing general tasks.
At the?Tesla Cyber Rodeo event ?today, Musk offered a production timeline for the first time
I think we have a shot at being in production for version 1 of Optimus hopefully next year.
Musk added about Tesla Optimus during the event:
It will upend our idea of what the economy is… it will be able to do basically anything humans don’t want to do. It will do it. It’s going to bring an age of abundance. It may be hard to imagine it, but as you see Optimus develop, and we will make sure it’s safe, no Terminator stuff, it will transform the world to a degree even greater than the cars.
Optimus comes from "Optimus Prime", a "Cybertronian, a fictional extraterrestrial species of sentient self-configuring modular robotic lifeforms (e.g.: cars and other objects), a synergistic blend of biological evolution and technological engineering"
Building a humanoid robot prototype with real superintelligence
Real, True or Genuine AI is not about computing tools or computational devices as GPUs, field-programmable gate arrays (FPGAs), and application-specific integrated circuits (ASICs), designed for either training or inference ONLY for special tasks, or data analytics techniques, as datasets, statistical learning, neural networks, ML algorithms or language models.
True [Human-Machine] Intelligence is emerging as a Transdisciplinary AI (Trans-AI) or Meta-AI or?the Man-Machine Supermind or Hyperintelligence, integrating symbolic AI and Neural ML, be it Artificial Narrow Intelligence, Artificial General Intelligence, Artificial Superintelligence, with Collective Human Intelligence.
Trans-AI is to be designed, developed and distributed as Man-Machine Hyper/Super/Meta/Trans-Intelligence Platform.
Trans-AI: How to Build True AI or Real Machine Intelligence and Learning
Encoding a 'unified world model' of how the world works, Trans AI is to organize the world’s data, information, knowledge and intelligence and make it all universally accessible and valuable.?
We are at the edge of colossal changes. This is a critical moment of historical choice and opportunity. It could be the best 5 years ahead of us that we have ever had in human history or one of the worst, because we have all the power, technology and knowledge to create the most fundamental general-purpose technology (GPT), which could completely upend the whole human history. The most important GPTs were fire, the wheel, language, writing, the printing press, the steam engine, electric power, information and telecommunications technology, all to be topped by real artificial intelligence technology. Our study refers to Why and How the Real Machine Intelligence or True AI or Real Superintelligence (RSI) could be designed and developed, deployed and distributed in the next 5 years. The whole idea of RSI took about three decades in three phases. The first conceptual model of TransAI was published in 1989. It covered all possible physical phenomena, effects and processes. The more extended model of Real AI was developed in 1999. A complete theory of superintelligence, with its reality model, global knowledge base, NL programing language, and master algorithm, was presented in 2008.
The RSI project has been finally completed in 2020, with some key findings and discoveries being published on the EU AI Alliance/Futurium site in 20+ articles. The RSI features a unifying World Metamodel (Global Ontology), with a General Intelligence Framework (Master Algorithm), Standard Data Type Hierarchy, NL Programming Language, to effectively interact with the world by intelligent processing of its data, from the web data to the real-world data. The basic results with technical specifications, classifications, formulas, algorithms, designs and patterns, were kept as a trade secret and documented as the Corporate Confidential Report: How to Engineer Man-Machine Superintelligence 2025.
As a member of EU AI Alliance, the author has proposed the Man-Machine RSI Platform as a key part of Transnational EU-Russia Project. To shape a smart and sustainable future, the world should invest into the RSI Science and Technology, for the Trans-AI paradigm is the way to an inclusive, instrumented, interconnected and intelligent world.
Trans-AI or Meta-AI running a Unified World Model Engine undergrids all the human-like AI models mimicking human brains/intelligence/cognition/mind/behavior, as Meta AI + Google AI + Transformers NNs+..., as briefed in the resource.
Tesla Optimus Prime as Real AI/Trans-AI/Meta-AI vs. Fake AI/ML/DL/ANNs
Technological Mind or Machine Intellect or Computing Intelligence, widely known as Artificial intelligence (AI), be it symbolic AI, neural AI or neuro-symbolic AI or strong, full and general AI, is largely a fictional construct of scientific imagination, being as imaginary as the sci-fi AI from Hollywood studios.
This is well-supported by the unreality of the so-called symbolic AI as well as the sub-symbolic ML, as large NLP ML models, Google's MUM or OpenAI's GPT-3, being numb and dumb or unintelligent by the very design.
AI has been sold for the last 70+ years as an umbrella term representing a wide range of techniques, programs, algorithms and models that would allow computing machinery to mimic, simulate, replicate or fake human body/brains/mind/intelligence/behavior.
As any big fictive construction, it has its hype cycles, boom and bust dynamics, winter and spring seasons, with its believers and unbelievers.
In reality, we need a truly intelligent AI powerful to model, understand and interact with any reality, physical, social or virtual, to reason about the world, while discovering and learning new things, [continually] expanding its knowledge and intelligence.
The AI Superintelligence Hardware
Let’s unravel the mystery and complexity of processors and so-called AI accelerators, as from Google, Nvidia, etc.
It is just simple data matrix-multiplication accelerators. All the rest is commercial propaganda.
Unlike other computational devices that treat scalar or vectors as primitives, Google’s Tensor Process Unit (TPU) ASIC treats matrices as primitives. The TPU is designed to perform matrix multiplication at a massive scale.
Here’s a diagram of Google’s TPU:
At its core, you find something that inspired by the heart and not the brain. It’s called a “Systolic Array” described in 1982 in “Why Systolic Architectures ?”:?https://www.eecs.harvard.edu/~htk/publication/1982-kung-why-systolic-architecture.pdf
And this computational device contains 256 x 256 8bit multiply-add computational units. A grand total of 65,536 processors is capable of 92 trillion operations per second.
It uses DDR3 with only 30GB/s to memory. Contrast that to the a Nvidia Titan X with GDDR5X hitting transfer speeds of 480GB/s.
Whatever, it has nothing to do with real AI hardware.
Some general reflections on processing units and Narrow AI coprocessors.
A central main processor is commonly defined as a digital circuit which performs operations on some external data source, usually memory or some other data stream, taking the form of a microprocessor implemented on a single metal–oxide–semiconductor integrated circuit chip (MOFSET).
It could be supplemented with a coprocessor, performing floating point arithmetic, graphics, signal processing, string processing, cryptography, or I/O interfacing with peripheral devices. Some application-specific hardware units include video cards for graphics, sound cards, graphics processing units and digital signal processors.
A central processing unit (CPU), also called a central processor, main processor or just processor, is the electronic circuitry that executes instructions comprising a computer program. The CPU performs basic arithmetic, logic, controlling, and input/output (I/O) operations specified by the instructions in the program.
Microprocessors chips with multiple CPUs are multi-core processors.
领英推荐
Array processors or vector processors have multiple processors that operate in parallel, with no unit considered central. Virtual CPUs are an abstraction of dynamical aggregated computational resources.
Microprocessors chips with multiple CPUs are multi-core processors.
We know a lot of processing units, as listed below.
Processors Taxonomy
Central Processing Unit (CPU). If designed conforming to the von Neumann architecture, containing at least a control unit (CU), arithmetic logic unit (ALU) and processor registers.
Graphics Processing Unit (GPU)
Sound chips and sound cards
Vision Processing Unit (VPU)
Tensor Processing Unit (TPU)
Neural Processing Unit (NPU)
Physics Processing Unit (PPU)
Digital Signal Processor (DSP)
Image Signal Processor (ISP)
Synergistic Processing Element or Unit (SPE or SPU) in the cell microprocessor
Field-Programmable Gate Array (FPGA)
Quantum Processing Unit (QPU)
A Graphical Processing Unit (GPU) enables you to run high-definition graphics on your computer. GPU has hundreds of cores aligned in a particular way forming a single hardware unit. It has thousands of concurrent hardware threads, utilized for data-parallel and computationally intensive portions of an algorithm. Data-parallel algorithms are well suited for such devices because the hardware can be classified as SIMT (Single Instruction Multiple Threads). GPUs outperform CPUs in terms of GFLOPS.
From Fake AI Accelerators ASIC to Real AI Accelerators
The TPU and NPU go under a Narrow/Weak AI/ML/DL accelerator class of specialized hardware accelerator or computer system designed to accelerate special AI/ML applications, including artificial neural networks and machine vision.
Big-Tech companies such as Google, Amazon, Apple, Facebook, AMD and Samsung are all designing their own AI ASICs.
Typical applications include algorithms for training and inference in computing devices, as self-driving cars, machine vision, NLP, robotics, internet of things, and other data-intensive or sensor-driven tasks. They are often manycore designs and generally focus on low-precision arithmetic, novel dataflow architectures or in-memory computing capability, with a typical NAI integrated circuit chip contains billions of MOSFET transistors.
Focus on training and inference of deep neural networks, Tensorflow uses a symbolic math library based on dataflow and differentiable programming
The former uses automatic differentiation (AD), algorithmic differentiation, computational differentiation, or auto-diff, and gradient-based optimization, working by constructing a graph containing the control flow and data structures in the program.
Again, the datastream/dataflow programming is a programming paradigm that models a program as a directed graph of the data flowing between operations, thus implementing dataflow principles and architecture.
Things revolves around static or dynamic graphs, requesting the proper programming languages, as C++, Python, R, or Julia, and ML libraries, as TensorFlow or PyTorch.
An Arms Race for Building General, Human-Level AI
The big tech ML/DL/AI labs, as Microsoft’s OpenAI or Meta’s AI or Tesla, are striving for a HLAI (human-level AI) beside of the prospective Tesla Optimus humanoid AI robot project.
At the recent Meta AI event , its chief scientist LeCun discussed possible paths toward human-level AI, challenges that remain and the impact of advances in AI, talking about its modular architecture.
The world model will be a key component of this architecture, coordinated with other modules. Among them is a perception module that receives and processes sensory information from the world. An actor module turns perceptions and predictions into actions. A short-term memory module keeps track of actions and perceptions and fills the gaps in the model’s information. A cost module helps evaluate the intrinsic — or hardwired — costs of actions as well as the task-specific value of future states.
And there’s a configurator module that adjusts all other modules based on the specific tasks that the AI system wants to perform.
LeCun believes that each one of these modules can learn their tasks in a differentiable way and communicate with each other through high-level abstractions. This is roughly similar to the brain of humans and animals, which have a modular architecture (different cortical areas, hypothalamus, basal ganglia, amygdala, brain stem, hippocampus, etc.), each of which have connections with others and their own neural structure, which gradually becomes updated with the organism’s experience
World models are at the heart of efficient learning
Now, only god knows how humans think,... and, may be, Aristotle, but he long ago is part of human history, who mentioned 256/512 logical reasoning types in his classical Analytics.
The so-called rules-based AI, as expert systems with RPA applying conditional rules, IT-TT, pretended to model human reasoning. Such a computer system is emulating the decision-making ability of a human expert, relying on two interacting modules:
the inference engine and the knowledge base.
The latter represents axioms, facts and rules. The former applies the rules to the known facts to deduce new facts, having explanation and debugging abilities.
Again, the sorts and types of human reasoning are known unknowns: logical and intuitive, deductive or inductive, pictorial or analytic, conscious or unconscious, statistic or determinate, specific or general, abstract or concrete,...or at best causal and model-based.
So to mimic human intelligence, automating thought processes, as supervised, unsupervised, reinforced or self-supervised machine learning, is hardly a rational way at all, softly speaking.
Again, the HLAI is about machines that not enhance but replace natural intelligence and perform every task that a human can, all leading to technological unemployment, singularity, runaway intelligence, and HLAI robot invasions.
Creating Real AI Platform Technology, Agents, Systems, Bots and Applications
Our humanoid GAI robotic platform, call it "Optimus Prime", is driven by the world model engine and master causal algorithms, unlike the mentioned HLAI models, as the Tesla Optimus bot, Microsoft's AGI or Meta's AGI, all relying on statistical ML algorithms.
It is supposed to embrace a robotic paradigm interrelating the three basic elements: Sensing, Planning, Acting:
As a real superintelligence (RSI), the prospective Optimus Prime humanoid AI robotic platform is integrating all possible intelligences as interdependent functional modules:
SUPPLEMENT: What does the history of AI tell us about its future?
It tells that a human-mimicking AI has no sustainable future, but only the hype cycles of boom/bust, AI springs and winters, never reaching real man-machine intelligence and learning.
The mainstream AI is a dull and dumb quasi-AI.
But I’d not classify it as utter nonsense or pure hogwash.
There are ANNs, ML and AI, weak and narrow.
ANNs are the basis of ML models and the building blocks of the Quasi-AI.
ML algorithms apply backprop and gradient descent algorithms with loss and activation functions to computationally simulate learning within neural networks.
ANNs “learn” from training datasets, as if simulating the learning identical the human brain, learning via synaptic plasticity.
It is not how humans really learn, involving experiences, mistakes, studies, reading, instructions or educational programming.
It is just applied mathematics with all sorts of optimization algorithms, where ANNs are tuned/trained by computing the errors of neural network predictions strengthening/weakening neural connections given the errors.
There is no real learning, intelligence, inferences or understanding the world here. It is mere a false or fake AI.
To be specific, here is a freshly deadly simple argument, which must be open for simpletons.
An international security conference explored how artificial intelligence (AI) technologies for drug discovery could be misused for de novo design of biochemical weapons. A thought experiment evolved into a computational proof.
We had previously designed a commercial de novo molecule generator that we called MegaSyn, which is guided by machine learning model predictions of bioactivity for the purpose of finding new therapeutic inhibitors of targets for human diseases. This generative model normally penalizes predicted toxicity and rewards predicted target activity. We simply proposed to invert this logic by using the same approach to design molecules de novo, but now guiding the model to reward both toxicity and bioactivity instead.