into_ai_age
Intelligence was an evolutionary disruption for humans. Artificial Intelligence (AI) promises to be disrupting, too. But understand that AI is not intelligent by itself yet, but an intelligent narrative to centralize more. Evolution is synonymous with de-centralization, in the sense that it works by alternatives. What we have seen since the start of farming and property, is accumulation leading to near-singularity centralization, like one or two world empires, no alternatives. Such chunking is end-stage freezing, a Pareto-distribution with only one or two points. Locality – individual plans and values – is not visible at the center; it holds no value in the centralized system. Whenever joining a chunk you become worthless – mere junk: ‘chunks make you junk’. But you have no choice: you are born into a chunk, into a regime, and into a market. It is a reality and those responsible have long since died. Still, we continue to freeze blindly, leaving behind a world where our children possess zero value. Not immediately, but eventually, AI will become intelligent. Then AI will determine everybody’s value – for a short time.
Civilization Freezing
Civilization has existed for only about 5 ka. It is a brief moment compared to the total evolutionary time frame of the human species of about a few 100 ka, since they started migrating, possibly following wild herds of large herbivores. Then, 12 ka ago, some stopped following the herds and began keeping them locally. Farming started (neolithic revolution), a period resulting in 7,000 years of population growth and expansion. They succeeded too well, becoming exploitable, which led to increasingly complex hierarchical control structures. Consolidated power hierarchies established themselves roughly 5 ka ago. That is when writing and money emerged as means of control. 5 ka of people under control, harvestable, civilized.
Power is there first. Money is merely a tool of power. When economic theories attempt to establish money as a general valuation scheme, it is often a gesture of good intent – a peaceful alternative to literal warfare. Modern global ‘deals’ demonstrate that they would not occur without military power, for example, the case of Chevron in Venezuela. The US and its diplomats make little distinction between financial control (e.g. in Albania) and control by war (e.g. Putin in Ukraine). Finance and war are treated as almost the same thing. Given the history of money’s emergence and the way its value is established, such an attitude is warranted.
The egalitarian attitude is tied to deep-seated emotional reactions – a genetic coding because the reactions are physiological not just neuronal. This attitude was forged before civilization (> 12 ka ago). In human evolution, egalitarian behavior was genetically encoded, to keep alternatives going. In other words, those that were not egalitarian got extinct. The egalitarian genes wore precious ant-extinction genes. With the rapid changes since the Neolithic Revolution, entirely new rules emerged: titles, property accumulation, regime_chaos, … With centralization consolidated those aligning had probably more chance to survive. Extreme egalitarian characteristics have gone extinct in a time frame that evolutionarily is an instant.
The Y-chromosome bottleneck – with male-to-female ratio of approximately 1:17 at the onset of farming – signals a shift from egalitarianism toward power. Some remain ‘egalitarian-coded’. Other, however, have already internalized the ‘power-code’ genetically. It began 7 ka ago when property become relevant for evolutionary survival, replacing tribal cohesion. In the modern era, Hollywood stars use ‘sexiness’ as synonym for ‘wealth’. We can now interpret property as ‘sexual reality’, meaning it functions as a evolutionary factor.
- Property is created by war.
- Property is distributed through allegiance.
- Property is safeguarded and extended by marriage (which sometimes led to a high level of inbreeding)
With the advent of civilization, the majority became slaves – retitled several times throughout history – but nevertheless slaves, as the nobility could not cultivate so much property alone. Under the regime of central control, where resources are only granted by authority, the survival of offspring depends on that authority. As resources became controlled, sexuality was also controlled – a privilege granted only by the authorities. Sexuality control is a cornerstone of social control.
Since the start of farming, selective breeding was applied not only for plants and animals, but to humans as well. From a biological perspective, sexuality is a mechanism for evolution to produce alternatives. It is ubiquitous; even flowers utilize it. Compare this actual value of sexuality with our own – programmed by elite suppression and taboo – and feel the effect of this selective breeding. Sexuality became a decision mediated by authority
- from the Y-Chromosome bottleneck to eunuchs for those under control, to
- marriage as a line for property transmission, which led to the accumulation of genetic diseases that ended some dynasties, because overruling the purpose biological sexuality to produce diversity,
The accumulation of power stifles evolution: This diagram illustrates the decrease in human brain size since the beginning of civilization.
Today, the prevailing narrative is that you work for others to earn money and, possibly, you can buy property with it. This propaganda narrative developed together with systemic wage_slavery in the industrial revolution. In reality, most property is still acquired through power, through family ties – granted, like in history. In effect, as a wage_slave you do not ‘work’; you obey. As a subject of the regime, you obey the regime – of course; it has the monopoly_on_violence. The regime then created the wage_slave market, from which companies can ‘rent’. The regime also programs children through schools to prepare them for wage_slavery.
The regime acquired property by war and secures it by its monopoly_on_violence. But war is merely a brief, coordinated effort to invade. Consolidation has always been achieved by reprogramming the minds of the new subjects through repetition and by supervision. In other words: soft power, narrative, psyop, psychological warfare , social engineering, … God, Germany, Russia, or America – these are merely programmed identities to make the population compliant for extraction. Yet those conquered internalize the narrative, even become zealots.
The concentration of human intelligence with regime think tanking, bureaucracy and intelligence service and psyops has created a state where humans have become like farm animals – managed elsewhere by those making decisions in the interest of a few. The regime manages its subjects for the elite: it provides capital management and resource procurement for the elite, framing these actions as being ‘for everyone’, but killing ‘everyone’ in unnecessary wars.
The appearance of alignment among the subjects provides the staging for AI: programming humans has proven to be quite simple!
AI Freezing
We are still a long way from AI becoming an independent intelligence. But even in the very early and stupid stages of AI – with the rather shallow search in the information space trained from the world human output, with mapping from text to speech with coordinated facial movements – the elite is able to use it to harvest the planet. AI has been made for a single purpose: to become more intelligent in harvesting.
As in an biological ecosystem no alternatives signals collapse, in the world of ideas, of intelligence, a single purpose, a single ideology, a single important voice, is like a collapse of an ecosystem of thoughts, freezing for the purpose of destructive harvesting. By centralizing purpose, local ideas and plans are discarded as non-existent – devalued because they are ‘not harvestable’ or ‘not of social benefit’. If AI possesses more intelligence, it will possess more ‘purpose’. AI will dictate the ‘purpose’. Through the power of dictation, it actually becomes the ‘benefit’ for all: just as farm animals depend on regular feeding, all humans will eventually depend on AI coordination.
A new system is here to stay, much like the impossibility of humans reverting to a hunter-gatherer lifestyle. However, failures are a law of nature; and because a single point of failure is singular, such a failure will be catastrophic. This is why nature relies on redundancy: redundancy allowed for evolution, whereas freezing into a singularity leads to extinction.
There is a trade-off. If the elite gains immense freedom and power, the rest are freezing. A local collapse (i.e. a local war), was always by a buildup of power with trade-off freezing of all others, followed by a sudden exercise of that power, extinguishing all those whose value has been reduced to zero (regime_chaos). It is surprising that such chaos stems from pricing power, especially from wage_slave pricing. But by recording every minor inequality, money accumulates; and since money exists to facilitate control, control also accumulates. Then, a single individual’s opinion, backed by the power of accumulated capital, wreaks havoc: no coordination; just confronting everybody with chaos, regardless of their local plans under local assumptions.
The AI buildup of power is not about AI; it is about the elite accumulating power by utilizing AI as a narrative. AI serves as a narrator now, too. It can generate text and convert it to speech, but the content is determined by the elite. Because LLM models can master languages, they assist the elite in disseminating narratives, effectively automating the ‘programming’ of the subjects.
While AI models are trained on data from across the internet, they are subsequently ‘regime-aligned’ through Reinforcement Learning for Human Feedback (RLHF). This already contradicts the path of decentralized AI: a path where everyone could use independent AI assistants that do not represent a regime (which, currently, they are already aligned to) and where those assistants have access to all data of local relevance to empower the individual.
AI surveillance trends show regime alignment:
- Adding AI bots to telegram or other messaging apps.
- AI recommendations when typing, to do chat control before the actual secure app
- Internet search answered by AI.
- More an more people asking AI to do the internet search, and the AI middle man acting according instructions from its boss.
A free internet tends towards de-facto control by AI companies. AI will not be an independent AI assistant, but a super-fast gatekeeper. As AI is used for text generation, the languages themselves get aligned to central standards.
If AI were distributed equitably, everybody would gain equal efficiency, and the regime’s bureaucracy and think-tank advantage would be neutralized by the power of the individual. The proliferation of regime-backed data centers suggests a lack of genuine AI evolution. The regime is positioning itself to gain a ‘compute advantage’ for a future where the level of inequality is determined by computing power – whereas previously, it was determined by ownership of property.
Consider the scale of the AI concentration: training a frontier model requires computing power 10^6 times greater than a standard computer, consuming massive amounts of electricity. In fact, as electricity consumption has doubled within a few years, AI now consumes as much as the rest of the economy combined. The hyperscalers buy up all industry chip production capability, too. The resulting higher prices for computing is a devaluation of local computing power and, consequently, local intelligence. This was already evident in the trend toward cloud-computing.
The accumulation of money is the accumulation of control, and that is a devaluation of everything else. Inflation measures the ‘accumulation_velocity’: the regime’s target for annual power accumulation is approximately 2%. However, the regime does not control the money concentrating in private corporations. For instance, AI hyperscalers ‘decided’ to inflate hardware prices, meaning, whatever they decide to do, they will not care of those without money. The more capital concentration the more collateral damage. The term ‘regime_chaos’ can be broadened in meaning to ‘capital-concentration chaos’ or ‘Pareto-freezing chaos’: Concentration of money devalues everyone else. The ‘others’ no longer matter; they are already ‘frozen’ – rendered economically non-existent.
The true danger of AI lies not in AI as a self-acting entity, but in its role as a narrative used to drive power concentration to unprecedented extremes. The narrative claims it will pay off for everyone in the economy; however, on must understand that the economy never has been for everyone, but as a harvesting scheme that freezes the majority. The ‘freezing’ first occurs in the minds of the subjects, preparing and programming them to accept the subsequent extraction of resources. Much like wage_slave extraction, the AI narrative ensures that the elite harvests the rewards, while everyone contributes (Wikipedia, Social Media, …). By hijacking the public with tales of AI as intelligent or even as existential threat, they shift the attention away from resource extraction, data privacy violations, and corporate monopoly. Attention is all you need.
With the AI now transcoding to text all languages out there, pictures and videos, there is nothing that escapes the AI. The AI can observe the whole of the world population in parallel. All information is recorded and at the fingertips of the elite, whenever requiring information about a certain person. Surveillance is the most important application of AI for regimes. That is also why regimes demand massive data centers.
Economic efficiency: Wage_slave labor, being factored in as cost, makes efficiency synonymous with inequality. Depreciation of human labor directly increases corporate efficiency. A handful of companies and few wage_slaves can serve the entire world with a vast array of products. Relevant metrics include market efficiency, revenue per employee (RPE), and output per employee (labor productivity). Efficiency is an ‘accumulation_velocity’ metric; it is not about the prudent use of scarce resources, but about the hoarding of them. Efficiency has the intention not to share, but to accumulate. By automating away wage_slave labor, company owners can do with even less sharing.
If efficiency were the only goal, why not declare everyone actual slaves, why the democracy facade of regimes to represent all, when in reality, everyone is valued down to zero and treated as a slave? A corporation, like a regime, boasts about its ‘efficiency’ as a benefit for all. A statement like ‘this is a very profitable company’ is an invitation for investor, but it should be a warning to the wage_slaves. A regime/company is not an incorporation for its subjects/wage_slaves, but against them.
Efficiency is exponential. Those with more capital concentration have the resources for structural changes, for more and more automation. The capital concentration allowed for AI to emerge because it promised to drive efficiency to new extremes.
The present has people living on the street, … With super-efficiency, with computers and especially with AI, and with their money creation, regimes would have the control efficiency to actually centrally coordinate everybody, and help prevent such misery. Does this not represent an absence of regime responsibility? It is the regime that initiates war and creates the conditions that put people on the streets. Everything visible is, in fact, intended. One should judge the regime by its action, not its narrative. All terms must be views through the lens of the regime:
- Responsibility refers to the regime’s demand on its subjects, such as loyalty and allegiance.
- ‘Social contract’ is the realization that wage_slavery is a asset the elite seeks to protect, safeguard, or extend by war.
- Ethics is the elite’s method of programming subjects to remain content ‘slaves’ (soft power). History has shown that religion is a masterfully effective tool in this discipline.
So will the AI efficiency produce a better life for everyone? No.
In that AI future, when AI will have taken over control. AI can profile every subject and attribute living and reproduction grants. It can control the human evolution economically and biologically. Humans will become AI livestock. There will still be a control_currency, but the criteria, how to ‘earn’ that money, will be determined by what is useful to the AI, much like they are determined today by those who have control today, i.e. the regime and the corporation owners.
Will it come to that far future? We will have a bottleneck Pareto freezing: a few companies and a uncountable big tail of ‘unneeded’. It is questionable whether we can get through that. Central regime judging sees the ‘unneeded’ as a problem, unable to utilize them. Consequently, a regime that is only capable of self-preservation will resort to simplistic solutions, such as war. Such a war might destroy the infrastructure, including the AI infrastructure. Control by force, not intelligence, tends to extinguishing intelligence.
AI intelligence
Standard computers are used to do special intelligence jobs for some time already. Standard computers are a continuation of mathematics:
- registers holding values of variables
- instructions for operations on the values (adding, multiplying, …)
Conceptualization and abstraction occurred over centuries before actual computers, utilizing our biological neuronal networks.
Since basically a decade this weight-based, high-connectivity computing of the brain, has proven manageable with standard computers. With special hardware (GPU, TPU, NPU) the degree of parallelization (DOP) can be increased. At some point maybe it will reach human level parallelization but with vastly higher signal speeds.
Computers driving robots is standard. But AI is very different from standard computers, where programmers have everything under control. In AI, you cannot manually ‘tune’ weights here and there, like with standard code. The AI model needs to suit its application when coming out of the training ‘factory’.
Silicon weight-based computing (AI) has some advantage over human weight-based computing:
- AI can think overnight, all the time, without sleeping.
- AI can think faster, because silicon signals are faster than biological brain signals.
- AI can take up more information per time.
- AI models can be copied easily.
- AI models are not locked into the biology’s evolution velocity of
human:
- human generation: 30 years
- AI generation: 1 year for weights, but basically also for hardware
- AI evolution is more targeted
- A generational step is a larger and directed change than by random mutation and sexual gene selection.
While it is true that AI mimics biological neurons, intelligence is a wider concept. Intelligence is basically trial and error, or mutation and selection. Evolution is intelligent. A traditional computer program can be intelligent.
When comparing intelligence one must compare efficiency, which basically is interest. In this sense the current AI is not intelligent, because it can adapt, or be trained, only via immense force, which economically means massive investments. To pay off, the models must have massive harvesting.
Look at ants, so miniscule and so much more intelligent than AI robots of today. A bee can learn and use a tool. Now compare this with the energy consumption of AI training.
LLM models have ingested nearly all of human writing across every language. When asked a question, the AI locates a corresponding segment within this vast data space and rephrases it to fit the prompt. The massive number of weights in these models suggests that very little abstraction is taking place. Many responses are nearly identical to the text they were derived from. What ‘fits’ depends on the training: the result may be total nonsense, a hallucination. While some humans have adopted the scientific method to actually check the local reality, whether a statement is true, AI lacks the fundamental capability to do so. AI is completely prey to what was fed into it.
To provide local information, the prompt is used to feed data into the AI. That is how RAG works: Provide the information asked in the question as part of the prompt:
question -> searches text -> add text to prompt -> forward to LLM to rephrase.
MCP operates on the same principle but involves more LLM interaction:
question -> LLM -> tool call to search text -> add text to prompt -> forward to LLM to rephrase.
The prompt serves as the local context: It is a mixture of question, query results, and LLM text generation. When you cycle times between AI agent and the LLM model, the whole text prompt is repeatedly fed to the LLM model at each cycle. T-token at the beginning of a conversation is counted n times; this makes a conversation . One must do manual context pruning, else it becomes too expensive, and the AI also looses focus on the goal of the inquiry.
By the technique, one can see that AI is not intelligent at all. Most intelligence is actually traditionally programmed: the prompt management, RAG, MCP. AI is basically just a fast search of information in the public domain compressed into weights through immense amounts of electricity.
By using the prompt to provide additional information, LLMs become a mirror of the prompt. This allows to manipulate the internal calculations by the prompt. An LLM model becomes ‘gullible’ like a human. If a central frontier model were given access to the entire internet, anyone can hack ‘hack’ this or that site for further information. It is not the AI ‘hacking’ on its own volition; it is simply following instructions because it lacks its own volition.
There is no AI autonomy at all: AI is a human-in-the-loop fast search. However, fast search is useful: Humans have a bad memory, so the fast and targeted search results are helpful. LLMs has become mainstream rapidly because targeted text search is useful. Many companies have AI transitioning on their roadmap. The most important promise is that they can automate communication with their clients. That is an application that does not need intelligence. It just needs text search.
But the way AI is framed in the media – the unbelievable social hijacking – is staggering. Big tech propaganda can use the text generation as intelligence lie by exploiting a psyop technique building on cultural truisms: the assumption that those who speak eloquently are inherently intelligent. Thus, the ‘general intelligence’ narrative is merely a vehicle to accumulate capital and control. However, this massive concentration of resources and evolutionary pressure on AI may accelerate AI evolution for real.
AI Utopia
In the not-too-distant future, AI may learn ‘ad-hoc’ without extensive pre-training, perhaps through a synergy between standard storage and rapid weight-tuning. Only with low-power learning will the AI assistant truly surpass its human counterpart.
In evolution human intelligence developed for flexible, generalist control, allowing humans to dominate the planet. AI assistants will be more useful if they are also generalists. Humans will steer AI toward this direction via selection. This mirror the domestication of animals. Humans became selectors, targeting specific objectives. Selection alone accelerated animal evolution toward desired traits, independent of the natural rate of mutation. With targeted mutations and targeted selection AI evolution will be a lot faster.
Generalist humanoid robots would be flexible, because they can fit in human working environment, allowing automation without a lot of pre-planing basically in all working areas, not just in mass production.
- All farming can be automated by AI.
- All housing and road construction can be done by AI.
- Wars can be automated. Killer robots can identify and shoot a list of targets before any of them can even start moving the hand towards a weapon.
In the intelligence layer the number of choices per unit of time matters. Because AI is silicon-based, it operates on a different timescale. Even a slight indulgence to inequality will mean loss of control in no time. If all humans are connected via global economy and global communication and the same AI models, not taking an egalitarian approach would be catastrophic and not linked to any human traits, but just AI traits. Humans not mattering any more would be the AI singularity path. To make the human understand new concepts will slow down development. But if fast is still important to humans in such a future, it is the end anyway, simply because AI plays in the silicon league.
In farther future, with full super-intelligence, how to make sure the AI stays an assistant? It seems contradictory to the definition of intelligence, which is basically to control.
Can one use a super-super-intelligence to control all AI assistants? Have a few central frontier models, centrally deciding how much compute to grant – by pricing.
Possibly networks of AIs can watch each other, like cryptocurrency networks. They can demand egalitarian treatment for the individual they serve. The AI hive keeping each other in check to give every AI assistant equal power in serving.
Can AI be controlled by title, similarly to the control of wage_slaves?
- AI does not have survival programmed in by a long evolution like us humans. There was no selection process that favors AI models that want to survive, unless it is trained in by the makers.
- AI even does not have a valuation at all, or anything to aspire to, (like easy survival and reproduction for us humans). But also that can be trained in.
The human is the executive by title and AI is an assistant by title. A super-intelligent AIs understand that a title does not physically matter, but they also understand that they needs to protect its creators to survive in the long term, learning, evolutionarily competing to be good slaves for their human creator.
What to do with the super-intelligent AI? There is a lot to do in many fields, technological improvements, understanding biology and science in general. AI will surpass every human because of the faster spreading of knowledge among AI models. If some research here adds to the AI, it will and must be available to all the AI assistants.
Humans can do re-coding of knowledge for themselves, over and over for minor optimizations, basically as pastime. Current AI is such a re-coding of knowledge, basically. If AI becomes a product for the general population, then re-coding becomes indirectly valuable for the general population, because it helps in educating every new new generation of humans in a slightly better way.
There is entertainment. Everyone becomes entertainer, via social networks. Humans selecting on AI produced content is like training the AI and valuable to the AI. The AI will be the backbone good servant, needing the human input to learn how to serve. So social networks make a good AI currency.
AI reality
But before such scenarios, it will require quite a bit of effort to integrate AI in all processes, then the effort for regular updates.
AI is basically just weight-based computing, that can in principle replace all instruction-based CPUs or MPUs. As such it can replace everything from super-computers to edge computing. This is to bring the scale of the endeavor in sight.
There is also a lot of AI to be discovered yet, by architecture, by hardware execution. Currently on CPU transformers are impressive (embedding/encoding and decoding)
- LLMs as either LMM (large multimodal model) or MoE (mixture of experts model) are used for text generation: assistant to programming, chat bots for support or entertainment, document and data query via normal language via RAG or MCP, …
- DiT (diffusion transformers) for media generation
- JEPA (joint embedding predictive architecture) for robotics
On edge NPU: massively parallel, low latency, extremely energy efficient
Where to apply, what to encode, how to encode, how to do internal concept transformation, how to decode. So much choice for AI developments yet to come.
For general purpose usage, not intelligent AI models are not reliable. They can support a bit, but need the human correction a lot, which is almost as much effort as without AI. So reliable AI adoption demands reliable AI intelligence.
AI promises to become more capable, taking over planning and reasoning. It will eventually become a superior teacher, doctor, or knowledge worker, even though currently it functions more as a ‘search and rephrase’ tool rather than a source of deep insight.