AI aspiration

30.07.2024

• Current AI development is driven bottom-up by ML advances. AI product definitions stem from the same tradition, which is limiting.

• A top-down approach helps to draw insights form a big picture and define better AI products, which are more profitable, create lesser externalities, and remain safe.

• AI can resolve a centuries long quest. It can help humans to ascend one step further. For that the AI shall become a cross-cultural unifier, instead of a subcultural magnifier. The technology is already here, it should be trained, assembled and packaged appropriately.

• The effort demands cooperation across multiple stakeholders. AI technology should not be left to technologists. They are going to squander the resources and likely lead the societal development into a dead end.

1. Introduction

The term artificial intelligence is frivolously used in popular culture. It generally means all synthetic things, which perform some unattended decision making. A simple light dimmer, that reacts upon ambient light, would fit the definition. And it is false.

Scientific literature tunes off the misconception by predominantly applying the term machine learning (ML). Machine learning is a particular method to facilitate synthetic decision making. 96 out of 100 AI-researchers are today purely ML-specialists. Journals carrying AI in their names are actually ML-shops. It is also a huge mistake. A water supply (here AI) cannot be reduced to plumbing (here ML). And the plumbing cannot be reduced to PVC pipes welding technique (here deep learning - the most popular ML domain).

The notion of AI, that would be correct, but remains exposed through lip service, is encompassing actual thinking. The synthetic thinking ability is an attack point for sceptics. Their outrage is bad for publicity and investments. So the ambitious authors keep the terminological discussions out as much as possible. It has gone that way for decades up to the extent that some journal editors don't know better. Unfortunately, the terminological skew significantly deforms our understanding about AI utility. 

I follow an uncommon approach in the article. 

• Firstly, I want to elucidate that the terminological skew creates a significant difference. The difference triggers much more than a variation in accounting. It impacts the very essence of our civilization. Aiming for an easy to target utility could be fatal. Aiming for a sensical utility could be a blessing. 

• Secondly, I want to point at the most investment-worthy AI-utility. It is deliberately a conversation starter about AI for people in business suits with ties. It was crafted in qualitative intelligence terms. The ML-guys in shorts may find it lacking direct instructions, mathematical quantitative terms, and TLDR. If the interest persists, the out of scope content could be partially found in my peer reviewed papers.

2. AI term 

AI is a technical system capable of synthetic thinking. The synthetic thinking isn't something out of the world, next thing after absurd. It just means that machines have to be capable to do knowledge synthesis. The synthesized knowledge introduces novelty. It allows to produce an "unexpected" decision making result. The unexpectedness is what differentiates AI from a programmatically set light dimmer. 

Any synthesized knowledge is a combination of multiple elementary knowledge components. A teacher, who installs the knowledge components into a pupil's head, wouldn't be surprised of any synthesized result. So do humans, who instruct machines. Controlling the initial set of knowledge components is also a programming process. It produces a fully predictable set of outcomes.

A true unexpectedness, i.e. true thinking, would be enabled by an original knowledge acquisition. That is, machines shall expand the set of knowledge components by their own effort. There should be no reliance on human help. Humans shall not take part in data processing. Humans can still be a facility that provides services, which are not related to data processing. 

Knowledge set expansion in human thinking is done by an analysis process. At the very least, a thinking person takes a data blob; decomposes it in separate elements; proves if they are mutually exclusive and collectively comprehensive, i.e. assemble a whole picture; stores the newly acquired knowledge tidbits. An IT system, that is capable of unassisted data decomposition and attribution, would do exactly that. 

Modern AI is capable of both analysis and synthesis. Thinking machines do belong to reality. The thinking may remain naive and simple, but it will catch up soon. Technological advancements are now focusing on capacity growth and systems architectures to deliver proper utility. The latter, the proper utility, is in focus of the article. 

AI helps to overcome some anthropologic deficiencies, as any tool. Thinking machines help to overcome thinking deficiencies. Economically viable thinking machines help to overcome thinking deficiencies that impact economic activity, i.e. management limitations.

3. Cognitive limits

A retrospective view over the evolutionary path of western civilization hints that societies have been periodically challenged by stagnation. The underlying reason was that the main tool, which underpins societal wealth, had been completely exploited. The growth potential had been completely depleted, or was approaching an end due to the effect of diminishing returns. Luckily, a decline of one society within western civilizational tradition has been compensated by growth of another society. That is, one tool form has been substituted by another, more advanced, tool form. Unfortunately, there were also events, when a fundamental tool approached its end of life. Then the very civilizational foundation cracked, and a decline was plaguing all social units. 

An instrumental limit hindered growth each time a local decline could have been observed. Invention of a method to overcome the instrumental limit was the condition to establish new big business successfully. There have been numerous instrumental inventions. They allowed to tapp the main productive principle more efficiently.

A conceptual limit triggers global demise. Discovery of a conceptual limit is an early warning for looming changes in socio-economic regime and big business as well. There have been two events to the date, as far as I've been informed. Today we are facing it for the third time. Methods, that are capable to overcome conceptual limit, solve a fundamental issue but remain also instrumental in nature. They are conceptual by being applied to conceptual issue, not by attributes. Any such method will eventually wear out and will require an upgrade by complimentary instrumentation. Depending on how well the initial method implements new productive principle, the upgrade will be required later, rather than sooner.

The relation between instrumental and conceptual limits is the same as between the symptoms and a root cause. Instrumental innovations alleviate the grip of symptoms. Conceptual innovations get rid of the root cause. By the time all possible symptoms have been treated, further improvement can be triggered exclusively by curing the root cause. Other issues come to the forefront after that. And the hunt for symptoms begins anew. 

Being a process at the civilizational scale, hitting a conceptual limit it is slow, but cruel. We had probably seen the limit as far as two hundred years ago. Waste of time and resources would probably mean to loose a lot. Success or failure both will reshape geopolitical and elite landscape. In the case of failure, we will additionally find us thrown back by several centuries. In contrast, a proper investment will realize enormous innovation opportunity. A well-executed solution to overcome conceptual limit can cover all inherited debts, provide resources for erecting new order, and leave every participant's net balance much better off. 

Being a civilizational process, the breaking through the limitations can adopt various solutions. An AI-based solution is not exclusive. Though, given the ongoing AI-arms race, the dice has been thrown. AI solution is a low hanging fruit. Whoever applies AI to overcome the actual conceptual limitation will define the prevailing civilization format, shape the next centuries, and possess untold wealth.

Let's consider some limits attributed to human thinking. The aim is to get a sense how the limits impact economy, social structures, and our lives.

Dunbar's number is a fundamental limit underpinned by finality of the head size. It's the number of items in a list that can be processed without splitting in independent batches, or using extrabodily tools. Surely, the number varies among individuals, but the expectancy stays within a couple of hundred entries. It hints the maximal overall complexity of a mental task that can be resolved. It can represent the number of considered variables, or it could represent the list entries to sort and optimize. The limitation handling was mastered in our mammalian past in the form of enforced hierarchies. Each hierarchical node delegates management of overflow items down the command line. 

A derivative from Dunbar's number is Graikunas' number. It describes the maximum number of direct subordinates. Assume each agent is a multifunctional asset. Each function under control is essentially a parametrized link from principle to agent. The total number of serviceable links is Dunbar's number bound. The more engaged are the agents, the lesser span of control would be available. Each agent will consume an increasing number of link slots, leaving less for others. For example, consider a lion pride ad extremum. A single rival at paar with the pride principal can completely incapacitate any principal's activity, demanding principal's devotion to keeping the war at bay. 

Coase limit adds a strategic perspective to the rivalry. Strategy helps to get rid of unimportant functional links, hence increasing effective link capacity. A strategically talented principal will be somewhat superior to its equally strong rivals. It is -though- not a 1:1 exchange. Strategic thinking forces to take more Dunbar's slots to process a single link of importance. It trades quantity for quality and introduces the notion of trade-off. Coase limit describes the trading rates. Getting conscious about them was a trigger for the neolithic age. For example, men adopted stone hummer. They traded additional weight to carry and additional labor to invest for a punch impulse that can crash the bones. Ancient humanity evolved into a casts society by exploiting the principle. For example, the casts were consisting of masters (rivals to pay attention to) and slaves (unimportant undistinguishable service units, i.e. no attention paid). The masters were left responsible for any cognition, including ideation and realization of wealth. Stagnation resulted in revolts and wars. Wealth surplus was assuring stability of the social system. 

A constant drive for wealth surplus supported emergence of Egypt and similar congregations. It also stimulated masters to design more complex (productive) and universally applicable (to reduce overall Dunbar's slots consumption) utilities, hence empires and pyramids. The complexity brought masters at the edge of thinkable. There was a need to increase cognitive conceptual capability. A purpose-driven breeding did not play out well. Humanity faced the Leibnitz limit. Up until invention of a functional, i.e. fair and diverse labor division, the ancestors of western civilization were predominantly decadent and declining. No instrumental breakthrough could really help. Instrumental changes do not change the fundamental balance. They provide suboptimizations capable of keeping conceptual setup afloat for a while. Nothing will help if it is fundamentally brocken. 

Hellenic period offered a solution, that was properly implemented in Rome. Up until our days we have been exploiting labor division, while intensifying its effectiveness by boosting efficiency. Every slowdown in implementing the principle had triggered a turmoil and did incentivize to implement an instrumental modification. 

The casts system has been practically abandoned as a reaction to Ross limitation. The mechanization and automation were reactions on negative effects imposed by Planck limitation. Whatever the method, they all aligned during the teameria period (see picture) with the principle: split any production process in as many parts as possible, while incentivize people to willingly participate in the process. 

The workgroup PK39 stipulates that modern systems have become so complex and include so many simple work stations, that we again are in the situation that is similar to 4000 years ago: a non-impactful crowd vs a few visionaries, super-integrators, rivals. There is no capacity left to supply wealth-creating ideas at a sufficient speed rate. Only this time around, the bottle neck is not the thinking out of the box, i.e. breaking out, but the thinking into the box, i.e. assembly in. The crowd possesses required skills and knowledge, but lacks initiative. We are at a Hodgson limit. 

Stumbled WTO globalization, failing multicultural approach in the EU, demise of communistic states are evidences for our inability to apply JOIN function. The root cause is that people at the far edges of a virtual assembly line don't understand each other. Thier world views are radically different. It is good for synthesis. But the radical difference has no place for commonality. The brain size doesn't allow to maintain a plethora of world views at the same time. Labor compartmentalization alienates people. A productive conversation is disabled. Synthesis fails. There are no proper organizational tools to aid the issue. People in the same boat work against each other. All sorts of logistical disarray effects hinder efficient production and economic growth. It is as if a howling dog bites its own tail continuing the vicious cycle. 

The most talented people are subjects to the same limits as imbeciles. There is nobody, who can capture the whole picture and preach the struggling individuals at their work stations. Big business efficiency dwindles.

4. Hodgson limit impact

Any introduction of a tool, that resolves issue at a civilizational scale, triggers a logarithmic growth. Its mpact is initially breath capturing. Results diminish over time and the transformational impact slows down up until a stagnation. The same dynamics is to observe for instrumental and conceptual tools. 

Conceptual tools shift the entire population into a better situation. There is a certain fluctuation around the norm. It is characterized by tolerance interval. 

Instrumental tools provide a trade-off, i.e. improve situation for a strata, while worsening the situation for others. Such tools significantly increase tolerance interval. If not concerning abut the reasons it looks like fluctuation is growing. There comes, indeed, a systematic disparity into play. An individual's situation likely drifts upwards or downwards, away from the average value. Instrumental tools reinvigorate efficiency of the conceptual tool. Therefore, the the total economic effect is usually larger than predicted. The conceptual boost component can compensate for instrumental drawback. The disadvantaged strata can end up in a better situation after all. 

Instrumental changes enjoy unrestricted support so long the overall boost is greater than the fallback due to wealth redistribution. Should a redistribution kick some pople out of their habitual comfort zone, there comes an opposition and armed destruction. The art of good governance is to secure the lowest life standard, and to develop individuals in the lowest strata so, that the life standard becomes self-sustainable and there will be no need for support in the future.  

Conceptual shifts create new culture. Each culture provides two answers: A1: why are we together? A2: Why somebody would wanna sacrifice for it? No change is sustainable and possible until both questions are resolved and settled. 

A1

It is a simple case. Civilizations are formed to enable reproduction and exploitation of a tool, which sustains life and improves survival chances. It all can be measured by a proxy metric - wealth. Every civilizational tool does boost wealth by several orders of magnitude. Conceptual tools boost more, instrumental less, both boosts are substantial. 

Should the Hodgson limit be resolved, the world global product will tripple, presumingly within the next generation's lifespan, reaching at least 300 trillions a year. Humanity could expect a strait of prosperity for the next 250 years, to say the least.

A2

A rapid expansion creates a strata of nova riches. The advancement is then institutionalized by a novel social system design. At the same time it forces to rapidly abandon the old design. This change pulls the rug under some individuals, who are destined to adapt slowly. There were two kinds of response on such interventions. The first kind is inertial scenario. It led inevitably to some sort of revolution and dethroning of the initial intervention beneficiary. The second kind is proactive scenario. It deploys a mechanism to conciliate the disadvantaged strata. 

One way or another, people, who have innate understanding of the disadvantaged strata life style, will be the most active defenders of the new world order. They would rather die, than fall back. Nobody values civilizational comfort as much, as a former primate, who was once on the list among endangered species. The mechanisms to upscale life standards for disadvantaged people are important to establish positive defender balance.

If somebody finds a way to overcome Hodgson limit and builds new big business upon the idea, they will have to deal with the disparity issue and participate in a reconciliation effort. If history will be left to its own devices, the business will face a massive counterforce. A technological solution will likely not survive the turmoil. The idea and knowledge may survive. The first-to-come business, that sits on vulnerable assets, will definitely be gone.

5. Ushering into the new era

At the very least we'd like to know what to do and what not. Each category has active and passive components. Considering the active components, we have two activity areas: 1) goals to pursuit, 2) outcomes to prevent. 

Key effect and competences

Let's take a historical perspective. Let's consider what uplifted humans in the past. 

Innovators were focusing on acquiring a set of master skills during the time period before Leibnitz limit. A better mastery directly transposed into prosperity. Productive surplus could easily feed the specialists, solo-enrtepreners. The time period gave birth to accountants, astronomers, agronomists, etc. Society vehemently protected their unique knowledge by putting those specialists behind invisible walls, accepting a casts system. Since approaching Leibnitz limit, the knowledge increasingly began to spread rather than accumulating. Universal availability of knowledge alleviated the need to protect knowledge carriers. Social system degraded the casts. The two remaining casts (masters and slaves) were enforced against the will by leveraging weapons. Productivity and strength of such society was steadily dwindling.

Overcoming Leibnitz limit required to split knowledge and tasks within one profession. That means, instead of competing against each other on the ground of better mastery, a number of equally competent people had to work in accord. Aristotelian solution was to decompose the tasks by ever finer grained subject matter. For example, the approach created various accountants. There is a book keeper, controller, financial officer, cashier,- to name some of them. Each specialist has some unique skills and domain knowledge that are not attainable to others. Thanks to the compartmentalization principle, we have way too many sciences and incompatible world views today. The compartmentalization is still producing a plethora of rare professions. The vast variety of specialists constitutes a set of problems. For example, to find matching specialists is to search for a needle in a haystack. Instrumental tools like social networks are modern takes on this problem. 

A key competence for the post Leibnitz limit period is manning teams. A bunch of professions like psychology, sociology, management, etc. serve the need. None of that helps if a set of proper specialists cannot work as a unit. People from far distanced disciplines don't understand each other enough, even by the help of intermediaries. The intermediaries themselves have become so diverse that rival each other for the right to dictate rules. This actually means to force others to assume one point of view. Whoever wins in the silos battle is as much productive, as the represented silo itself. End of story. There is no benefit any more in further professional compartmentalization. Compartmentalization is narrowing one's tune and breeds deafness. Modern world is a set of blind people, who describe an elephant from their unique perspectives, but it does not help, since blind people have also become deaf. We are constrained by the Hodgson limit.

The hindering impact of Hodgson limit can be reduced by instilling active cooperation. Firstly, the specialist's mentality could be shaped so that no stone is being held in the pockets. Secondly, a culture of innate unsettled interest for others could be established. Then one individual would give everything, another take everything possible. Both unreserved expression and perception capabilities have to be developed. This has been attempted as follows:

• Religious confessions in their esoteric practices have investigated training methods to become super humane. That is possessing a high moral, following the principle to serve others first. It is -in principle- a move from conscious to sentient being. Sentient beings cooperate easily if it helps the teamsters. They give everything, resolving the expression question. They also possess empathy that facilitates understanding even if there is no common ground for it. That resolves the perception question. Sentient being is biologically destined, but extremely slow in coming. The problem, that we face now, has to be treated by a universally available trait. Otherwise betting on sentiency will establish fascism. Long live Christian democrats!

• Communists have attempted to tear down the expression barrier by unthreatening knowledge exchange. The wealth of specialists was promised to be decoupled from holding up to unique tidbits. Unfortunately, there was no constructive approach whatsoever to incentivize knowledge acquisition. It concerns both new knowledge and learning (perceiving) the known. An individual on welfare has no incentives to put any productive effort forth. There was no due effort even in socialistic systems. They say, socialistic is a pre-stage before communistic system. People were fed there proportionally to delivered effort. It is similar to capitalistic approach, albeit somewhat skewed in defining the proportions. People did not like the jokes at all. The approach failed plausibility test.

• Europeans have attempted within the framework of European Union (EU) to train empathy for other nations and trigger congruency among various value systems. This goes under multiculturalism agenda. EU is also remaining a capitalistic system. The latter helps to retain professional interest in knowledge acquisition, including perception. The perception capability development would be perfectly served herewith. Unfortunately, EU has not yet established a positive feedback to boost expression capability. Hillbillies hold on their belongings and positions more firm than ever. Crowd is the primary target for socio-economic changes if we advance starting from an elite-crowd social model. Yet the crowd, the underprivileged people, have not responded after the elite demonstrated its good will. To the date EU-multiculturalism created some tolerant net payers, but further radicalized the net receivers. Hungary and other countries evolved into egoistic solo signers with the only song: "me first, give me, what is in for me?" It seems the multicultural approach is failing to hit the most important target. Long live the EU!

Apart from social experiments there is a technological option too. A method that can unrestrictedly elicit knowledge and translate it across knowledge domains could work around the closed ears. An LLM model, that captures vast amount of knowledge domains, could potentially zoom into some distant knowledge and rephrase it appropriately. LLM-based products could help a narrow specialist to fathom others, while staying in comfort of her professional determinants. AI could become a unifier, a teacher, a digital repatriation tool. If AI is architected and trained primarily for automated unattended linking across-domains, it can do resolve the greatest civilizational challenge of our time. The required technology to usher into sociora era is already here. It has to be packaged appropriately.

Inclusive development strategy

AI applications are teamerisitc today. They resolve some pressing issues, applying various instrumental boosts at the long outdated concept. In a greater sense they try to put a decaying body on life support in order to win a race. I think this activity is of negligent importance and meager outcome. 

There are three reasons to stick to the hopeless plan:

• Centuries of engineering tradition. There is a greater efficiency in doing the same thing repetitively. Engineers have been instrumental to current paradigm. A lot of knowledge goes without saying and critical reconsideration. Looking under each stone will be costly. Setting a novel task will disqualify many vested reputations. It will kick off a revolt movement. Many technologists will argue that any pursuit for a different AI is expensive and wrong. A rare politician will have the stamina to persist. No civilizational solution will ever evolve without political stakeholders.

• Engineering disciplines are subjects to compartmentalization. The aforementioned blind spot that AI is not the same as ML can be drawn as an example. ML is mostly using a micro-view bottom up approach. AI benefits the most from a macro-view top-down approach. Decades ago ML split off and became a separate science that still wants to tap the larger money pool. ML today eats lunch for ML and AI. Give an ML-scientist a hope, she will chew the entire computer science and natural philosophy for breakfast. It would be hard to get rid off this compartmentalized idiocy. The community is barely steerable. A significant collective effort would be required to turn around. It is much simpler to let it be.

• Public justification. A massive turn around in AI sector will consume trillions of dollars. There will be requirement to justify the spendings. Private sector financing enjoys lesser attention. But imposing a regulatory guidance upon it will cause the same effect. There are many inherent unknowns if digging down the balance sheet. It will annoy the general public. Worse yet, the negative effects will be exposed. It will be a daunting task to defend an alternative development course. While letting things go, as they do, is a hands off, no fuzz activity.    

A tempting idea would be to let all flowers bloom, i.e. inclusive development strategy. Both AI approaches could be supported. It would be a substandard decision for three reasons:

• The amount of investment is prohibitive. It costs $100M to increase model complexity by one order of magnitude (OOM) at the current rate of technology advancement. The costs grow in a geometrical progression. The actual GPT4 complexity is at 6 OOM. We need to reach at least 15 OMM to lift corporations at the orbit with better efficiency. It will consume $.25 trillion if accounting for efficiency improvements. Given that successful project has at least three contenders there will be investments at $1 trillion. The sum is comparable with the total annual FDI. There will be finance shortages and overspendings under an arms race condition, that currently involves more than five counties. It would be hard to run a program with two different ideas developing in the same time.

• Much to loose. Firstly, if a rival jurisdiction bets all-in on a potentially better technology and succeeds sooner, it will change the geopolitical status quo. There is no solace in sponsoring one's own demise. Secondly, an uncompensated inequality growth will instill constituency to ask if politicians are fit for the job. The more focus on instrumental paradigm, the larger the problem. Pursuing the instrumental option is outright politically dangerous. Pursuing a conceptual option will immediately provide positive use cases up to complete alleviation of the problem over time.

• Existential risk. Current AI paradigm is capability magnification, i.e. AI reinforces specialists in their own vertical domains. It makes labor competition more fierce. That maintains demand for ever more powerful reinforcement. A corollary is that a drive for more capable AI has no cap. AI will eventually arrive at ASI. Such omnipotent systems won't be controllable. In contrast, an AI, that facilitates cooperation, does not have to be overly intelligent. It requires versatility and universality rather than leverage. It will be productive approaching and not exceeding AGI stage. It will remain controllable and safe. 

A strategy to serially develop one and then another product is not plausible, if the naturally evolving instrumental AI is put forth. The knowledge magnification quality will be assessed by the compartmentalized specialists. The greater compartmentalized specialists will have greater impact in reinforced learning. It is an echo-chamber that breeds narrowmindedness. At the same time, this AI will allow underdeveloped people to skip basis understanding and rely on AI to magnify what's left. That will lessen overlap with others even more and cause atrophy of fathoming skills. AI within teameria business paradigm will have a net zero effect, furthermore intensifying compartmentalization. There will be no productive force left that maintains the skills to come over the Hodgson limit. That being accompanied by a long lasting disparity impact forms a hopeless proposition.

Negative effects reconciliation 

The amount of resources required to reconcile negative effects is lesser for conceptual tool than for instrumental ones. Instrumental tools require welfare programs for an indefinite period of time. Conceptual tools require substantial care for the transitional period only. Moreover, the fundamental property to universally link people form diverse corners, will supply positive use cases at steady pace. A substandard safety net could hold if people see a brighter future.

The most common historical reconciliation mechanism was to export a problem outside of the ecumene. That is, the disadvantaged strata within the intervention region was not the net payer. Aliens, such as Neanderthals, Incas, Chinese, were paying for instrumental ascend of humans. The disadvantaged strata was herewith requalified into underprivileged strata. The underprivileged strata could enjoy spillover effects, or some financial aid, which was financed by robbing indigenous people outside of civilizational realm. For example, the opium wars did cover the robbing operations in China. Given that the entire Earth is spanned by the UNO accords, there is no exporting option in earnest. Either a long war, or a third option have to come up. 

The third option could have been a universal income. Considering the stigma of social policy in communistic states; and the conservancy principle that force is equaling the counterforce, i.e. capitalistic financial incentive equals the amount required for redistribution, the universal income in a nonstarter. The cumulative surplus will certainly outpace the negative impact. But the expected payback time would be significantly delayed. I doubt, any private office would invest for so long. The national state governments have not been noticed to be that decisive in their regulatory policy either. 

Under the circumstances, a conceptual tool should be preferred over the instrumental one.