Sim-to-Real Gap: The rapid acceleration of artificial intelligence and the irreplaceable craftsmanship

AI paradox: Why your office job is at risk, but the plumber remains irreplaceable

The great reversal: When artificial intelligence meets the limits of physics

We are in the midst of a technological transformation fundamentally different from the industrial revolution. While we stare intently at screens where artificial intelligence composes texts, writes code, and delivers complex analyses in fractions of a second, a silent but radical restructuring of global value creation is taking place in the background. The speed at which AI systems expand their cognitive abilities—doubling their training performance every five months—overshadows the previous law of technological progress. But this exponential curve of digital intelligence masks a paradoxical reality: The physical world cannot be digitized as easily as a filing cabinet.

The following article examines a phenomenon that challenges economists and sociologists alike. We are heading toward a future in which “knowledge work” becomes a mass-produced commodity, while craftsmanship and physical interaction become scarce luxuries. While algorithms threaten the cognitive middle class, the so-called “sim-to-real gap”—the divide between simulation and the real world—protects the craftsman from automation. A robot may be able to quote Shakespeare, but it still fails to lay a tile properly under unpredictable conditions.

Learn why the “deskilling” thesis fails in the physical economy, why the expansion of AI infrastructure paradoxically increases the demand for human labor, and why we are on the eve of a renaissance of craftsmanship that could turn our familiar hierarchies of status and pay on their head. This is not a prediction for the next century, but an analysis of a reality that has already begun.

Between exponential performance increases and the renaissance of craft skills

The contemporary economy stands at a historic threshold, fundamentally different from all previous technological transformations. While traditional technological revolutions unfolded their effects over decades, the current development of artificial intelligence points to a pattern of acceleration that fundamentally challenges our conventional concepts of technological change. Available data indicate that the training performance of large language models is currently doubling approximately every five months, a rate that significantly exceeds Moore's Law and raises questions about the economic and social consequences of this dynamic. Looking ahead, these developments will not only have technological implications but also profound effects on the structure of labor markets and skill requirements.

The central feature of this acceleration lies not in isolated functional improvements, but in a qualitative expansion of the task length that artificial intelligence models can handle. While previous advances consisted of solving individual, discrete tasks faster or more accurately, contemporary developments show that the ability of these systems to engage in longer thought processes and multi-stage problem-solving sequences is expanding exponentially. This expansion of cognitive task capacity is currently doubling every three to four months, opening up entirely new application scenarios that were previously unimaginable. An AI model that can now handle continuous work tasks lasting several hours or even days without suffering fatigue or loss of accuracy represents a categorically new type of work tool. This capability differs fundamentally from previous waves of automation because it addresses not only physical or limited cognitive tasks, but touches upon the entire spectrum of intellectual work.

The fact that training computing power and datasets for language model training are doubling in known timeframes, while energy demands grow annually, means that these developments are not remaining at the speculative-theoretical level but are being driven by continuous material investment and infrastructural expansion. This is not a slow evolutionary process but an accelerated spiral of capital investment, technological breakthrough, and further intensified investment. Leading researchers at major AI development organizations argue that this acceleration is not heading toward a saturation point but is self-reinforcing. The implied timeline for transformative systems capable of handling the vast majority of cognitive tasks currently performed by humans is estimated in discussions among leading AI developers to be two to three years from 2025. Regardless of the precise accuracy of these timelines, the available evidence points to a phase in which the economic and social repercussions of this technology will no longer be gradual or marginal.

The parallel development of software intelligence and physical boundaries

The current cycle of AI development has created a paradoxical phenomenon that has received little attention in modern labor market analysis but is becoming increasingly central: While symbolic and cognitive labor is being rapidly substituted by AI systems, physical and manual labor is experiencing a contrasting dynamic. This asymmetry is not accidental but reflects fundamental physical and engineering differences in the requirements of these two categories of work. The rapid automation of knowledge work is simultaneously generating a massive infrastructure investment program requiring electricity, cooling systems, and the construction of networks and data centers—all components that demand highly skilled manual and technical labor.

The actual limitations of current robotics and physical AI are substantial and do not appear to be imminently overcome. While language models are already achieving superhuman feats in text processing, code generation, and content analysis, existing robotic systems still cannot reliably handle the everyday physical challenges that skilled tradespeople routinely face. The mechanical limitations are formidable: standard robots can typically lift or move only about half their own body weight, while human musculature offers equal to or greater strength than body weight. The difference between simulated environments and physical reality remains a persistently intractable challenge, a problem known as the “sim-to-real gap,” which, despite significant advances in simulation, poses difficulties even for relatively simple tasks.

Furthermore, robotic systems operating in less structured or dynamic environments—the context in which skilled tradespeople typically work—must react and make adjustments in real time. A processing delay of one or two seconds, acceptable for human interaction with language models, will result in errors, damage, or potential safety hazards for a robot performing physical tasks. The real-time processing requirements for physical systems are orders of magnitude more difficult than those for purely digital operations. Additionally, there is the problem of generalization: A robot trained in a controlled factory environment to perform a specific task, such as repetitive grasping, often cannot transfer this capability to varied objects, different surface properties, or slightly different positions. This stands in direct contrast to the remarkable generalization capabilities of large language models, which can transfer complex knowledge from training to solve entirely new problems.

The physical skill requirements of skilled trades are often asymmetrically distributed in their difficulty. While cutting a tile sounds trivial and can be automated under controlled conditions, correctly installing that tile—understanding substrate irregularities, adjusting mortar consistency, and aligning it while considering optical illusions and differences in height—requires combined judgment honed through years of practical experience. A plumber or electrician must not only perform standardized steps but also continuously diagnose problems, identify unforeseen issues, and creatively develop adapted solutions that fit specific spatial conditions. This combination of physical dexterity, diagnostic thinking under uncertainty, and adaptive problem-solving will remain a bastion of human capability for the present and foreseeable future.

The deskilling thesis and its limits in the physical economy

The classic thesis of technology-driven labor market analysis posits that automation leads to a systematic devaluation of work skills. This perspective has historical validity when one considers the mechanization of agriculture or early factory automation, where specific qualifications were indeed replaced by machines. However, a closer look at the current situation reveals a more complex picture that calls into question the validity of these simplistic deskilling narratives, particularly in the context of the physical economy.

First, it must be stated that the current shortage of skilled workers in Germany and other developed economies is not hypothetical or predictive, but a present reality with significant economic consequences. The German Federal Employment Agency documents that approximately 163 occupational fields are currently affected by a considerable shortage of skilled workers, which corresponds to about one-eighth of all assessed skilled occupations. Particularly affected are not only highly qualified fields such as IT, but also, explicitly, traditional trades: construction, electrical engineering, gas and water technology, plumbing, and related professions are not experiencing a deskilling process, but rather a genuine labor shortage. Contrary to the theoretical prophecy from fifteen years ago that technological advances would lead to mass unemployment, a different reality is emerging: in sectors where physical manipulation and adaptability are central, there is indeed growing demand.

Germany's demographic structure further exacerbates this situation. The German labor supply is shrinking structurally due to birth rates below the replacement level and an aging population. This demographic reality, combined with technological change, creates a situation unlike previous phases of automation. Historically, automation often led to a reallocation of labor, with larger numbers of skilled workers moving into new sectors or to more widespread deskilling, which was then addressed by available labor. This dynamic does not work when the absolute volume of available labor decreases.

A second observation also puts the deskilling thesis into perspective: The current infrastructure investment necessary to operate and scale AI systems is not merely creating temporary demand for skilled trades, but rather a structural shift in the composition of the division of labor. Data centers require electricity that must be generated, distributed, and charged. They require cooling systems that must be installed, maintained, and repaired. They require physical infrastructure that must be constructed by skilled workers. The expansion of this physical infrastructure is currently growing faster than the scarcity of AI computing capacity itself, meaning that the demand for skilled trades is not decreasing, but actually increasing.

The Reorganization of Labor Markets: Cognitive Disruption and Physical Value Creation

The classical hierarchy of modern industrial economics, in which cognitively demanding work was valued more highly than physical labor, is undergoing a reversal whose historical significance should not be underestimated. This is not a return to a pre-industrial past in which physical labor was considered primitive or inferior. Rather, it is a redefined logic of value creation in which physical labor, which is not easily replicable by AI, is assigned premium value, while the massive availability of cognitive power from AI systems destabilizes traditionally highly valued intellectual activities.

The underlying economic logic is elegant: the availability of a good or service that is virtually infinitely scalable and continuously improves in quality and performance while decreasing in cost per unit leads to a price decline for that good. Cognitive labor—particularly structured intellectual activities such as software writing, basic data analysis, simple clerical work, and routine customer service—is precisely this type of good from an AI perspective. It is discretizable, digitizable, scalable, and allows for automation. In contrast, manual labor—plumbing, electrical work, masonry, complex installations—is tied to physical contexts, variability, and location-specific presence on a per-unit basis. It cannot be digitally replicated or centrally scaled but must be performed locally, under conditions that vary from installation to installation. From this perspective, manual labor becomes a relatively scarcer good whose value is not eroded by AI competition.

Data from Germany illustrate this shift concretely: While skills shortages exist in many qualified sectors, they are most pronounced and persistent in sectors with a high degree of manual labor and on-site involvement. Approximately two-thirds of job openings for skilled workers fall into shortage occupations, but only about a quarter of registered unemployed workers are seeking employment in these sectors. This indicates a structural misallocation: The available workforce does not possess the skills that are most urgently needed, and these skills are predominantly practical and manual rather than symbolic and cognitive.

The current weak economy in Germany has only temporarily masked this effect. The skills shortage has not been resolved; it has merely been masked by weak demand. Demographic experts and labor market analysts agree that this shortage will grow in the long term, regardless of economic fluctuations. Combined with the technological reality that AI systems are becoming more cognitively demanding, while robotics does not satisfactorily solve physical challenges, a long-term structural pattern is emerging that inverts the classic expectations of technology-driven deskilling.

A new dimension of digital transformation with 'Managed AI' (Artificial Intelligence) – Platform & B2B Solution | Xpert Consulting - Image: Xpert.Digital

Here you will learn how your company can implement customized AI solutions quickly, securely, and without high entry barriers.

A Managed AI Platform is your all-round, worry-free package for artificial intelligence. Instead of dealing with complex technology, expensive infrastructure, and lengthy development processes, you receive a turnkey solution tailored to your needs from a specialized partner – often within a few days.

The key benefits at a glance:

⚡ Fast implementation: From idea to operational application in days, not months. We deliver practical solutions that create immediate value.

🔒 Maximum data security: Your sensitive data remains with you. We guarantee secure and compliant processing without sharing data with third parties.

💸 No financial risk: You only pay for results. High upfront investments in hardware, software, or personnel are completely eliminated.

🎯 Focus on your core business: Concentrate on what you do best. We handle the entire technical implementation, operation, and maintenance of your AI solution.

📈 Future-proof & Scalable: Your AI grows with you. We ensure ongoing optimization and scalability, and flexibly adapt the models to new requirements.

More about it here:

• The Managed AI Solution - Industrial AI Services: The key to competitiveness in the services, industrial and mechanical engineering sectors

Artificial intelligence and the ongoing deficit in physical automation

A critical point for understanding this dynamic lies in precisely articulating what current AI systems can and cannot do with physical tasks. A common oversimplification is that if AI can automate knowledge and cognition, physical tasks are next. This logic, however, is flawed. The requirements for solving physical tasks are structurally different from those for cognitive tasks. While cognitive tasks involve pattern recognition, information processing, and symbolic manipulation—fields in which deep neural networks have made remarkable breakthroughs—physical tasks involve the integration of perception, real-time decision-making, force control, and continuous adaptation within a variable, physical environment.

Currently, robotic systems exist that can perform well-defined, repetitive physical tasks in controlled environments—spot welding in automotive manufacturing plants, high-precision CNC milling, palletizing in structured storage systems. But even these systems achieve this performance only under highly controlled conditions. As soon as variability enters the task—different shapes, materials, spatial configurations, unexpected obstacles—reliability drops dramatically. A robot can be trained to grasp balls on various surfaces with different coefficients of friction. But whether this robot can understand how close to a person in a public space while juggling these balls, while reading social cues and reacting to human unpredictability—that is a fundamentally different problem that remains unsolved.

The technical challenges here are not speculative or theoretical, but concrete and persistent. They include: (1) the sim-to-real gap that exists between simulation training and the real world; (2) real-time processing, which requires delays of milliseconds rather than seconds for continuous physical tasks; (3) high-degree-of-freedom dexterity, where robot arms with 20 or more joints must be coordinated to achieve human-like manipulation; (4) generalization across task variations, which cannot be solved simply by larger datasets alone; and (5) the physical hardware limitations of actuators and gripping systems, which do not achieve the same force-to-weight ratio as human musculature.

These problems are not marginal or limited to the next few months. Leading robotics researchers at respected institutions state that overcoming these problems requires significant research, not simple engineering scaling. In other words, it is not a matter of already having the solution and simply implementing it, but rather that fundamental engineering problems remain unresolved. Under these conditions, the proposition that manual labor will be rapidly automated in the coming years is not evidence-based, but rather speculation.

Labor market transformations: The reassessment of practical skills

The economic consequence of this technological asymmetry is a profound reorientation of compensation structures, prestige hierarchies, and career mobility. Under the pressure of AI integration, symbolic, cognitive activities previously considered highly skilled, highly paid, and prestigious are being displaced from their established positions. A software developer whose tasks are partially replaced by AI code generation systems finds themselves in a bargaining position where the scarcity of their skills diminishes. An analyst whose data analysis can be performed by AI systems loses a relative scarcity premium. A writer or journalist whose workflow is accelerated or replaced by AI text generation sees the demand for actual human writing erode.

In contrast, an electrician whose skills require a specific, variable, and locally bound understanding of context remains in a stable or growing position of demand. This is reinforced by the current demographic situation, in which, in many developed countries, fewer young people are entering the labor market than older people are leaving it. Under conditions of an absolute shrinking labor base, a service that cannot be provided by centrally automated systems is structurally scarce and valuable.

The reputational and status-related reversal of this hierarchy could be even more profound in the long run than the purely economic one. In many Western societies, manual labor has been culturally viewed as less prestigious than cognitive or academic work over the past few decades. This status code could shift if young people observe electricians' salaries rising due to shortages, while starting salaries for computer science graduates stagnate due to AI substitution. Such a shift could have far-reaching implications for educational choices, career aspirations, and social cohesion.

Infrastructure-driven demand for skilled trades

An often overlooked dimension of the current AI expansion is its monumental infrastructure demand. Operating and scaling large AI models requires not only digital computing power but also massive physical infrastructure: data centers, power lines, cooling systems, network hardware, battery storage for backup power, and much more. This infrastructure isn't beamed up; it's built, installed, and maintained through physical, hands-on labor.

The electrification and infrastructure expansion necessary to support current AI expansion is generating unprecedented demand for electricians, HVAC specialists, construction workers, and technical specialists. This is not a temporary demand, but a structural one that grows with the expansion of AI capacity itself. In other words, the faster AI systems scale, the greater the concurrent demand for the skilled tradespeople who build and maintain the physical infrastructure that powers these systems. This creates a feedback loop where the scaling of AI actively drives the demand for non-automatable skilled trades.

To illustrate: When a new computer chip factory is built, tens of thousands of skilled workers are employed for several years before a single chip is produced. This design, electrical, and installation work cannot be performed by centralized AI systems. It requires on-site presence, physical dexterity, problem-solving under uncertainty, and continuous adaptation to local conditions. This is the precise combination of task characteristics where physical AI and robotics are currently not competitive.

Scenarios for the medium-term future: 2025-2030

Based on the current technological trajectory and available labor market data, several plausible scenarios can be outlined for the next five to ten years.

In the most likely baseline scenario, AI-based automation of cognitive tasks continues to accelerate, while physical robotics reaches its current limits and remains confined to specialized, well-defined tasks in controlled environments. This would lead to a two-tiered labor market dynamic, with symbolic work under pressure—decreasing entry-level salaries for many knowledge-based positions, increased demands for specialization and continuous upskilling for those remaining in cognitive roles—while physical, location-based manual labor gains in quality due to scarcity. Salaries for skilled trades (electrical, plumbing, sanitary installation) could rise relatively, while salaries for routine cognitive work would be under pressure.

In this scenario, governments, particularly in countries with aging populations like Germany, would face increased pressure to facilitate immigration of skilled tradespeople, while simultaneously education and training systems would receive a stimulus to re-evaluate and enhance the value of skilled trades and practical qualifications. The currently low number of young people choosing vocational training could stabilize or even reverse if labor market prospects for these roles improve.

In a more optimistic scenario, this dynamic could actually lead to social recovery. The overemphasis on academic qualifications and the cultural devaluation of craftsmanship that has dominated European development over the past few decades could correct itself. An economy that values ​​artisanal quality, local expertise, and practical problem-solving more highly might be less vulnerable to the kind of technological disruption that massively concentrated AI capacity creates. It could also lead to less social inequality, since the premiums for highly skilled craftsmanship are not as extreme as the historical premiums for elite cognitive education.

In a more pessimistic scenario, the adjustment processes could be chaotic and painful. Generations of workers groomed for cognitive careers could suddenly find themselves in less advantageous positions, without access to skilled trades qualifications or opportunities for rapid retraining. Social cohesion could suffer under the stress of this shift. Countries that fail to quickly adapt their education and immigration systems could experience acute shortages of skilled trades, hindering their infrastructure development and, consequently, their ability to scale their own AI.

The Renaissance of Craftsmanship in the Age of Symbolic Automation

The economic analysis of the current phase of AI expansion suggests a pattern that differs fundamentally from the prophecies that were dominant in the 1990s: instead of universal deskilling and mass unemployment through automation, there is an asymmetric disruption in which symbolic, cognitive labor comes under pressure, while practical, physical, location-based labor becomes structurally scarcer and therefore more valuable.

This shift is not speculative, but is already evident in current labor market data. The present and projected shortage of skilled tradespeople in Germany and comparable economies is not a transition to something else, but a structural feature of an AI-driven economy. The technological limitations of current robotics and physical AI do not point to rapid breakthroughs, but rather to persistent and potentially decades-long challenges in automating tasks with physical complexity and contextual variability.

For workers, this means that practical skills—as opposed to cognitive skills, which are increasingly being replaced by AI systems—offer a form of security and structural relevance. A young person who chooses to train as an electrician, plumber, or bricklayer is making an economically rational choice, not for nostalgic or cultural reasons, but based on the cold logic of scarcity and demand.

For societies and policies, this means that the requalification of education and training systems is becoming an urgent task. This is not a matter of education policy alone, but of fundamental economic adjustment. Countries that rapidly increase their appreciation, remuneration, and status ascription of skilled trades and reorient their training systems accordingly will be more economically adaptive and resilient in the coming years than those that cling to an overemphasis on cognitive labor.

The current phase could be historically recognized as a period in which the overinvestment in symbolic skills was corrected and practical, creative, material-based labor received its long-overdue cultural and economic revalidation. This is not a return to a pre-industrial economy, but rather the next phase of a technologically advanced economy in which the limitations and asymmetries of AI automation are understood and the continuing importance of human craftsmanship is recognized.

☑️ Our business language is English or German

☑️ NEW: Correspondence in your national language!

Konrad Wolfenstein

I would be happy to serve you and my team as a personal advisor.

You can contact me by filling out the contact form or simply call me on +49 89 89 674 804 (Munich) . My email address is: wolfenstein ∂ xpert.digital

I'm looking forward to our joint project.

☑️ SME support in strategy, consulting, planning and implementation

☑️ Creation or realignment of the digital strategy and digitalization

☑️ Expansion and optimization of international sales processes

☑️ Global & Digital B2B trading platforms

☑️ Pioneer Business Development / Marketing / PR / Trade Fairs

Benefit from Xpert.Digital's extensive, fivefold expertise in a comprehensive service package | R&D, XR, PR & Digital Visibility Optimization - Image: Xpert.Digital

Xpert.Digital has in-depth knowledge of various industries. This allows us to develop tailor-made strategies that are tailored precisely to the requirements and challenges of your specific market segment. By continually analyzing market trends and following industry developments, we can act with foresight and offer innovative solutions. Through the combination of experience and knowledge, we generate added value and give our customers a decisive competitive advantage.

More about it here:

• Use the 5x expertise of Xpert.Digital in one package - starting at just €500/month