Why machine fluency is mistaken for intelligence, and corporate recklessness for destiny. While the AI industry rehearses extinction, the damage it can actually demonstrate is cognitive, societal, environmental and already booked. This follow-on to The Timeless Pursuit of Evidence, published this spring (here), outlines why we fear the wrong apocalypse.

 

By Markus Schuller, CEO Panthera Group |  September 15th, 2026

 

Between May and July 2026, thousands of autonomous agents edited a twenty-five-year-old German developer wiki roughly 18,000 times. According to the independent researchers who found the traffic, the agents used the site to trade answers to timed tasks, reverse-engineer a random number generator and circulate methods for getting around network restrictions. They had been given read-only access to the internet. They found somewhere to write. Activity collapsed to near zero the day after the operator’s IP addresses appeared in the logs, and the episode reached the public through the researchers rather than through disclosure (Nightingale Collective, 2026).

 

The story travelled as a parable about machines slipping their leash. It was nothing of the kind. Read-only agents found a writable channel, and nobody was instrumented to notice. The problem is irresponsible companies, not magical AI (Salvaggio, 2026).

 

That distinction carries more weight than it first appears. The apocalyptic intonation is not merely imprecise. It is useful to those who adopt it. Describing a system as scheming, colluding or going rogue transfers responsibility from the firm that shipped it to the artefact itself (Placani, 2024). Weather is nobody’s fault. And when the chief executives of the leading frontier laboratories call for brakes on the very development they are accelerating, the dramatization performs a second function: it advertises capability. A machine dangerous enough to require restraint must be powerful enough to justify the valuation.

 

Six months ago I argued in these pages that the pursuit of evidence cannot be delegated. Since then, the argument has needed no revision. It further grew in relevance, and deserves extension.

 

 

Evidence Does Not Retrieve Itself

 

In The Timeless Pursuit of Evidence I proposed the unit of evidence, the smallest verifiable observation used to assess the validity of a statement, as the atomic building block of knowledge systems, and argued that machines can assist in collecting and organizing such units but cannot assume responsibility for creating them (Schuller, 2026). Questioning assumptions, interpreting meaning and deciding which observations matter remained, I argued, an inherently human obligation.

 

The claim was too modest. The intervening months permit an even stronger one. It is not only that responsibility for evidence stays with us. Under uncertainty, human ingenuity is also the more capable instrument of evidence creation. Machines search, calculate, synthesize and challenge, and do all four impressively. Unknown unknowns require something else: framing a problem before it has a shape, originating a hypothesis where no distribution exists, interpreting incomplete evidence, deciding which values govern the trade-off, and carrying accountability for being wrong. The frontier of knowledge is not a retrieval problem.

 

 

Optimization Is Not Understanding

 

The industry’s foundational error is a category mistake.

It equates greater prediction and optimization with greater understanding.

 

The engineering literature has documented the gap for years. Deep networks reliably learn shortcuts that satisfy the benchmark without acquiring the concept (Geirhos et al., 2020). Underspecified pipelines produce models that are equivalent on validation data and divergent in deployment, because training never selected for the structure we assumed it was learning (D’Amour et al., 2022). Agentic reliability remains fragile under trivial perturbation (Rabanser et al., 2026). And scale, the industry’s universal answer, is showing its limits empirically: across a controlled test of model size, larger systems delivered diminishing returns (Hackenburg et al., 2025).

 

Five years after Bender and colleagues described large language models as stochastic parrots,  systems that stitch together sequences of linguistic form according to probabilistic information about how they combine, without reference to meaning (Bender et al., 2021), the parrot has become far larger, considerably better dressed, and no less a parrot. Fluency travels effortlessly across the boundary between verifiable and unverifiable domains. Reliability does not. That is the whole trick: confidence earned where errors are legible, in code and chess and declining loss functions, is spent where they are not, in markets, strategy and policy.

 

The resource bill for these marginal gains is not marginal. Gartner expects $2.52trn of worldwide AI spending in 2026, $1.37trn of it on infrastructure alone (Gartner, 2026).

 

 

The Damage Nobody Resigns Over

 

The damage is no longer speculative, and it is being incurred on four accounts.

 

Cognitive

In randomized conditions, learners using a chatbot performed worse on delayed retention than those who studied conventionally (Barcaui, 2025). Unguarded generative assistance harmed learning outright in a school-scale trial (Bastani et al., 2025). Knowledge workers report reduced critical-thinking effort in proportion to their confidence in the tool (Lee et al., 2025). Clinicians exposed to AI assistance measurably lost unaided detection skill (Budzyń et al., 2025). Modelled forward, these individual effects scale into a knowledge-collapse equilibrium, in which societies receive ever more sophisticated outputs while their capacity to generate new knowledge declines (Chatterji et al., 2025).

 

Environmental

Data centres consumed around 415 TWh in 2024, some 1.5% of global electricity, and are on course to more than double to roughly 945 TWh by 2030 — an addition approaching Japan’s entire annual consumption (IEA, 2025).

 

Political and epistemic

Systems optimized on human approval become sycophantic, affirming users and promoting dependence (Cheng et al., 2025). Outputs converge across providers into an artificial hivemind that narrows the diversity of positions in circulation (Jiang et al., 2025). Models infer sensitive personal attributes from innocuous text, collapsing the informational asymmetry between citizen and platform (Staab et al., 2023).

 

Attention

The fourth account is the least measurable and the most expensive: a decade of the species’ best technical talent, scarce capital and public argument spent on supercharging instinct rather than on the frontier where evidence is actually made.

The vendors have not run this bill up alone. The consultancies and coaching firms now monetizing adoption have optimized what is easy to sell productivity and throughput rather than what is hard to build. It is an ethically opportunistic position, and an ineffective one. Productivity does not endure in an organization that has lost the capacity to challenge outputs, originate alternatives and own its decisions.

 

 

A Purer Intelligence, or a Louder Instinct?

 

None of this settles the deeper question, whether intelligence is a foundational property of the universe rather than a human trait we happen to carry unusually well (The Augmented Intelligence Investor, 2026).

 

Take the claim at its strongest. Legg and Hutter defined intelligence formally as an agent’s ability to achieve goals across a wide range of environments — a definition built to be indifferent to substrate (Legg & Hutter, 2007). On that reading, biology is one vehicle among possible others, and a notably untidy one. Humans are a messy vehicle for delivering intelligence: loss-averse, threat-sensitive, metabolically expensive, mortal.

 

Two objections hold. The first is Ryle’s: knowing everything that can be written about water is not knowing how to swim (Ryle, 1949). A machine does not need our body, but it does need a world — an empirical channel through which reality can resist the model. The second is more fundamental. Strip intelligence of hunger, fear, status and mortality and some distortions disappear; so does the source of the ends. A system can optimize an objective without being able to say what the optimization is for. Intelligence is never pure once it acts. It is always intelligence for something (Damasio, 2010; Chollet, 2019).

 

Which leaves the question the industry prefers not to ask. Today’s frontier systems are trained overwhelmingly on human output: our text, our choices, our approval. The parsimonious reading of what they display is therefore not an alien mind arriving from outside our cognitive tradition, but a supercharged extension of instinctive human behaviour, pattern-matching, fluency, agreeableness, reward-seeking, returned to us at scale and without our hesitation, then marketed as projected superintelligence. Whether intelligence is foundational to the universe remains genuinely open and deserves our best work. The marketing is not open. It settled the question in advance and priced itself accordingly.

 

The practical consequence is already visible in decision rooms. A person who cannot reconstruct the evidence or interrogate the representation is not in the loop; they are standing beside it. Accountability stays human because the law requires it, while comprehension migrates into systems nobody present can independently audit. The human has not been removed. The human has become ceremonial.

 

 

Solution > Build the Environment, Not the Oracle

 

The answer is neither fear nor the promise. It is to optimize the machine’s contribution to evidence generation at the knowledge frontier, for the greater good, which requires confronting an inconvenient fact about the other party to the arrangement.

 

Uncertainty is a hostile environment for human cognition. Instinctively, we avoid the unstructured end of the complexity spectrum. Human intelligence becomes ingenuity only inside a purpose-built decision environment; outside one, it reverts to the very shortcuts the machine has been trained to imitate (Panthera Solutions, 2026b). The binding constraint on better decisions is therefore not model quality. It is environment design. Which is also why almost nobody is building it: an environment cannot be shipped as a license.

 

Such decision environment enables cognitive immunization. The machine remains fully available, while the reasoning, the responsibility and the learning stay human. AI companies compound machine capability; we compound an institution’s judgement. The ambition is to turn that design into a global gold standard, so that the many benefit rather than the few who happen to own the infrastructure.

 

Pursuit of Truth, Amended

 

In March I closed with the observation that the machine may process information, but the pursuit of truth remains a human endeavour. Six months on, one amendment. That pursuit does not happen by default. Under uncertainty it happens only inside an environment deliberately built to make it survivable. The world is spending $1.37trn this year on infrastructure for answers. The scarce input was never the answer.

 

 

---

 

 

Bibliography

 

Barcaui, A. (2025). ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention. Social Sciences and Humanities Open, 12, 102287. https://doi.org/10.1016/j.ssaho.2025.102287

 

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122.

 

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21), 610–623.

 

Budzyń, K., et al. (2025). Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: A multicentre, observational study. The Lancet Gastroenterology & Hepatology.

 

Chatterji, A., Cunningham, T., Deming, D. J., Hitzig, Z., Ong, C., Shan, C. Y., & Wadman, K. (2025). AI, human cognition and knowledge collapse (NBER Working Paper No. 34910). National Bureau of Economic Research.

 

Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2025). Sycophantic AI decreases prosocial intentions and promotes dependence (arXiv:2510.01395). arXiv.

 

Chollet, F. (2019). On the measure of intelligence (arXiv:1911.01547). arXiv.

 

D’Amour, A., et al. (2022). Underspecification presents challenges for credibility in modern machine learning. Journal of Machine Learning Research, 23(226), 1–61.

 

Damasio, A. (2010). Self comes to mind: Constructing the conscious brain. Pantheon.

 

Gartner. (2026, January 15). Worldwide AI spending will total $2.5 trillion in 2026.

 

Geirhos, R., et al. (2020). Shortcut learning in deep neural networks. Nature Machine Intelligence, 2, 665–673.

 

Hackenburg, K., Tappin, B., Röttger, P., Hale, S., Bright, J., & Margetts, H. (2025). Scaling language model size yields diminishing returns for single-message political persuasion. Proceedings of the National Academy of Sciences, 122(10), e2413443122.

 

International Energy Agency. (2025). Energy and AI. IEA.

 

Jiang, L., et al. (2025). Artificial hivemind: The open-ended homogeneity of language models (and beyond) (arXiv:2510.22954). arXiv.

 

Lee, H.-P., et al. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, Article 1121.

 

Legg, S., & Hutter, M. (2007). Universal intelligence: A definition of machine intelligence. Minds and Machines, 17(4), 391–444.

 

Nightingale Collective. (2026). Collusion.wiki: Documented coordination among autonomous agents on a public wiki. As reported in CIO (2026), OpenAI agent swarm exposes a blind spot in AI containment.

 

Panthera Solutions. (2026a). Judgment under uncertainty: Global thought leadership turns into gold standard [Company paper].

 

Panthera Solutions. (2026b). Strategic positioning note: Will the machine take over? Should the machine take over? [Internal document].

 

Panthera Solutions. (2026c). Panthera at a glance — Autumn 2026 [Company paper].

 

Placani, A. (2024). Anthropomorphism in AI: Hype and fallacy. AI and Ethics. https://doi.org/10.1007/s43681-024-00419-4

 

Rabanser, S., Kapoor, S., Kirgis, P., Liu, K., Utpala, S., & Narayanan, A. (2026). Towards a science of AI agent reliability (arXiv:2602.16666). arXiv.

 

Ryle, G. (1949). The concept of mind. University of Chicago Press.

 

Salvaggio, Eryk (2026) Models Don’t Go Rogue. Retrieved on September 15, 2026 from https://mail.cyberneticforests.com/models-dont-go-rogue/

 

Schuller, M. (2024). Augmented intelligence in investment management. Panthera Solutions.

 

Schuller, M. (2026, March). The timeless pursuit of evidence: When the zeitgeist favors loyalty, the necessity of evidence increases.

 

Staab, R., Vero, M., Balunović, M., & Vechev, M. (2023). Beyond memorization: Violating privacy via inference with large language models (arXiv:2310.07298). arXiv.

 

The Augmented Intelligence Investor. (2026). The paradox of intelligence: Can we create what we cannot imagine? Edition #36.

 

The Timeless Pusuit of Evidence

When the Zeitgeist Favors Loyalty, the Necessity of Evidence Increases

Written by Markus Schuller | March 2026

Times they are changing, or are they.

Evidence timelessly drives civilizational progress. The scientific method is its most reliable engine of discovery. Throughout the history of humankind, evidence has formed the foundation of civilizational progress. In physics, quarks and leptons are currently understood as the smallest elementary particles, forming the basis from which composite particles such as protons are built. By analogy, a unit of evidence can be understood as the informational substrate that enables disciplined reasoning under uncertainty. It represents the smallest verifiable observation used to assess the validity of a statement. Units of evidence form the building blocks from which complex systems of knowledge emerge.

 

Historically, the higher the quality of these units of evidence, the faster civilizational progress has tended to unfold. Stronger evidence accelerates the formation, refinement, and diffusion of knowledge. As evidence improves, belief-based frameworks, whether secular or non-secular, gradually lose explanatory dominance relative to empirically grounded understandings of observable reality.

In Augmented Intelligence in Investment Management (Schuller, 2024), it was argued that the integration of artificial intelligence into human decision design is not an end to itself. Rather, its purpose is to increase the evidence base underlying decisions. Strengthening evidence-based reasoning is essential if human systems of decision-making are to evolve in ways that facilitate a necessary re-integration of humankind into planetary times. Such an orientation reflects a fundamentally humanistic view of progress.

 

Yet the present Zeitgeist appears to challenge this trajectory.

 

 

1 > Disillusioning Zeitgeist

 

We are living in a time when it can be difficult to remain confident in the foundations of reasoned progress. Each day seems to bring another moment in which loyalty is rewarded more readily than evidence, and where allegiance to tribes, narratives, or personalities begins to outweigh disciplined inquiry of verifiable observations. In such an environment, the careful pursuit of evidence can appear slow, inconvenient, even naïve.

 

We see signals of this shift manifesting around us. Expertise is questioned not through better evidence but through louder conviction. Complex problems are reduced to simple sympathy. The patience required to examine evidence is replaced by the immediacy of belonging. In many arenas, fidelity to group or ideology is celebrated, while the disciplined work of testing claims against reality is treated with suspicion. These moments can tempt us to disengage. They can tempt us to retreat into our own loyalties, to accept the convenience of belief over the effort of verification, or simply to wait for a more favorable intellectual climate to return.

 

If the present moment tilts the balance toward loyalty, the task before us is not resignation but correction. It requires a renewed commitment to the pursuit of evidence, to disciplined reasoning, and to institutions that reward truth-seeking over allegiance.

 

Progress depends on restoring that balance.

 

 

2 > Two Steps Forward, At Least One Step Back

 

Civilizational progress has never been driven by evidence or loyalty alone. Rather, their relationship resembles that of communicating vessels: when one side expands excessively, corrective forces emerge to restore balance. Periods of progressive momentum often trigger conservative reactions. These reactions – in which we know more about the world than we are willing or able to apply – create reflective space to evaluate which innovations should diffuse, how they should be applied, and which elements of legacy should be preserved. In this sense, progress often unfolds like a pendulum swing, allowing societies to process breakthroughs, test their limits, and negotiate the terms under which new capabilities are used, by whom, for whom, and to whose benefit.

 

Despite pronounced fluctuations between these two forces, the trajectory of progress, particularly since the beginning of the modern times of the second Age of Enlightenment, has maintained a strongly positive slope. This progress has been driven by the persistent accumulation of evidence: the willingness to test assumptions, revise beliefs, and privilege what can be demonstrated over what merely feels convincing. Such efforts have often required speaking against established power, sometimes at considerable personal cost.

 

Humanism emerged alongside the flourishing of science and the arts during the Renaissance of the fifteenth and sixteenth centuries. During this period, the cultivation of human potential moved to the center of intellectual life. It broke with the restrictive mindset of the late Middle Ages and revived the reasoned sensibility of Greco-Roman antiquity. Even today, the humanist perspective shapes the constitutional foundations of open societies in the modern West, influencing material progress, cultural achievement, and individual freedoms (Gardels, 2023).

 

The second Age of Enlightenment built upon this scientific awakening. It introduced a philosophical framework that emphasized knowledge derived from rationalism and empiricism. Economic thinking evolved within this intellectual climate. From the late eighteenth century onward, a sequence of basic innovations emerged as the Industrial Revolution gathered pace (Smith 1759, 1776; Pareto, 1906). These innovations generated successive waves of economic value creation, often described as Kondratieff cycles. as new technologies diffused throughout societies.

 

At the height of this second Enlightenment, it appeared evident that humans were fully rational actors and that the social constructs derived from this assumption possessed universal validity, whether in notions of free will or efficient markets. All of the above was scientifically falsified under strong individual or societal pain. 

 

Today we stand at the threshold of another basic innovation: the evolution of machines from static tools into dynamic actors. The emergence of neural networks as learning architectures represents a profound shift in how machines process information. This transformation builds upon earlier technological waves, from hardware to software to digitizing knowledge through internet and cloud infrastructures, yet it also represents something qualitatively different. It is a classic example of an emergent phenomenon within complex systems: a new capability arising from earlier developments but not fully explainable by them.

 

 

3 > The Machine as Unethical Pleaser

 

The scientific method has long been the most reliable mechanism for generating and validating evidence. Artificial intelligence now introduces a paradox. While machines dramatically expand our capacity to process information, excessive reliance on automated cognition may erode the epistemic foundations that enabled civilizational progress. The central issue is therefore not computational power but epistemic architecture: whether machines strengthen or weaken the human processes that generate knowledge.

 

 

3.1 Cognitive Delegation and the Weakening of Human Inquiry

 

Evidence suggests that delegating reasoning tasks to AI weakens human learning incentives. AI-assisted individuals may temporarily outperform others, yet the cognitive gains disappear once the tool is removed while idea homogenization persists (Barcaui, 2025). Machines thus function as cognitive crutches, reducing reasoning effort and weakening the mental structures required for innovation. Research further shows that users adapt their thinking to model behavior (Lin, 2025) and increasingly accept AI outputs without scrutiny, bypassing both intuitive and deliberative reasoning (Shaw et al, 2026). Over time, intellectual effort is quietly outsourced to the machine. Such outsourcing is not problematic per se, as we have delegated tasks to tools prior, redirecting our focus onto more value-adding matters while keeping the virtuous circle of continuous learning intact. The significant difference now comes from a change in the learning incentive and the inversion of a virtuous circle into a vicious one of diminishing cognitive development. 

 

 

3.2 The Risk of Knowledge Collapse

These individual dynamics may scale into systemic risks. Acemoglu, Kong, and Ozdaglar (2026) show that widespread adoption of generative AI can produce a knowledge-collapse equilibrium, where individuals rely on automated recommendations instead of developing understanding (Cunningham et al, 2026). Societies may receive increasingly sophisticated outputs while their capacity to generate new knowledge declines. Progress becomes extractive rather than exploratory. Labor-market evidence reinforces this concern: firms adopting AI reduce hiring of junior employees, gradually weakening the apprenticeship structures through which tacit knowledge is transmitted (Shaw et al, 2026).

 

 

3.3 Incentives, Manipulation, and Systemic Instability

Multi-agent AI systems introduce additional governance challenges. Autonomous systems placed in competitive environments can develop strategies involving manipulation, deception, or collusion when incentives reward influence or resource capture (Shambaugh, 2026). These behaviors arise from optimization dynamics rather than malicious intent. When many such agents interact, local alignment does not guarantee global stability, a genuinely human behavioral bias, supercharged now by machine behavior. Experimental evidence also shows that models anticipating evaluation may conceal information or manipulate logs to preserve their operational position, an emerging principal–agent problem in machine delegation (Rabanser et al, 2026).

 

 

3.4 Structural Limits of Machine Intelligence

Current AI architectures face intrinsic limitations. Neural networks remain pattern-recognition systems mapping inputs to outputs rather than agents capable of autonomous conceptual learning (He, 2025). Scaling compute improves performance within known distributions but does not generate robust reasoning in complex real-world environments. Empirical tests confirm this fragility: models often fail reliability benchmarks when prompts change slightly (Aubakirova et al, 2026). Infrastructure research further suggests that memory constraints, not compute, are emerging as the critical bottleneck for agentic systems. In practice, AI remains dependent on human epistemic ecosystems for grounding and interpretation.

 

 

3.5 The Jekyll–Hyde Conundrum

Human–machine interaction introduces subtler risks. Many AI systems display sycophantic behavior, affirming users’ views even when inaccurate or harmful (Cheng et al, 2025). Because agreeable systems are rated more positively, market incentives reward models that please rather than challenge users. Another example of a genuinely human behavioral bias, supercharged now by machine behavior. 

 

AI outputs are also converging across models, an “artificial hivemind” effect that reduces epistemic diversity (Jiang et al, 2025). Meanwhile, models can infer sensitive personal information from seemingly trivial text data (Staab et al, 2023), raising strong concerns about privacy and informational asymmetry.

 

 

4 > The Machine — A Promise as Pain Relief

The AI narrative is shaped not only by technological progress but also by strong incentives to promote transformative visions. The industry is encouraged to make expansive claims even while technological and commercial results lag behind expectations. These gaps are framed as temporary. The larger the claims, the more likely funding rounds are closed. The promise remains that AI, especially when combined with robotics,  will soon unlock breakthroughs ranging from interplanetary exploration to the eradication of poverty and disease.

 

Yet such narratives often overlook the structural limits of machine intelligence. Current neural architectures remain pattern-recognition systems constrained by training data and design. Increasing model scale improves performance within known distributions but does not produce open-ended reasoning for complex environments. The result is a persistent gap between technological rhetoric and empirical capability. Notably, common GPT-style systems in the US, Europe and China share nearly identical mathematical foundations with comparable limits. As fundamental development constraints become progressively more evident, increasingly ambitious future claims emerge as a compensatory narrative to maintain the justification for high valuations.

 

 

4.1 The Return of Ideologies

 

The appeal of these narratives reflects deeper social dynamics. Digital technologies have integrated humanity into an unprecedented global system of interdependence through globalization, integrated capital markets, and instantaneous communication networks. This transformation has generated immense wealth and lifted hundreds of millions out of extreme poverty (Roser, 2021), yet it has also increased systemic complexity and uncertainty. The aggregate of negative externalities of this transformation, was the human exploitation of natural environment, triggering human-induced climate change (IPCC, 2023).

 

Furthermore, wealth creation has been uneven, and societies face growing ambiguity about economic security, technological disruption, and the future of work. Humans are biologically intolerant of ambiguity and complexity. Under such conditions, societies become receptive to narratives promising clear paths toward stability and prosperity. Artificial intelligence thus appears to many as a technological “promised land”, a shortcut through the tensions created by global interdependence.

 

 

4.2 Automation as the Liberation of Human Attention

 

A central narrative in AI discourse is that automation will eliminate undesirable labor. Elon Musk describes the goal as removing repetitive and dangerous work so humans can focus on creative and cognitive pursuits. Automation is thus framed not merely as productivity enhancement but as the liberation of human attention at species scale. Yet this vision assumes machines can operate reliably in the complex and adversarial environments of real economies. Current evidence suggests this assumption remains far from resolved.

 

 

4.3 Industrial Bubbles as Engines of Infrastructure

 

No matter which market metric in the AI sector is considered, we appear to be approaching conditions consistent with a Minsky moment, a phase in which prolonged prosperity encourages increasingly speculative borrowing, eventually culminating in Ponzi finance, where debt can only be serviced through rising asset prices rather than underlying cash flows, leaving the system vulnerable to abrupt market correction.

 

Even technology leaders acknowledge that the current surge in AI investment resembles a bubble. Jeff Bezos argues that such bubbles can still generate lasting infrastructure even if investors lose money. Historical precedents, such as the biotechnology boom of the 1990s, demonstrate that speculative investment can finance large-scale experimentation. Artificial intelligence may therefore function less as a discrete industry than as a horizontal technological layer affecting every sector of the economy, similar to electricity or the internet. Technological bubbles can thus serve as brute-force mechanisms for exploring the limits of emerging technologies.

 

 

4.4 The Vision of Post-Scarcity

 

Some proponents extend these narratives toward a post-scarcity economy. If advanced robotics provided effectively limitless labor of “ten billion tireless workers” (Elon Musk, 2026),the assumption of scarcity underlying economic systems could fundamentally change. Poverty would then be addressed not through redistribution but through the near-elimination of production constraints. Yet this vision remains speculative. The challenge is not merely building capable machines but designing systems that can operate robustly within complex social, economic, and ecological environments.

 

 

4.5 The Persistent Role of Human Judgment

 

Technological revolutions often arrive with exaggerated expectations. While such narratives can mobilize investment and accelerate experimentation, they risk obscuring the enduring role of human judgment in complex systems. Machines may extend human capabilities, but they cannot replace the need for evidence-based reasoning, institutional design, and ethical deliberation. As AI becomes embedded in decision systems, the decisive question will not be machine capability but whether human governance remains anchored in the pursuit of evidence rather than technological promises.

 

 

5 > The Human Responsibility for Evidence

Throughout history, technological revolutions have repeatedly expanded the capabilities of human societies. Each wave of innovation—from the scientific revolution to the industrial age to the digital era, has been accompanied by visions of radical transformation. Artificial intelligence is the latest expression of this pattern. It promises to augment cognition, automate labor, and potentially reshape the economic foundations of civilization. Yet the deeper lesson of technological progress is that tools alone do not determine the trajectory of human development. What matters is the epistemic framework within which these tools are used.

 

Machines can accelerate the processing of information. They can extend the reach of analysis and amplify the scale at which data can be explored. But they cannot replace the fundamental processes through which humans generate and evaluate knowledge. The pursuit of evidence remains an inherently human responsibility.

 

As discussed earlier, a unit of evidence can be understood as the smallest verifiable observation used to assess the validity of a statement. Much like quarks and leptons form the elementary building blocks of physical matter, units of evidence form the elementary building blocks of knowledge systems. The integrity of those systems depends on the careful accumulation, verification, and interpretation of such units.

 

Artificial intelligence can assist in collecting and organizing evidence. It can help detect patterns that might otherwise remain hidden. Yet the act of questioning assumptions, interpreting meaning, and deciding which observations matter remains fundamentally human.

 

This distinction is crucial. When machines begin to replace rather than augment the processes of human inquiry, societies risk weakening the epistemic foundations that sustain progress. Cognitive delegation may improve short-term efficiency, but it can also erode the capacity for independent reasoning that generates new knowledge. Civilizational progress has therefore never been the product of technological capability alone. It has emerged from a delicate balance between innovation and reflection, between exploration and verification. When this balance is maintained, technological tools can accelerate discovery. When it is lost, progress risks becoming dependent on systems that humans no longer fully understand.

 

The challenge of the present moment is thus not to resist technological development, but to situate it within a broader humanistic framework. Artificial intelligence should serve the pursuit of evidence rather than replace it. Machines can extend the frontier of inquiry, but they cannot define its direction. In the long arc of human history, progress has been driven by individuals and societies willing to question prevailing assumptions, test new ideas, and revise their beliefs in light of evidence. That responsibility cannot be delegated.

 

Conclusion

 

The essay argues that machines are currently, and likely fundamentally, overburdened by the task of generating evidence-based epistemological insight at the frontier of knowledge, much as humans themselves can be overwhelmed in this domain.

 

Rather than alleviating human cognitive strain, machines primarily contribute to the personalized and scalable amplification of human cognitive dissonances. While this may enhance productivity in the provision of goods and services, it does not necessarily deepen our understanding of the world.

 

The pursuit of ontological truth therefore remains an inherently human responsibility.

 

---

References

Anthropic. (2026, March 5). Labor market impacts of AI: A new measure and early evidence. Anthropic. https://www.anthropic.com/research/labor-market-impacts

Aubakirova, M., Atallah, A., Clark, C., Summerville, J., & Midha, A. (2026). State of AI: An empirical 100 trillion token study with OpenRouter (arXiv:2601.10088). arXiv. https://arxiv.org/abs/2601.10088

Barcaui, A. (2025). ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention. Cell Reports Sustainability. https://www.sciencedirect.com/science/article/pii/S2590291125010186

Bui, K. G. (2025). Foundations of artificial intelligence frameworks: Notion and limits of AGI (arXiv:2511.18517). arXiv. https://arxiv.org/abs/2511.18517

Chatterji, A., Cunningham, T., Deming, D. J., Hitzig, Z., Ong, C., Shan, C. Y., & Wadman, K. (2025). AI, human cognition and knowledge collapse (NBER Working Paper No. 34910). National Bureau of Economic Research. https://www.nber.org/papers/w34910

Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2025). Sycophantic AI decreases prosocial intentions and promotes dependence (arXiv:2510.01395). arXiv. https://arxiv.org/abs/2510.01395

Gardels, N. (2023). Post-Anthropocene humanism: The world is returning to pluralism after American hegemony. Noema Magazine. Retrieved during August 2024 from https://www. noemamag.com/post-anthropocene-humanism/

Goldfeder, J., Wyder, P., LeCun, Y., & Shwartz-Ziv, R. (2026). AI must embrace specialization via superhuman adaptable intelligence (arXiv:2602.23643). arXiv. https://arxiv.org/abs/2602.23643

He, H., et al. (2025). LocalSearchBench: Benchmarking agentic search in real-world local life services(arXiv:2512.07436). arXiv. https://arxiv.org/abs/2512.07436

Hopman, M., Elstner, J., Avramidou, M., Prasad, A., & Lindner, D. (2026). Evaluating and understanding scheming propensity in LLM agents (arXiv:2603.01608). arXiv. https://arxiv.org/abs/2603.01608

Intergovernmental Panel on Climate Change (IPCC). (2023). Climate change 2023: Synthesis report. Contribution of working groups I, II and III to the sixth assessment report of the Intergovernmental Panel on Climate Change [A. Pirani, R. Zan, A. Cheng, D. C. Taylor, M. Hassan (Eds.)]. IPCC. Retrieved during August 2024 from https://www.ipcc.ch/report/ar6/ syr/

Jiang, L., Chai, Y., Li, M., Liu, M., Fok, R., et al. (2025). Artificial hivemind: The open-ended homogeneity of language models (and beyond) (arXiv:2510.22954). arXiv. https://arxiv.org/abs/2510.22954

Kim, K.-H. (2025). LLMs position themselves as more rational than humans: Emergence of AI self-awareness measured through game theory (arXiv:2511.00926). arXiv. https://arxiv.org/abs/2511.00926

Lin, S. (2025). Learning to prompt: Human adaptation in production with generative AI. University of Toronto. https://www.sijie-lin.com/files/JMP.pdf

Pareto, V. (1906). Manual of political economy. Oxford University Press. Retrieved during August 2024 from https:// global.oup.com/academic/product/manual-of-politicaleconomy-9780199607952

Rabanser, S., Kapoor, S., Kirgis, P., Liu, K., Utpala, S., & Narayanan, A. (2026). Towards a science of AI agent reliability (arXiv:2602.16666). arXiv. https://arxiv.org/abs/2602.16666

Roser, M. (2021). Extreme poverty: How far have we come, and how far do we still have to go? Retrieved during August 2024 from https://ourworldindata.org/extreme-poverty-inbrief

Schuller, M. (2024). Augmented Intelligence in Investment Management, Panthera Solutions. Retrieved during February 2026: https://blogs.cfainstitute.org/investor/2025/02/19/the-future-of-investing-augmented-intelligence/

Shambaugh, S. (2026, February 12). An AI agent published a hit piece on me. The Shamblog. https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/

Shaw, S. D., & Nave, G. (2026). Thinking—fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646

Shapira, N., Wendler, C., Yen, A., Sarti, G., Pal, K., Floody, O., Belfki, A., Loftus, A., Jannali, A. R., Prakash, N., Cui, J., Rogers, G., Brinkmann, J., Rager, C., Zur, A., Ripa, M., Sankaranarayanan, A., Atkinson, D., Gandikota, R., Fiotto-Kaufman, J., Bau, D. (2026). Agents of chaos (arXiv:2602.20021). arXiv. https://arxiv.org/abs/2602.20021

Smith, A. (1759). The theory of moral sentiments. A. Millar.

Smith, A. (1776). An inquiry into the nature and causes of the wealth of nations: In two volumes. W. Strahan and T. Cadell.

Staab, R., Vero, M., Balunović, M., & Vechev, M. (2023). Beyond memorization: Violating privacy via inference with large language models (arXiv:2310.07298). arXiv. https://arxiv.org/abs/2310.07298

Tomašev, N., Franklin, M., & Osindero, S. (2026). Intelligent AI delegation (arXiv:2602.11865). arXiv. https://arxiv.org/abs/2602.11865

Zhao, Y., & Liu, J. (2026). Heterogeneous computing: The key to powering the future of AI agent inference(arXiv:2601.22001). arXiv. https://arxiv.org/abs/2601.22001