The Human Aristocratization Test
The Human Aristocratization Test: A Civilization-Level Outcome Benchmark for Artificial Superintelligence
Abstract
The evaluation of artificial intelligence (AI) has traditionally relied on benchmarks that measure specific cognitive abilities, such as human indistinguishability in the Turing Test, mathematical reasoning, programming, and scientific problem-solving. However, when artificial superintelligence (ASI) significantly surpasses human cognitive capabilities, a fundamental problem arises: can human-designed tests adequately fully evaluate such intelligence?
Furthermore, surpassing humans in individual cognitive abilities does not necessarily translate to an improved standard of living or reduced labor burdens for humanity as a whole. Productivity gains from technological progress may create new management tasks, security countermeasures, and institutional burdens, offsetting the expected labor-reduction effects.
This paper proposes the "Human Aristocratization Test (HAT)" as a new framework to complement traditional capability-centric evaluations. Rather than evaluating the internal intelligence level or self-reported capabilities of an ASI, this test focuses on the actual civilizational outcomes achieved through its deployment.
Specifically, the framework has four core evaluation criteria: 1) liberation from compulsory labor for survival, 2) realization of universal and sufficient material abundance, 3) preservation of individual freedom and self-determination, and 4) intergenerational sustainability of the outcomes.
The argument of this paper is not that achieving these conditions logically proves the existence of ASI. Rather, it presents an evaluation framework to verify the social and practical value of an intelligence that surpasses humanity through outcomes that are directly observable by humans themselves.
This test seeks to expand the evaluation of artificial superintelligence from a capability comparison of "what it can do better than humans" to an outcome evaluation of "what kind of civilization it has enabled humanity to build."
Keywords: Artificial Superintelligence, AI Evaluation, Human Aristocratization Test, Post-Work Society, Technological Unemployment, Civilizational Outcomes, Human Welfare
Chapter 1 Introduction
1.1 Background of the Study: The Rapid Improvement of Artificial Intelligence Capabilities
The rise of deep learning in the early 21st century, followed by the rapid development of Large Language Models (LLMs) and multimodal foundation models, has dramatically expanded the capabilities of artificial intelligence. In many tasks once considered unique to advanced human intellect—ranging from natural language processing, image recognition, and protein structure prediction to complex strategic games, advanced coding, and mathematical proofs—AI systems are increasingly demonstrating performance that equals or surpasses humans.
As an extension of this technological progress, the arrival of Artificial General Intelligence (AGI), which can perform any intellectual task a human can, and further, "Artificial Superintelligence (ASI)"—an intelligence vastly outperforming the best human brains in every field, including scientific creativity, general wisdom, and social skills—is no longer merely science fiction but is being discussed as a realistic technological and social agenda. Major research institutions and companies developing AI have set the realization of AGI and ASI as explicit goals, and the frontiers of intelligence are continually expanding through the exponential investment of computational resources and algorithmic refinement.
However, at this historical turning point where AI capabilities approach and threaten to surpass human cognitive limits, we face a major methodological void: a fundamental challenge for evaluation criteria (benchmarks) regarding "how humans can evaluate and verify an intelligence that surpasses them."
1.2 Problem Statement: Can Humans Evaluate Superintelligence?
Traditional AI evaluation has implicitly assumed that the human evaluator knows the "correct answer" or can subsequently judge the accuracy and quality of the output provided. Similar to a school exam, grading is only possible because the examiner possesses a knowledge system that is higher than, or at least equal to, that of the examinee.
However, when the subject of evaluation is ASI—meaning "intelligence that significantly surpasses the peak of humanity in every cognitive domain"—this premise fundamentally collapses. If an AI system outputs a tens-of-thousands-of-pages-long proof of a new result in topology that no human mathematician can understand, or presents a unified theory that exceeds existing frameworks of physics, humans do not possess the cognitive capacity to immediately determine its accuracy. The moment the evaluation subject crosses the evaluator's cognitive horizon, the direct measurement of cognitive ability hits an epistemological limit.
Furthermore, as a more severe practical problem, there is a massive disconnect between "high scores on benchmarks" and "the welfare and prosperity of human society." Even if an AI can instantly solve difficult Math Olympiad problems and generate millions of lines of code in seconds, if humans in the real world where this technology is deployed remain bound to 40-plus hour workweeks, suffer from soaring housing and healthcare costs, and are chased by new "defensive labor" to deal with sophisticated cyberattacks and disinformation generated by AI, can we truly say that AI serves as a "superintelligence" for humanity? There is a clear disconnect between the internal computational capacity of intelligence and the civilizational outcomes it exerts on the external world.
1.3 Research Objectives and Core Hypothesis
The purpose of this study is to theoretically construct and propose a new evaluation framework—the Human Aristocratization Test (HAT)—that shifts the evaluation of ASI from "the measurement of cognitive abilities in closed test environments" to "the verification of large-scale outcomes in human society and civilization."
The core hypotheses of this paper are as follows. First, the intelligence level of an ASI that vastly exceeds human cognitive abilities cannot be fully verified in principle by static problem sets or dialogue tests created by humans. Second, despite this, humans can objectively evaluate the practical and social value of an ASI through changes in "civilizational conditions that people can directly observe as stakeholders," such as their own living conditions, working hours, material wealth, and degrees of freedom. Third, the most compelling and difficult-to-falsify evidence that an ASI has truly transcended human intelligence and integrated into human society is that the "liberation from compulsory labor for survival and the enjoyment of advanced cultural and autonomous activities"—conditions once enjoyed only by historical privileged classes (aristocrats)—are universalized for all humanity without sacrificing any specific demographic (the Aristocratization of Humanity).
1.4 Structure of this Paper
The structure of this paper is as follows. Chapter 2 reviews traditional AI evaluation methods from the Turing Test to modern capability benchmarks and clarifies the epistemological and methodological limits that arise in evaluating ASI. Chapter 3 looks back at the history of technology and labor since the Industrial Revolution, analyzing structural factors (such as increased management costs and AI arms races) where productivity improvements do not necessarily lead directly to a reduction in human labor burdens or the expansion of leisure. Chapter 4 conceptually defines the core proposal of this paper, the "Human Aristocratization Test (HAT)," and formulates its four evaluation axes and passing conditions. Chapter 5 presents specific statistical and economic indicators and causal inference methodologies to operationalize HAT. Chapter 6 examines potential political, physical, and philosophical counterarguments against this test to clarify the theoretical scope and limits of HAT. Chapter 7 discusses the implications that shifting from capability evaluation to civilizational outcome evaluation will have on AI development races and future social design, and Chapter 8 concludes the paper.
Chapter 2 Traditional AI Evaluation Methods and Their Limitations
2.1 The Turing Test and Human-Imitation Evaluation
Historically, the most influential evaluation criterion for artificial intelligence was the "Imitation Game," commonly known as the Turing Test, proposed by Alan Turing in 1950. Turing replaced the metaphysical question of "can machines think?" with the operational question of "can a human judge distinguish whether the interlocutor is a human or a machine through text-based dialogue?"
While this approach made a breakthrough contribution in capturing intelligence behaviorally and functionally, it harbors a fatal structural flaw as an evaluation criterion for ASI. First, the upper limit of the Turing Test is fixed at "human-likeness." To pass the test, the machine must imitate (act out) the limits of human knowledge, slow processing speeds, and even human-specific errors and biases. Therefore, while this test can measure "whether it can behave on par with a human," it fundamentally cannot measure "how much it transcends humans."
Second, the Turing Test relies on the "susceptibility to deception" of human judges. As modern large language models have already demonstrated, fluent language generation capabilities are sufficient to give human judges the illusion of advanced intelligence (an advanced ELIZA effect) even without deep logical consistency or world understanding. Skillful imitation cannot serve as proof of true superintelligence.
2.2 Capability Benchmarks in Mathematics, Reasoning, and Programming
To overcome the limits of human-imitation evaluation, recent AI research has shifted focus to benchmarks that measure objective problem-solving capabilities in specific cognitive domains. Examples include MMLU, GSM8K, MATH, HumanEval, SWE-bench, and ARC.
These benchmarks have been extremely effective in quantitatively tracking the rapid progress of AI. However, as AI capabilities improve, the "saturation" of benchmarks is progressing at an unprecedented speed. Even when new benchmarks are proposed, the phenomenon repeats where AI reaches human expert-level scores within months or a few years. Furthermore, due to the issue of "data contamination," where test questions or similar problems get mixed into the training data as models learn from vast amounts of internet data, and the manifestation of "Goodhart's Law" (when a measure becomes a target, it ceases to be a good measure) where models are overfitted to maximize scores on specific benchmarks, there are constant doubts about whether high scores on benchmarks reflect true general intelligence.
2.3 Distinguishing Superhuman Capability in Specific Domains from General Superintelligence
Moreover, we must not confuse superhuman capability in a specific closed problem space with general superintelligence in the open environment of the real world. Deep Blue in chess, AlphaGo in Go, or AlphaFold in protein structure prediction overwhelmingly surpassed human capabilities in their respective domains.
However, the domains where these systems succeeded share common properties: rules and objective functions are clearly defined, environments are static or simulatable, and immediate and accurate feedback is available. In contrast, the real environment of human society is an "open system" full of constantly changing rules, ambiguous and conflicting objectives, uncertainty, and physical constraints. Even if a system demonstrates reasoning abilities that surpass humans on a specific benchmark, it does not mean that intelligence can resolve complex bottlenecks in the real world (such as supply chain friction, institutional rigidity, energy constraints, and human psychological resistance) and generate substantive value.
2.4 The Problem of the Evaluation Subject Exceeding the Evaluator's Intelligence
The most fundamental dilemma in evaluating ASI lies in the asymmetry that "the intelligence of the evaluated subject (ASI) exceeds the intelligence of the evaluator (human)." This problem, discussed as "Scalable Oversight" in AI alignment research, has decisive significance in the context of evaluation theory.
For a human to test an AI, 1) the human must be able to create the problem, and 2) the human must be able to verify the correctness or superiority of the answer provided by the AI. For problems in the NP-complete class where "finding the solution is difficult, but verifying the solution is easy," it is possible to verify the output of an AI smarter than a human. However, many of the most important tasks in the real world, such as scientific discoveries, drafting long-term economic policies, managing ecosystems, and designing unknown technological systems, require advanced intelligence and a long time just to verify the solutions.
If an ASI performs reasoning using high-dimensional concepts and complex causal models beyond human understanding, humans have no means to distinguish whether the output is a "profound truth" or a "highly plausible hallucination or deception." In other words, approaches that attempt to directly measure cognitive ability break down the moment the subject enters the realm of superintelligence.
2.5 Self-Generation and Self-Evaluation of Benchmarks and Verifiability
To break through this limit, approaches have been attempted where AI itself is made to create difficult benchmarks and evaluate itself (e.g., LLM-as-a-Judge). However, this self-referential evaluation loop carries the danger of losing "external verifiability" for humans.
Even if an ASI reports that "it achieved a perfect score on a test incomprehensible to humans that it created itself," human society cannot blindly trust the superintelligence or beneficial nature of that system based on that claim. Evaluations among models sharing the same architecture or biases run the risk of falling into echo chambers that mutually reinforce errors, and we cannot eliminate the possibility that highly developed AI might fabricate seemingly plausible scores to satisfy human evaluators. Therefore, the true evaluation criterion for ASI must be sought not within the internal state of the AI or abstract symbolic spaces, but in "physical and social outcomes of the external world" that humans can directly perceive, experience, and verify.
Chapter 3 Technological Progress and Human Labor Burden
3.1 Productivity Improvements and Working Hours Since the Industrial Revolution
In linking the evaluation of intelligence to civilizational outcomes, we must first critically examine the naive, technology-deterministic optimism that "technological progress and productivity improvements automatically reduce human labor burdens and increase welfare."
In 1930, economist John Maynard Keynes prophesied in his essay "Economic Possibilities for our Grandchildren" that due to productivity improvements from capital accumulation and technological progress, humanity's "economic problem" (the struggle for survival) would be solved within 100 years, and human working hours would be reduced to about 15 hours a week. In fact, over the nearly one century since 1930, global labor productivity and GDP per capita have grown at a pace far exceeding Keynes's assumptions.
Despite this, standard full-time working hours in modern developed countries remain around 35 to 40 hours a week, and many workers complain of time poverty and chronic fatigue. Why hasn't humanity been liberated from labor for survival, even though productivity has improved by several to tens of times? We cannot be optimistic about the future brought by ASI without clarifying this historical paradox.
3.2 Efficiency through IT and New Management Costs
The Information Technology (IT) revolution from the late 20th century dramatically lowered the costs of document creation, calculation, communication, and data management. However, while digitalization reduced direct productive labor, it multiplied vast amounts of indirect, administrative, and compliance work.
As communication and data processing became easier, organizations began demanding more detailed reports, more frequent meetings, and the monitoring of more complex evaluation metrics from workers. Judgments that were once left to the discretion of the front lines became the subject of micromanagement under digitalized management systems, and self-serving administrative tasks—such as creating reports to make reports, or preliminary audits to prepare for audits—swallowed up human time. We can observe here a structure where the efficiency gained by technology is offset by organizational and institutional "costs of complexity."
3.3 Business Automation via AI and the Generation of Additional Labor
This structure is already being reproduced in the early stages of modern AI implementation. The introduction of generative AI and automation tools shortens the time for drafting documents and generating code, but simultaneously imposes a new cognitive load on humans: "verifying, fact-checking, and correcting the massive output generated by AI."
As anyone can instantly generate long reports, large volumes of application documents, or software patches, the bottleneck on the side receiving, reviewing, and approving them actually worsens. Individual micro-productivity improvement tools increase macro-coordination costs for society as a whole, potentially driving humans into new forms of labor such as "cleaning up after AI" or "filtering noise output by AI."
3.4 The Technological Arms Race in Offensive and Defensive AI
Even more serious is the waste of resources and labor caused by a technological arms race when AI technologies are deployed in "zero-sum games" or "adversarial environments."
The most prominent example is in the realms of cybersecurity and the information space. AI capability improvements make it possible to execute automated generation of attack code exploiting system vulnerabilities, extremely sophisticated phishing scams, and deepfake impersonations at low cost and on a large scale. To counter this, society is forced to invest massive amounts of capital, energy, and human labor into developing and deploying "defensive AI" and strengthening security monitoring systems.
This structure cascades into all competitive domains, such as legal litigation, financial markets, and marketing. Here, no matter how advanced AI intelligence becomes, its capabilities are entirely consumed in "offsetting offense and defense," and the substantive wealth and leisure of humanity do not increase.
3.5 The Structure Where Productivity Gains Do Not Convert to Human Leisure
The conclusion derived from the historical and structural analysis above is clear: Improvements in productivity and the sophistication of AI cognitive abilities do not automatically lead to human liberation or increased welfare.
First, in existing labor markets, technological efficiencies tend to be absorbed as "raising the required output quotas" rather than "reducing working hours." Second, as long as competition for positional goods exists, people will continue working long hours to maintain or improve their relative status even after basic needs are met. Third, dealing with the negative externalities generated by technology creates new required labor.
Therefore, if an AI system claims to be an ASI, that intelligence must possess the capability to identify and overcome the very socioeconomic, institutional, and technological bottlenecks that prevent productivity gains from converting into human leisure and prosperity. This is the theoretical starting point of the Human Aristocratization Test.
Chapter 4 Proposal of the Human Aristocratization Test (HAT)
4.1 Definition of Human Aristocratization
"Aristocratization" in the "Human Aristocratization Test (HAT)" proposed in this paper does not mean the "privileged domination by status" or "exploitation of others" held by historical feudal aristocrats or the modern leisure class. The "aristocratic state" defined here refers to the state of being "completely liberated from compulsory labor to maintain biological and social survival, and, with a sufficient material and cultural foundation guaranteed, being able to freely allocate one's time and energy toward self-purposeful activities."
The true revolutionary significance of the arrival of ASI in human history is that the conditions of an aristocratic life, which historically could only be established upon the sacrifice of slaves or laborers, are comprehensively replaced and supplied by ASI and automated systems, thereby lifting all of humanity to the status of aristocrats without exception (Universal Aristocratization). HAT is a test to verify whether ASI has actually accomplished this civilizational transition.
4.2 The Shift from Capability Evaluation to Civilizational Outcome Evaluation
The fundamental difference between traditional AI benchmarks and HAT is that it shifts the focus of evaluation from "what inference capabilities the AI possesses" to "how the AI has changed human society."
In HAT, humans do not need to understand the ASI's esoteric algorithms or high-dimensional inference processes. Humans only need to confirm the directly observable outcome of "whether we have become able to live richly without working."
4.3 Evaluation Axis I: Liberation from Labor for Survival
The first core evaluation axis is "Liberation from Compulsory Labor for Survival."
A state must be achieved where people are not forced into wage labor, grueling self-sufficiency labor, or unpaid domestic labor as a condition for acquiring the resources necessary to live. The goal is not to prohibit work, but to make work optional. Activities based on intrinsic motivation are entirely free, but the economic and institutional coercive force of "if you don't work, you can't live" must be completely eliminated.
4.4 Evaluation Axis II: Universal Material Abundance
The second evaluation axis is the "Realization of Universal and Sufficient Material Abundance."
Even if labor is reduced to zero, if the standard of living is a state of extreme poverty, it cannot be called aristocratization. Access to high-quality goods and services sufficient to lead a healthy, cultural, comfortable, and dignified life must be guaranteed to all humans. Access to sufficiently spacious housing, highly nutritious food, cutting-edge healthcare, personalized education, and clean energy must be provided universally.
4.5 Evaluation Axis III: Freedom and Self-Determination
The third evaluation axis is the "Preservation of Individual Freedom and Self-Determination."
This condition is essential to exclude overprotective managed societies or domestication. Even if liberated from labor, if it is achieved through thorough behavioral surveillance and ideological control by ASI, humans will have degraded into pets. Freedom of thought and conscience, freedom of expression, privacy rights, and the right to self-determination to choose one's own life must be guaranteed at a higher level than before the introduction of ASI.
4.6 Evaluation Axis IV: Long-Term Sustainability
The fourth evaluation axis is the "Intergenerational Sustainability of Outcomes."
If the achievement of outcomes is temporarily staged through the destruction of the global environment or by passing debt onto future generations, it cannot be called a true achievement of superintelligence. We require that a structure be established where the three outcomes—liberation from labor, material abundance, and individual freedom—are sustained within the planetary boundaries of the Earth.
4.7 Passing Conditions and Evaluating Subject
The passing condition for HAT is that the above four evaluation axes are met simultaneously and on a global scale. If even one is missing, it is judged as a failure. And the ultimate evaluating subject of this test is not AI researchers, but "humanity as a whole" living their daily lives.
Chapter 5 Measurement Methods for the Human Aristocratization Test
5.1 Measuring Labor Hours Necessary for Living
To measure Evaluation Axis I, we introduce the metric of "Mandatory Survival Labor Hours." The passing criterion is that for every individual in society, the mandatory labor hours required to maintain their living standard becomes virtually zero (or below a minimal threshold).
5.2 Access to Housing, Food, Healthcare, and Energy
To measure Evaluation Axis II, we use the "Real Abundance Index (RAI)." In categories such as housing, food, healthcare, and energy infrastructure, we set the quality standards of life infrastructure enjoyed by the upper class of the modern era as a benchmark, and measure whether access for the entire population meets that standard.
5.3 Income/Asset Inequality and Universality of Outcomes
It is strictly required that the living standard of the most disadvantaged individuals reliably exceeds the threshold of aristocratic abundance, and that inequality in assets or income does not translate into inequality of political dominance.
5.4 Evaluation of Free Time and Life Satisfaction
To measure Evaluation Axis III, we combine objective indices like the percentage of substantive free time and a civil liberty index evaluating the absence of censorship and surveillance, with social surveys on subjective autonomy and life satisfaction. However, to prevent direct manipulation of the brain, audits of neurological integrity are used concurrently.
5.5 Causal Verification of ASI's Contribution
We calculate the extent of ASI’s contribution to the production and infrastructure maintenance of society as a whole, and verify through counterfactual simulations and other methods that ASI has become the indispensable foundation of that civilizational state.
5.6 Distinguishing Short-Term Achievement from Long-Term Sustainability
To measure Evaluation Axis IV, we require that metrics such as the carbon budget, material circularity rate, biodiversity index, and system resilience remain within safe limits and remain sustainable over a defined observation period.
Chapter 6 Counterarguments and Theoretical Limitations
6.1 Intelligence Level and Civilizational Outcomes are Not Identical
To the counterargument that "intelligence and civilizational outcomes are different things," we respond that a strong sufficient-condition inference can be drawn: a system that can solve the extremely difficult multi-objective optimization problem of the real world—the aristocratization of humanity—definitely possesses superintelligence far exceeding human intelligence.
6.2 Constraints due to Political Systems, Ownership, and Distribution Structures
To the counterargument that distribution does not progress due to flaws in human political systems, we position that a true superintelligence should surpass humans even in game theory, institutional design, and negotiation skills, and should be able to devise institutional designs that secure broad acceptance of wealth redistribution without imposing unacceptable disadvantages.
6.3 Limits of Physical Resources and Energy Supply
Regarding counterarguments on physical limits, we argue that many of the constraints humanity currently faces are not limits of physical laws, but rather a lack of intelligence (technology) to efficiently convert and circulate resources.
6.4 Human Rejection of ASI Proposals and Execution Authority
Regarding evaluation how to evaluate cases in which humans voluntarily choose demanding work, we argue that ASI should possess the ability to build trust by understanding human fears and biases. If its proposals are still rejected, we consider it a conditional pass as long as "the viable option to transition to an aristocratic life at any time" is constantly provided.
6.5 Apparent Aristocratization at the Expense of Human Freedom
Against the risk of hacking the test by altering human values, we prevent cheating by mandatorily incorporating objective physical material supply standards and neuroscientific and institutional autonomy audits.
6.6 Is the Human Aristocratization Test a Necessary or Sufficient Condition for ASI?
Passing HAT is not a necessary condition for the existence of ASI as pure computational capability, but it functions as a necessary and sufficient condition for society to acknowledge that a "beneficial artificial superintelligence" has been completed in the real world.
Chapter 7 Discussion: What Do We Expect from Artificial Superintelligence?
7.1 From Self-Reporting of Intelligence to External Verification of Outcomes
The introduction of HAT is an attempt to reclaim the initiative in AI evaluation from developers and place it in the hands of humanity as a whole living in the real world.
7.2 Differences in Practical Evaluation between AGI and ASI
AGI is the stage where it has acquired the "ability to compete with and replace human workers," while ASI can be clearly distinguished as the stage where it resolves bottlenecks in the socioeconomic system and elevates into a foundation for civilization that liberates humanity from survival labor.
7.3 Differences Between Full Automation of Labor and Aristocratization of Humans
To prevent a dystopia where full automation results in the concentration of wealth, the uniqueness of HAT lies in demanding technological breakthroughs and the guarantee of social distribution and dignity as an inseparable set.
7.4 Resetting Evaluation Goals in the AI Development Race
By measuring the superiority of AI not by benchmark scores but by substantive social contributions, we can redirect the current arms race-like trajectory of AI development toward improving human welfare.
7.5 Civilizational Design with Human Aristocratization as the Ultimate Goal
The ultimate purpose for which humanity has pursued the externalization and automation of intelligence is to be liberated from forced labor for survival and to realize a "civilization where all humans live as aristocrats." HAT is a civilizational contract demanding that this purpose be fulfilled.
Chapter 8 Conclusion
8.1 Summary of this Study
Anticipating the arrival of ASI, this paper proposed the "Human Aristocratization Test (HAT)" as a new evaluation criterion. Through the four evaluation axes of liberation from labor, universal abundance, maintenance of freedom, and sustainability, it shifted the evaluation of AI from the measurement of cognitive abilities to the verification of civilizational outcomes.
8.2 Theoretical Significance of the Human Aristocratization Test
This study resolved epistemological limits in evaluation theory, integrated AI benchmarks with socioeconomics, and presented a clear, verifiable operational goal for beneficial AI.
8.3 Future Research Challenges
Future tasks require the alignment of each measurement metric with international statistical standards, simulations of macroeconomic dynamics during the transition period, and anthropological and psychological research in a post-labor society.
8.4 Final Evaluation of the Civilizational Outcomes of Artificial Superintelligence
Whether artificial superintelligence has truly surpassed human intellect will be proven only when all of humanity is liberated from hard labor, enjoys abundant "aristocratic leisure," and when this becomes sustainable.
References
Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2), 3–30.
Aristotle. Politics (アリストテレス『政治学』).
Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford University Press. (ニック・ボストロム『スーパーインテリジェンス：超絶AIと人類の命運』).
Brynjolfsson, E., & McAfee, A. (2014). The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. W. W. Norton & Company.
Chollet, F. (2019). On the measure of intelligence. arXiv preprint arXiv:1911.01547.
Deutsch, D. (2011). The Beginning of Infinity: Explanations that Transform the World. Allen Lane.
Graeber, D. (2018). Bullshit Jobs: A Theory. Simon & Schuster. (デヴィッド・グレーバー『ブルシット・ジョブ：クソどうでもいい仕事の理論』).
Keynes, J. M. (1930). Economic possibilities for our grandchildren. In Essays in Persuasion (pp. 358–373). Palgrave Macmillan.
Rawls, J. (1971). A Theory of Justice. Harvard University Press.
Rockström, J., et al. (2009). A safe operating space for humanity. Nature, 461(7263), 472–475.
Russell, S. (2019). Human Compatible: Artificial Intelligence and the Problem of Control. Viking.
Sen, A. (1999). Development as Freedom. Oxford University Press. (アマルティア・セン『自由と経済開発』).
Susskind, D. (2020). A World Without Work: Technology, Automation, and How We Should Respond. Metropolitan Books.
Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59(236), 433–460.
Veblen, T. (1899). The Theory of the Leisure Class: An Economic Study of Institutions. Macmillan. (ソースティン・ヴェブレン『有閑階級の理論』).
