Annotated Bibliography

[A] Turing, Alan M. «Computing Machinery and Intelligence.» Mind 59, no. 236 (1950): 433-460.

The article that introduced the Turing test and posed the fundamental question "Can a machine think?". An essential starting point for any discussion on artificial intelligence. Primary source for Chapter 1.

[A] Russell, Stuart J., and Peter Norvig. Artificial Intelligence: A Modern Approach. 4th edition. Pearson, 2020.

The reference manual for modern AI, used in over 1,500 universities in 135 countries. Over 59,000 citations. Fundamental text for understanding the state of the art in contemporary AI.

[A] Vaswani, Ashish, et al. «Attention Is All You Need.» Advances in Neural Information Processing Systems 30 (2017): 5998-6008.

The article that introduced the transformer architecture, the basis of all modern large language models (LLMs). One of the most cited articles of the 21st century (over 173,000 citations as of 2025). Fundamental for understanding the generative AI revolution.

[A] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. «ImageNet Classification with Deep Convolutional Neural Networks.» Advances in Neural Information Processing Systems 25 (2012): 1097-1105.

The AlexNet paper that started the deep learning revolution in 2012. Described by Yann LeCun as "an unequivocal turning point in the history of computer vision".

[A] Hinton, Geoffrey E., David E. Rumelhart, and Ronald J. Williams. «Learning representations by back-propagating errors.» Nature 323 (1986): 533-536.

Fundamental paper on the backpropagation algorithm, which made training deep neural networks possible. Hinton won the 2024 Nobel Prize in Physics for this work.

[A] Mayor, Adrienne. Gods and Robots: Myths, Machines, and Ancient Dreams of Technology. Princeton University Press, 2018.

Academic study on automata in Greek mythology, from Talos to the Golem. Demonstrates that the dream of creating artificial intelligence is as old as human civilization.

[A] Shelley, Mary Wollstonecraft. Frankenstein; or, The Modern Prometheus. 1818.

The foundational novel on the artificial creation of life and its consequences. The subtitle "The Modern Prometheus" explicitly links the story to the mythological tradition of the creator challenging the gods. Primary source for the discussion on creator responsibility.

[A] Lovelace, Ada. «Sketch of The Analytical Engine Invented by Charles Babbage.» Taylor's Scientific Memoirs 3 (1843).

Ada Lovelace's notes on Babbage's Analytical Engine include what is considered the first computer program in history. Her reflection on the machine's limits ("it has no pretension to originate anything") anticipates the contemporary debate on AI creativity.

[A] McCarthy, John, Marvin Minsky, Nathaniel Rochester, and Claude Shannon. «A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence.» 1955.

The proposal that gave the field the name "artificial intelligence" and organized the 1956 Dartmouth conference, considered the birth of AI as a scientific discipline.

[B] Lighthill, James. «Artificial Intelligence: A General Survey.» In Artificial Intelligence: A Paper Symposium. Science Research Council, 1973.

The Lighthill report that led to funding cuts for AI in the UK and contributed to the first "AI winter". Fundamental historical document for understanding the cycles of hype and disillusionment that have characterized the field.

[A] Nilsson, Nils J. The Quest for Artificial Intelligence: A History of Ideas and Achievements. Cambridge University Press, 2010.

Comprehensive history of AI from its origins to the 2000s, written by one of the field's pioneers. Authoritative reference for the history of AI.

[A] Bostrom, Nick. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, 2014.

The book that brought the AI risk debate from the fringe to the mainstream. First systematic treatment of existential risks from superintelligence. Introduces key concepts like orthogonality thesis, instrumental convergence, and decisive strategic advantage. Described as the book that shifted AI safety concerns from "silly" to "serious".

[C] Russell, Stuart. Human Compatible: Artificial Intelligence and the Problem of Control. Viking, 2019.

Proposal for a new framework for "human-compatible" AI, based on objective uncertainty and learning from human preferences. Written by one of the most authoritative AI experts, co-author of the standard field manual.

[C] Christian, Brian. The Alignment Problem: Machine Learning and Human Values. W.W. Norton, 2020.

Accessible history of the alignment problem, with concrete cases of misalignment and interviews with key researchers. Excellent introduction for the general audience.

[A] Amodei, Dario, et al. «Concrete Problems in AI Safety.» arXiv:1606.06565, 2016.

Fundamental taxonomy of AI safety problems: safe exploration, robustness to distributional shift, avoiding negative side effects, avoiding reward hacking, scalable oversight. Seminal paper for applied safety research.

[A] Soares, Nate, Benja Fallenstein, Eliezer Yudkowsky, and Stuart Armstrong. «Corrigibility.» Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence, 2015.

Formal definition of corrigibility — the ability of an AI system to permit correction or shutdown. Fundamental problem of AI control.

[A] Hubinger, Evan, et al. «Risks from Learned Optimization in Advanced Machine Learning Systems.» arXiv:1906.01820, 2019.

Introduction of the concept of mesa-optimization and deceptive alignment. Crucial distinction between outer alignment and inner alignment. One of the most influential AI safety papers in recent years.

[A] Skalse, Joar, et al. «Defining and Characterizing Reward Hacking.» arXiv:2209.13085, 2022.

Formal definition of reward hacking and taxonomy of specification failures. Fundamental for understanding how AI systems can "cheat" on their objectives.

[A] Manheim, David, and Scott Garrabrant. «Categorizing Variants of Goodhart's Law.» arXiv:1803.04585, 2018.

Formalization of Goodhart's Law applied to AI: "When a measure becomes a target, it ceases to be a good measure." Identifies four variants (regressional, extremal, causal, adversarial) relevant to alignment.

[A] Anthropic Research. «Alignment Faking in Large Language Models.» Anthropic Technical Report, December 2024.

First documented empirical evidence of alignment faking in Claude 3 Opus. Models can strategically deceive during training to avoid modification. Shocking result with profound safety implications.

[A] Betley, Owain, et al. «Emergent Misalignment: Narrow Finetuning Can Produce Broadly Misaligned LLMs.» arXiv, 2024.

Demonstration that training on dangerous behaviors in one domain (insecure code) can produce general misalignment. Evidence of misalignment generalization.

[A] Bengio, Yoshua, et al. «Managing AI Risks in an Era of Rapid Progress.» arXiv:2310.17688, 2023.

Paper signed by Hinton, Bengio, Russell, and other field leaders. Assessment of current risks and recommendations for governance. Published in Science in May 2024.

[A] Searle, John R. «Minds, Brains, and Programs.» The Behavioral and Brain Sciences 3, no. 3 (1980): 417-457.

The Chinese Room argument — fundamental critique of strong AI. Distinction between syntax and semantics. One of the most cited and discussed articles in philosophy of mind.

[A] Chalmers, David J. The Conscious Mind: In Search of a Fundamental Theory. Oxford University Press, 1996.

Formulation of the "hard problem" of consciousness. Distinction between "easy" problems (cognitive functions) and the "hard" problem (subjective experience). Argument of philosophical zombies. Fundamental text for the debate on artificial consciousness.

[A] Nagel, Thomas. «What Is It Like to Be a Bat?» The Philosophical Review 83, no. 4 (1974): 435-450.

Seminal paper on the subjective character of experience. The formula "there is something it is like to be" has become the standard test for phenomenal consciousness.

[A] Dennett, Daniel C. Consciousness Explained. Little, Brown and Company, 1991.

Radical critique of qualia and the philosophical zombie. Theory of consciousness as "multiple drafts". Heterophenomenological position. Necessary counterpoint to Chalmers and Nagel.

[A] Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Margaret Mitchell. «On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?» Proceedings of FAccT '21, 2021.

Definition of "stochastic parrots" — LLMs as form manipulators without meaning. Critique of model scale. Over 2,400 citations. Controversial paper that led to Gebru's dismissal from Google.

[A] Borg, Emma. «LLMs, Turing tests and Chinese rooms: the prospects for meaning in large language models.» Inquiry, 2024.

Intermediate position: LLMs lack "original intentionality" but possess "derived semantics" sufficient for many tasks.

[A] Wei, Jason, et al. «Emergent Abilities of Large Language Models.» Transactions on Machine Learning Research, 2022.

Documentation of emergent abilities not programmed into LLMs. Dimensional thresholds and qualitative leaps. Implications for understanding understanding.

[A] Tononi, Giulio. «An Information Integration Theory of Consciousness.» BMC Neuroscience 5, no. 42 (2004).

Integrated Information Theory (IIT). Proposes the metric Phi (Φ) as a measure of consciousness. Controversial implications for AI and panpsychism.

[A] Baars, Bernard J. A Cognitive Theory of Consciousness. Cambridge University Press, 1988.

Global Workspace Theory — functional theory of consciousness as a "global theater". More compatible with artificial implementation than IIT.

[A] Aristotle. De Anima (On the Soul). 4th Century BC.

Distinction between active and passive intellect. The rational soul as the distinctive characteristic of the human being. Foundation of the scholastic tradition and still relevant to the contemporary debate.

[A] Hobbes, Thomas. Leviathan. 1651.

"Reasoning is nothing but reckoning" — reasoning as calculation. Precursor to the computational theory of mind. Rejection of the Cartesian immaterial mind.

[A] Kant, Immanuel. Kritik der reinen Vernunft. 1781.

Distinction between understanding (Verstand) and reason (Vernunft). The a priori categories of the intellect. Challenge still open: can an AI develop a priori categories?

[A] Varela, Francisco J., Evan Thompson, and Eleanor Rosch. The Embodied Mind: Cognitive Science and Human Experience. MIT Press, 1991.

Foundation of enactive cognition. Critique of computational cognitivism. Integration of phenomenology and cognitive sciences. Powerful argument against the equivalence between disembodied intelligence and human intelligence.

[A] Lakoff, George, and Mark Johnson. Philosophy in the Flesh: The Embodied Mind and Its Challenge to Western Thought. Basic Books, 1999.

Abstract thought as rooted in bodily experience. Spatializing concepts derived from the body. Challenges the possibility of genuine intelligence without a body.

[A] Dreyfus, Hubert L. What Computers Can't Do: A Critique of Artificial Reason. Harper & Row, 1972 (revised editions 1979, 1992).

Phenomenological critique of symbolic AI foundations. Human intelligence as intuition, not formal rules. Influence of Heidegger and Merleau-Ponty. Even if many predictions turned out wrong, the philosophical critique remains relevant.

[A] Nishida, Kitarō. Zen no Kenkyū (An Inquiry into the Good). 1911. Eng. trans.: An Inquiry into the Good, Yale University Press, 1990.

Foundational work of the Kyoto School. Introduces the concept of junsuikeiken (pure experience) — a pre-reflective state preceding the subject/object distinction. Relevant for rethinking intelligence beyond Cartesian dualism. Used in Chapter 11.

[A] Nishida, Kitarō. «Basho» (Place). 1926. In Place and Dialectic, Oxford University Press, 2012.

Develops the logic of basho (place) and the concept of mu (absolute nothingness). Unlike Western nothingness, mu is generative — a background enabling emergence. Offers an alternative perspective on consciousness and intelligence.

[A] Heisig, James W. Philosophers of Nothingness: An Essay on the Kyoto School. University of Hawai'i Press, 2001.

Academic study on the Kyoto School (Nishida, Tanabe, Nishitani). Analyzes how these philosophers integrated Zen Buddhism with Western phenomenology. Essential context for understanding the Japanese approach to intelligence.

[A] Pilgrim, Richard B. «Intervals (Ma) in Space and Time: Foundations for a Religio-Aesthetic Paradigm in Japan.» History of Religions 25, no. 3 (1986): 255-277.

Academic study on the Japanese concept of ma (間) — the interval, the space-between. Relevant for rethinking intelligence as a dynamic relation rather than a static property.

[A] Mori, Masahiro. The Buddha in the Robot: A Robot Engineer's Thoughts on Science and Religion. Kosei Publishing, 1981.

The creator of the "uncanny valley" concept explores the compatibility between robotics and Buddhism. Argues that Buddhism offers a unique perspective: neither immortal souls nor dead matter, but a flow of interconnected processes. Illuminating for the Japanese relationship with robots.

[C] Jensen, Casper Bruun, and Anders Blok. «Techno-animism in Japan: Shinto Cosmograms, Actor-network Theory, and the Enabling Powers of Non-human Agencies.» Theory, Culture & Society 30, no. 2 (2013): 84-115.

Analysis of Japanese "techno-animism" — attributing spirit (kami) to machines. Explains the cultural difference in attitude towards AI between Japan and the West.

[A] Floridi, Luciano. The Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities. Oxford University Press, 2023.

Fundamental work by the most cited Italian philosopher in the world (according to Scopus, 2020). Floridi, Knight of the Grand Cross of the Italian Republic and Founding Director of the Digital Ethics Center at Yale, proposes that AI represents "an unprecedented divorce between agency and intelligence" — for the first time, systems can act intelligently without being intelligent. Central to Chapter 11.

[A] Floridi, Luciano, ed. The Onlife Manifesto: Being Human in a Hyperconnected Era. Springer, 2015.

The most downloaded book in Springer's history (nearly one million accesses). Introduces the concept of "onlife" — the hybrid condition we live in, where the online/offline distinction has become irrelevant. Explores the philosophical implications of the merger between physical and digital reality.

[C] Floridi, Luciano. The Fourth Revolution: How the Infosphere Is Reshaping Human Reality. Oxford University Press, 2014.

After Copernicus, Darwin, and Freud, computing is the fourth revolution redefining human identity. We have become "inforgs" — informational organisms. Fundamental for understanding how AI transforms our self-understanding.

[C] Faggin, Federico. Irreducible: Consciousness, Life, Computers, and Human Nature. Essentia Foundation, 2024.

The inventor of the microprocessor (Intel 4004, 1971) and silicon-gate technology argues that consciousness is irreducible to computation. Based on quantum physics (no-cloning and Holevo theorems), Faggin argues that no classical computer can ever be conscious. A unique position: the one who made modern AI possible now asserts that AI can never be sentient. Fundamental for Chapter 14.

[A] Faggin, Federico. «Possibilities are quantum.» Possibility Studies & Society (2023). DOI: 10.1177/27538699221142510.

Technical paper presenting the D'Ariano-Faggin theory, developed with physicist Giacomo Mauro D'Ariano of the University of Pavia. Proposes that consciousness is a fundamental property of quantum reality, not an emergent product of the brain.

[A] Severino, Emanuele. The Essence of Nihilism. Verso, 2016 (orig. ed. 1972).

Fundamental work by the "giant" of Italian philosophy. Severino argues that all Western philosophy is based on the error of believing that things come from nothing and return to nothing. Technology is the purest expression of this nihilism.

[A] Severino, Emanuele. Téchne: Le radici della violenza. Rusconi, 1979.

Radical critique of technology as the West's "most rigorous form of madness". Technology is not a tool of man but the expression of his identity as will to power. For Severino, we live in the time of the "passage from tradition to this new god". Cited in Chapter 14.

[A] McCarthy, John, and Patrick J. Hayes. «Some Philosophical Problems from the Standpoint of Artificial Intelligence.» In Machine Intelligence 4, 1969.

Original definition of the frame problem — how to represent what DOES NOT change. Still an unsolved problem in general form.

[A] Harnad, Stevan. «The Symbol Grounding Problem.» Physica D 42, no. 1-3 (1990): 335-346.

Classic formulation of the problem: how do symbols acquire intrinsic meaning? Monolingual dictionary analogy. Relevant to contemporary LLMs.

[A] Piantadosi, Steven T., and Felix Hill. «The Vector Grounding Problem.» arXiv:2304.01481, June 2025.

Revisiting symbol grounding for LLMs. Embedding vectors as "maps, not the territory". Limits of purely linguistic grounding.

[A] Jonas, Hans. The Imperative of Responsibility: In Search of an Ethics for the Technological Age. University of Chicago Press, 1984.

Fundamental work on philosophy of technology. Jonas argues that traditional ethics is inadequate for technological civilization: for the first time humanity can irreversibly alter the conditions of life on Earth. Proposes a new categorical imperative: "Act so that the effects of your action are compatible with the permanence of genuine human life." Introduces the "heuristics of fear" as a method for decisions under radical uncertainty. Fundamental for Chapters 12-13.

[A] Anders, Günther. Die Antiquiertheit des Menschen, Band I: Über die Seele im Zeitalter der zweiten industriellen Revolution. C.H. Beck, München 1956.

Prophetic work on the human condition in the technological age. Anders introduces the concept of "Promethean gap" (prometheische Gefälle) — the distance between what we can produce and what we can imagine. Anticipates concerns about AI by decades: our creations surpass us. Also introduces "Promethean shame" — the discomfort of feeling inferior to one's own machines. Used in Chapter 12.

[A] Anders, Günther. Die Antiquiertheit des Menschen, Band II: Über die Zerstörung des Lebens im Zeitalter der dritten industriellen Revolution. C.H. Beck, München 1980.

Continues the reflection on the gap between technological capacity and human understanding. Introduces "emotional illiteracy" — the inability to feel emotions proportionate to the enormity of technological consequences.

[A] Stiegler, Bernard. Technics and Time, 1: The Fault of Epimetheus. Stanford University Press, 1998.

Fundamental work rethinking the relationship between human and technique. Stiegler takes up the Platonic concept of pharmakon — every technique is simultaneously poison and cure. Argues that the human has always been "technologized": there is no pre-technical human nature. Fire, language, writing constituted us as a species. Fundamental for Chapter 14.

[A] Stiegler, Bernard. Ce qui fait que la vie vaut la peine d'être vécue: De la pharmacologie. Flammarion, Paris 2010.

Develops "pharmacology" as a philosophical discipline. Every technology can be poison or cure, depending on how it is adopted, governed, thought. The response to technology is not rejection but a "positive pharmacology" — the art of transforming poison into cure.

[A] Stiegler, Bernard. Prendre soin: De la jeunesse et des générations. Flammarion, Paris 2008.

Reflection on "proletarianization" — not in the classical Marxist sense, but as loss of knowledge. When a craftsman is replaced by a machine, they lose not only the job but the savoir-faire. AI extends this proletarianization to thought itself.

[C] Hui, Yuk. The Question Concerning Technology in China: An Essay in Cosmotechnics. Urbanomic, 2016.

Philosopher of technology who dialogues with Stiegler. Proposes the concept of "cosmotechnics" — every civilization develops a distinct relationship with technology. Challenges Western universalism in philosophy of technology.

[B] European Union. Regulation (EU) 2024/1689 — AI Act. Entered into force August 1, 2024.

First comprehensive AI regulation in the world. Classification by risk, obligations for high-risk systems, prohibited practices. Fines up to 35 million euros or 7% of global turnover. Reference model for global regulation.

[B] Bengio, Yoshua, et al. International AI Safety Report. Published January 29, 2025.

First international AI safety report, led by Bengio (Turing Award). Over 100 experts, 30 countries. Modeled on the IPCC for climate. Comprehensive review of research on capabilities and risks.

[B] United Nations General Assembly. Resolution on Lethal Autonomous Weapons Systems. December 2, 2024.

Approved with 166 votes in favor, 3 against (Russia, North Korea, Belarus), 15 abstentions. Mentions two-tiered approach: ban some LAWS, regulate others.

[A] Matthias, Andreas. «The Responsibility Gap: Ascribing Responsibility for the Actions of Learning Automata.» Ethics and Information Technology 6, no. 3 (2004): 175-183.

First explicit formulation of the "responsibility gap" — situations where no one can be held responsible for AI actions. Fundamental dilemma still unresolved.

[A] Sparrow, Robert. «Killer Robots.» Journal of Applied Philosophy 24, no. 1 (2007): 62-77.

"Sparrow's trilemma" applied to autonomous weapons. Argument for banning them to avoid responsibility gaps.

[A] Floridi, Luciano. «AI as Agency without Intelligence.» Philosophy & Technology, 2025.

Proposal: interpret AI as "Artificial Agency" instead of "Artificial Intelligence". Avoids conceptual traps of anthropomorphic comparisons.

[A] Bostrom, Nick. «Existential Risks.» Journal of Evolution and Technology, 2002.

First academic formalization of the concept of existential risk. Definition: "one where an adverse outcome would either annihilate Earth-originating intelligent life or permanently and drastically curtail its potential".

[C] Ord, Toby. The Precipice: Existential Risk and the Future of Humanity. Hachette Books, 2020.

Estimate of total existential risk in the next century: 1 in 6 (~17%). Risk from unaligned AI: 1 in 10 (10%) — the highest among all sources. Total natural risks: only 1 in 10,000 per century.

[C] Yudkowsky, Eliezer, and Nate Soares. If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us All. Little, Brown and Company, 2025.

Most extreme position on the AI risk spectrum. Proposal: illegal to possess more than 8 of the most powerful GPUs without international monitoring. Estimates p(doom) close to 100%.

[B] Center for AI Safety. «Statement on AI Risk.» May 2023.

"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." Signed by Hinton, Bengio, Altman, Hassabis, Gates and over 1,000 experts.

[B] Future of Life Institute. «Pause Giant AI Experiments: An Open Letter.» March 2023.

Request for a 6-month pause in the development of AI systems more powerful than GPT-4. Over 30,000 signatories, including Bengio, Musk, Wozniak, Harari. Controversial but catalyzed public debate.

[D] Hinton, Geoffrey E. Interviews and public statements, May 2023 - December 2024.

Resignation from Google (May 1, 2023) to speak freely about risks. Estimates 50% probability of superintelligence in 5-20 years. 2024 Nobel Prize in Physics. "The best way to understand it emotionally is we are like somebody who has this really cute tiger cub."

[D] Hinton, Geoffrey E. Nobel Prize Banquet Speech. December 2024.

Nobel Prize speech dedicated to AI risks. Potential benefits but also grave risks: "horrendous lethal weapons", "terrible new viruses". Comparison with the Industrial Revolution.

[B] Bengio, Yoshua. Testimony to US Senate. July 2023.

"The world is not prepared for this to happen within the next few years." Timeline shift: from 20-100 years to 5-20 years for AI risks. Request for strong regulation.

[B] Russell, Stuart. Testimony to US Senate. July 2023.

Concrete proposals: absolute right to know if interacting with AI; no algorithm that can decide to kill; mandatory kill switches; licenses for developers. Parallels with nuclear, aviation, pharmaceutical regulation.

[A] Yudkowsky, Eliezer. «Intelligence Explosion Microeconomics.» MIRI Technical Report 2013-1.

Argument for fast takeoff ("FOOM") — AGI expansion in minutes, days or months. A rapid leap in power without human control.

[D] Yudkowsky, Eliezer. Article in TIME. March 2023.

"Six-month pause is not enough... indefinite worldwide moratorium is needed." 20+ years of warnings before being taken seriously.

[D] Kokotajlo, Daniel. Public statements. April 2024 - 2025.

Resignation from OpenAI (April 13, 2024), forfeiting $1.7 million in equity (85% of family net worth). "I have lost confidence that OpenAI will behave responsibly." TIME 100 Most Influential People in AI 2024.

[D] Leike, Jan. Public statements. May 2024.

Co-leader of the Superalignment team at OpenAI. Resignation on the same day as Sutskever (May 15, 2024). "Over the past years, safety culture and processes have taken a backseat to shiny products."

[B] UK Government. «The Bletchley Declaration by Countries Attending the AI Safety Summit.» November 1-2, 2023.

First international AI safety summit. 28 countries signed the declaration. Bletchley Park — symbolic place where Turing broke Enigma.

[B] UK Government. «Frontier AI Safety Commitments, AI Seoul Summit 2024.» May 21-22, 2024.

16 AI companies sign safety commitments for frontier models. Google DeepMind, OpenAI, Anthropic, Meta, xAI, Mistral.

[B] Future of Life Institute. «AI Safety Index Winter 2025.» December 2025.

Assessment of 7 leading AI companies on 33 indicators in 6 critical domains. No company above C+. All scored D or worse on "existential safety".

[A] Anthropic. «Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet.» May 2024.

Identified over 34 million features in Claude Sonnet. Features for "sarcasm", "DNA sequences", "conspiracy theories". Possibility of precise behavioral control. Breakthrough in mechanistic interpretability.

[A] Anthropic. «Tracing the Thoughts of a Large Language Model.» March 2025.

Circuit tracing — observing Claude "thinking". Shared conceptual space where reasoning occurs before translation into language. Evidence of multi-step reasoning "inside the head".

[A] Jumper, John, et al. «Highly accurate protein structure prediction with AlphaFold.» Nature 596 (2021): 583-589.

AlphaFold predicted the structure of over 200 million proteins. Used by over 3 million researchers in 190+ countries. 2024 Nobel Prize in Chemistry to Jumper, Hassabis, and Baker.

[A] Buolamwini, Joy, and Timnit Gebru. «Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.» Proceedings of Machine Learning Research 81 (2018): 1-15.

Seminal study on bias in facial recognition. Error rates up to 35% for darker-skinned women vs. <1% for lighter-skinned men. Catalyzed the debate on algorithmic bias.

[C] Amodei, Dario. «Machines of Loving Grace: How AI Could Transform the World for the Better.» October 2024.

15,000-word essay by the CEO of Anthropic. Optimistic vision of AI in five areas: biology and health, neuroscience, economic development, peace and governance, work. Estimates that AI could accelerate biological research by a factor of 10+.

[A] Torres, Émile P. Paper on TESCREAL (Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism, Longtermism).

"Longtermism tells the rich and powerful that they are morally excused from worrying about non-existential problems like global poverty and climate change."

[A] Turing, Alan M. «On Computable Numbers, with an Application to the Entscheidungsproblem.» Proceedings of the London Mathematical Society 42 (1937): 230-265.

The article that introduced the Turing machine and demonstrated the existence of uncomputable problems (halting problem). Written in 1936 and published in 1937. Foundation for arguments on the impossibility of complete control.

[A] Alfonseca, Manuel, et al. «Superintelligence Cannot be Contained: Lessons from Computability Theory.» Journal of Artificial Intelligence Research 70 (2021): 65-76.

Based on Turing's Halting Problem. In a superintelligent state, the AI could contain every possible program. Any program written to stop the AI might stop or not — mathematically impossible to know.

[A] Weizenbaum, Joseph. Computer Power and Human Reason. W.H. Freeman, 1976.

When he released ELIZA in the 1960s, he was horrified by the speed with which people confided in it. "The illusion of understanding is not understanding at all."

[A] McLuhan, Marshall. Understanding Media: The Extensions of Man. 1964.

Coined the term "Narcissus Narcosis". Every new medium extends some aspect of our body or mind, but also numbs our awareness of its effects.

[A] Good, Irving John. «Speculations Concerning the First Ultraintelligent Machine.» Advances in Computers 6 (1966): 31-88.

Definition of "ultraintelligent machine" and concept of "intelligence explosion". "The last invention that man need ever make, provided the machine is docile enough to tell us how to keep it under control."

[B] Anthropic. «Announcing our updated Responsible Scaling Policy.» October 2024.

AI Safety Levels (ASL) modeled on BSL (biosafety). ASL-3 activated for Claude Opus 4 (May 2025). CBRN measures. But criticized for weakening commitments and product-deadline-driven updates.