Faces*, Madge Gill, 1920–1960. Newham Archives and Local Studies Library
I. The Will of Angels
"Once upon a time a great Cabbalist lived in Prague, called the Rabbi Löw. He made a human figure of clay, and left a small aperture in the lesser brain in which he laid a parchment with the unutterable name of God written on it. The clod immediately arose and was a man; he performed all the duties of a servant for his creator, he fetched water, and hewed wood. All through the Jews' quarter he was known as the Golem of the great Rabbi Löw."
A common topic of interest in 18th-century French salons was the theodicy:[3] given an omnipotent, omnibenevolent God, why is there evil in the world?[4] This excited a range of answers. There was the Panglossian theodicy, the pollyannaish view that, "all is for the best [in] [...] the best of all possible worlds".[5] Others subscribed to the Irenaean theodicy that evil is necessary for full human development.[6]
A particularly prominent theory though was one originally argued by Saint Augustine of Hippo. Pulling from the Neoplatonists, Augustine first contended that evil is not a thing in itself, but rather a privation of the good.[7] Then, in exercising free will, a great gift in itself, humans can move themselves away from the good.
However, there seems to be a slight hole in this notion. While the theodicy is adequate for explaining human-derived evil, it does not resolve the 'problem of evil' in regard to natural sources; if evil extends from human freedom, where do we place natural disasters, diseases, or independent animal suffering?[8]
To patch this issue, an addendum can be made to the theodicy in what is commonly referred to as the 'two falls theodicy'. In standard Catholic metaphysics, angels are aeviternal beings which apprehend 'intuitively' and whose wills, therefore, adhere immovably.[9][10] When an angel makes a choice, it persists forever. In the 'two falls theodicy', the non-human mediated ills of the world are downstream from these singular decisions of angels who reject the 'good'.[11]
A simple choice, made by a being of sufficient capacity, is enough to irrevocably distort the world. For most of human history, consequences were limited to our immediate positional and temporal surroundings. Today, actions engraved into the embeddings of systems have the capability of far outlasting their original context.[12]
Small changes to the starting state of a complex system can incite highly divergent outcomes. An object with a constant 'pop'[13] of 1 m/s⁶, assuming no other initial conditions, travels the circumference of the Earth in 56 seconds and achieves a velocity faster than the speed of light in 130.[14]
We stand now at an inflection point in history, where present choices can ramify forward without recall.[15]
The Supper (La Cena), Belkis Ayón, 1991. Tate Modern, London
II. Why the West Has (Not) Won;
or, The Incoherence of the Incoherence
Scientific development requires a specific epistemology. In order to proceed over time in scientific inquiry, one has to believe both that the world is knowable and that knowing the world is good.
Around 1027, ibn Sīnā[16] published The Book of Salvation. Within its second 'treatise', over two centuries before Aquinas' Five Ways, ibn Sīnā describes a proof of God arising from the contingent and yielding a necessary existent.[17] Given a necessary existent, then all else stems from it[18] and insofar as things stem from it they are knowable[19] and good.[20]
Less than 100 years later, al-Ghazālī, the 'mujaddid'[21] of the 11th century,[22] wrote the Incoherence of the Philosophers. Here al-Ghazālī argues for an Ash'arite occasionalism:
"Let us, then, take a specific example—namely, the burning of cotton, for instance, when in contact with fire. [...] The one who enacts the burning by creating blackness in the cotton, [causing] separation in its parts, and making it cinder or ashes is God [...] As for fire, which is inanimate, it has no action. For what proof is there that it is the agent? They have no proof other than observing the occurrence of the burning at the [juncture of] contact with the fire."[23][24]
While the world may be good, it is not knowable.
Though not without dissent,[25] al-Ghazālī's approach proved ascendant. The madrasa system displaced the older patronage networks that sustained the Bayt al-Ḥikmah tradition. Natural philosophy made way for a curriculum which prioritized jurisprudence and Qur'ānic exegesis.[26] The Islamic world, which had preserved and expanded on the Western tradition in philosophy, math, and science[27] for half a millennium while Europe languished in darkness, became instead the recipient of European advancement.
The Latin West maintained the twin epistemologic pillars for slightly longer. Kepler presented astronomical inquiry as a way of knowing God.[28] A century later, Boyle's Christian Virtuoso defended experimental philosophy as a way to illuminate God's goodness.[29] In the General Scholium to the Principia, Newton held that the cosmic system "could only proceed from the counsel and dominion of an intelligent and powerful Being."[30]
However, with the Enlightenment, the idea that knowing the world is good began to fall. A possible theodicean response is to break down the premises. One cannot establish an 'ought' from an 'is',[31] so without an explicit moral ordering of the universe where is value derived? Scholastic vestiges remained, yet unmoored from any metaphysical anchor.[32]
In ethics, the good became restructured around the "two sovereign masters" of pain and pleasure;[33] efficiency the dominant signal in markets.[34] Without a telos, purpose came from what could be measured: "anything which does not conform to the standard of calculability and utility must be viewed with suspicion".[35] The value of discovery limited to its instrumentalization.
Modernity still has its triumphs. While the tractable is not identical to the good, the mode of orientation can be near enough to appear aligned in non-pathologic scenarios. In the 20th century alone, life expectancy doubled.[36] Famine has become a peripheral phenomenon primarily associated with war rather than agriculture.[37] Smallpox, the killer of over half a billion, has been eradicated.[38]
Its failure in sensing the inherent dignity of the person, the temptation towards the merely legible, though, curses the regime of modernity to eventually misalign from the good.
In Buck v. Bell, the United States Supreme Court, in an 8-1 decision,[39] upheld the forced sterilization of the 'unfit' to promote the general health and welfare of society:
"It would be strange if it could not call upon those who already sap the strength of the State for these lesser sacrifices [...] in order to prevent our being swamped with incompetence. It is better for all the world, if instead of waiting to execute degenerate offspring for crime, or to let them starve for their imbecility, society can prevent those who are manifestly unfit from continuing their kind. The principle that sustains compulsory vaccination is broad enough to cover cutting the Fallopian tubes. Three generations of imbeciles are enough."[40][41]
In the Holocaust, perhaps the central tragedy of the modern age, modernity again showed its fruits. Though history is littered with stories of mass killings, the Holocaust was not a harkening back to a barbaric past. It is distinguished, not just in magnitude but also in kind, by a particular rationality. The genocide was taken as a means, and correct from within its own axiomatic structure.[42] Without an external understanding of the good, there is no end toward which to correctly direct one's rationality.[43][44]
Our current tech landscape has provided remarkable achievements, yet there still remains an absence at its core. As we pursue greater advancements, the divergence between the efficient and the good becomes increasingly pronounced.[45] While in an age of relative abundance, we are tightroping across a path where, with rising capability, a small deviation can quickly cascade into a fall.[46] Yet the system, having lasted hundreds of years, resists correction; corrigibility to the numinous remains strained.
The Rabbis (Rabinos), Leonora Carrington, 1960. Rose Art Museum, Brandeis University
III. Sed Contra
While the development of artificial intelligence carries grave risk and must be taken with the utmost care, many standard objections serve to obscure the genuine concern rather than clarify.[47] These critiques both inflate and diminish capability while positing issues orthogonal to any real capacity.
On Stochastic Parrots
A common critique[48] given to modern generative AI systems is that they are merely 'stochastic parrots'[49] or 'next token predictors'. The tricky part of this statement is that there is some level where this is correct.
There are two main phases in the life-cycle of an LLM. The 'building' phase[51]and the 'usage' phase.[52]This building phase can then be split further into constructing out the 'foundation'[53]and then 'sculpting' it.[54]
In the initial foundation building phase, a large corpus of text is taken and strictly from how words[55] are used in relation to each other in the corpus, a high-dimensional 'raw' 'model of the world'[56] is formed.[57] Then, in the sculpting phase, this raw model of the world is shaped to be more amenable as a human assistant.[58] Finally, in the 'usage' phase your input goes into this sculpted model of the world and based on the structure within, emits a response.[59]
There is a sense that when prompting, an LLM is 'predicting' the next token in a sequence. However, emergent phenomena are the norm in complex systems, and when exhibited, generally make a description of parts to explain the whole not 'untrue' but misleading or non-helpful. Outside of highly specific scenarios, it is not useful to explain a computer as a series of transistors or further to describe any matter according to its fermionic makeup.
Most commonly, this type of argumentation seems to occur when one is trying to make a claim to separate human cognition from what occurs in LLMs. However, mereological observation does not constitute ontological proof. It is clear from even minor use that LLMs are not just stochastic parrots. Statements on what LLMs 'are' should be made directly, not laundered through a framing which falls apart on contact.
On the Environment
Despite apocalyptic hyperventilation to the contrary from media,[60][61] the environmental footprint for AI is fairly small. Obviously, as something that exists in the world, LLMs and their associated infrastructure have some environmental effect, but relative to their use this is negligible and possibly even negative when considering displacement costs.[62]
A similar story exists for emissions. Even in a 'lift-off' case with extremely strong AI uptake, limited local constraints, and low efficiency improvements in hardware or software, the IEA estimates data center emissions to only be at 475 Mt CO₂ by 2035[63][64] or less than ⅓ the output of the cement industry.[65]
This is not to say that data centers are immune to causing grid strain at a local level.[66] However, even here, energy supply is historically highly receptive to demand.[67] For those genuinely concerned about environmental degradation, the returns on attention are orders of magnitude higher in other areas.
On Intellectual Property
It has been a mainstay of internet culture, especially within tech circles, to be against intellectual property or believe it overly restrictive;[68] the operating assumption being that information should be free.[69] However, with the advent of LLMs, this notion has started to shift. There is a feeling that in AI's unmediated hoovering of data, some implicit boundary has been violated.
In the United States, copyright has always been something granted for the public benefit: "[t]o promote the Progress of Science and useful Arts".[70] Due to the inherent non-excludability of ideas, copyright exists as a way to incentivize their creation and discovery.[71]Copyright is merely a statutory concession for the good of the public and not a protection that flows from one's natural rights.[72][73]
To maximize the public welfare, even in the period of exclusivity provided, an author's rights are not absolute. The bounds on the exclusive rights of works were codified[74] in the Copyright Act of 1976 which established a four-factor test that utilized the purpose of use, the nature of the copyrighted work, the amount used, and the market impact on the copyrighted work to determine if a secondary work was a 'fair use' of the copyrighted material.[75]
With regards to AI, the use of copyrighted materials for training has been found to constitute fair use due to its highly transformative nature.[76] While whether the outputs of models themselves can constitute copyright infringement is under ongoing litigation,[77][78] copyright is a poor mechanism for addressing the displacement AI creates.
If we take AI displacement in specific industries to be bad, there are more direct and efficient ways to alleviate the harms. A Pigouvian tax on LLM providers, earmarked for affected industries, could compensate creators without requiring the increasingly strained application of a framework designed for a different issue.[79][80]
On Hallucination
One of the most tangible concerns people have of LLMs is related to having factual errors in outputs or 'hallucinations'. To clarify against what is a slightly common misconception, hallucinations do not come from some inherent 'randomness'[81] of LLMs, but rather through 'mistakes' in the models of the world being inferred into.[82]
The upside of this is that in using larger pre-training corpora and more robust post-training processes, successive model generations have significantly reduced hallucination rates while simultaneously becoming better calibrated in expressing uncertainty.[83]
A common mistake in judging digital output is comparison to an ideal form rather than existing alternatives. Compared to the counterfactual, existing deployment delays on autonomous vehicles have already cost thousands of lives.[84][85] Even in fields such as medical diagnosis with high potential consequences for inaccuracy, LLMs have been found, in certain circumstances, to exceed physician accuracy.[86]
This is not to say that hallucinations are not an issue, but rather that any sufficiently complex system will exhibit failure modes. When examining AI errors, it should be done in cognizance of their current rate of improvement and relative to the systems they stand to augment.
A final critique often levied against the AI industry as a whole is that it represents a speculative bubble whose eventual correction will inflict widespread economic harm. Stories of Allbirds, the shoe company, whose shares climbed over 600% after announcing a pivot to AI compute infrastructure,[88][89] make it hard to deny some mania in the markets.
However, it is perhaps clarifying here to view the AI industry in relation to its last major speculative episode: the dot-com bubble. In March of 2000, directly prior to the dot-com burst, information technology forward P/E stood at ~48.3x.[90] Today, due to massive earnings increases, the mark stands at only ~24.2x.[91] If we look at the 'shovel' play in both periods, Cisco had a P/E of ~222.4x, while today Nvidia still only stands at 41.8x.[92][93] Insofar as AI is a bubble, its 'irrational exuberance' still has room to grow.[94]
Even after any excess froth gets wrung out, the fallout is unlikely to follow the same patterns as some previous recent corrections. For established tech companies, AI is best thought of as a 'sustaining' rather than a 'disruptive' innovation.[95] Following from this, a majority of capex has come from free cash flow of highly profitable existing tech companies.[96] Compared to the dot-com crash, which was funded by retail speculation,[97] or 2008, which was built on leveraged consumer debt,[98] the current situation carries less structural risk. Further, in the case of a downturn, the majority of the build-out here is in compute and energy, which besides being highly fungible also historically have high positive externalities.[99]
Ten Virgins, Tove Jansson, 1953. Teuva Church, Finland
IV. Prohairesis
"The Rabbi ordered the precentor to pause at the end of the prayer. It was yet possible to save all, but later naught would avail — the whole world would be destroyed. He hastened home, and saw the Golem already seizing the joists of his house to tear down the building; he sprang forward, took the parchment out, and dead clay again lay at his feet."
Despite their inadequacies, there is a reason why these anti-AI arguments persist. People have an intuitive perception that there is something hollow at the center of the modern world and so glom onto the first rationale with even a façade of logical coherence.
Those who make claims of ushering in a new age are of the same group that brought us Soylent,[100] engagement-optimized feeds,[101] and claims of massive AI job displacement.[102][103] Even when AI leaders make remarks pointing to more humanistic ends,[104] it is hard to take these outside the milieu they operate within: "[W]ho could endure the Gracchi railing at sedition?"[105]
It is well understood across a range of fields that negative signals carry more weight than positive signals.[106] In decision-making under uncertainty, losses carry twice the weight of equivalent gains.[107] A common response then is to treat the asymmetry as a bias to be corrected and so actively discount any negative stimuli.[108] However, occasionally, several hundred million years of evolutionary selection encodes valuable information; Taleb distributions are real and often non-obvious.[109]
So where do we go from here? One must walk away from St. Petersburg; silence the martingale. In acting, we immediately shape our soul before any external consequence. To be truly in zugzwang requires a fully staid environment; however, we can change the conditions under which we operate.
Technology qua technology is not a good in itself. Its pharmakonic nature is decided in relation to the telos on which it is produced. We currently exist in an axial moment where choices made today can have an outsized effect on the future. In proceeding, one must act with caution, humility, and an understanding that in building, the 'ends' must be aligned to and in respect of the fundamental dignity of the human person.
Thanks to my family for reviewing earlier drafts of this work.
For any comments on this piece, please message me at joseph@babbo.dev.
The classic golem narrative has no singular authoritative source. However, Berthold Auerbach's biographical novel of Spinoza is thought to contain one of its earliest literary renderings.[110]↩︎↩︎
Berthold Auerbach, Spinoza: A Novel, trans. E. Nicholson (New York: H. Holt and Company, 1882), 216–217. ↩︎↩︎
Coined by Leibniz in the eponymous Essais de Théodicée (1710).[111]↩︎
Especially after the 1755 Lisbon Earthquake, where on All Saints' Day an estimated 8.7-magnitude[112]earthquake struck Portugal, killing tens of thousands, destroying up to 48% of Portuguese GDP,[113]and generating waves felt as far as Brazil,[114]inspiring writings from Voltaire,[115]Rousseau,[116]and Kant.[117]↩︎
From Dr. Pangloss in Candide,[115]whom Voltaire introduces to caricature Leibniz's rosy position from the Théodicée.[111]↩︎
"Just in the same way is the heavenly kingdom honourable to those who have known the earthly one. But in proportion as it is more honourable, so much the more do we prize it; and if we have prized it more, we shall be the more glorious in the presence of God. The Lord has therefore endured all these things on our behalf, in order that we, having been instructed by means of them all, may be in all respects circumspect for the time to come, and that, having been rationally taught to love God, we may continue in His perfect love: for God has displayed long-suffering in the case of man's apostasy; while man has been instructed by means of it, as also the prophet says, 'Your own apostasy shall heal you;'[118]God thus determining all things beforehand for the bringing of man to perfection, for his edification, and for the revelation of His dispensations, that goodness may both be made apparent, and righteousness perfected, and that the Church may be fashioned after the image of His Son, and that man may finally be brought to maturity at some future time, becoming ripe through such privileges to see and comprehend God."[119]↩︎
This was required as, if evil itself were a substance, it would necessitate either origination in God, undermining omnibenevolence; or a Manichaean dualism: "And, because my soul dared not be displeased at my God, it would not allow anything to be Yours which displeased it. Hence it had gone into the opinion of two substances, and resisted not, but talked foolishly."[120]↩︎
Even if one presumes that non-human evil propagates from 'the Fall', this would not explain, in light of evolution, pre-fall animal suffering (or even the serpent in the traditional narrative) without positing some exotic causality. ↩︎
Another way to put this is that angels understand non-discursively; everything is grasped immediately and in full, not through a chain of syllogisms. As there is nothing left to further understand, there is no basis for a change of will. "Such is the condition of the angels, because in the truths which they know naturally, they at once behold all things whatsoever that can be known in them."[121] "Now the angel's apprehension differs from man's in this respect, that the angel by his intellect apprehends immovably [...] Consequently [...] the angel's will adheres fixedly and immovably."[121]↩︎
This view, however, is not entirely uncontested. Scotus argued non-discursive knowing is non-corollary to immovable will: "[A]lthough the angel understand non-discursively what – according to them – man understands discursively, yet the intellect of man does not movably adhere to that which it reaches discursively; for it so holds with certitude (that is without doubting) the conclusion it discursively reaches as the angel holds it by seeing it in the principle non-discursively; therefore this immobility of the human intellect (that is, certitude) would have an equally immovable will just as that other [immovability] posited in the angel."[122][123] but rather, there is still a permanency due to lack of grace from God to further move oneself back to the good; a divine dereliction: "[F]or springing back meritoriously there is required another principle than the will, namely grace, which a bad angel cannot have of himself, – and God, according to the fact he deserted him, has disposed not to give it to him [...] [;] there not be another cause of their permanence in bad save divine dereliction."[123] Though, it should be noted, the word 'dunce' originated to describe the followers of Duns Scotus.[124]↩︎
More recent formulations posit this as an angelic 'omission' (or 'archon abandonment'), in which natural evil is traced not to direct angelic action but to the culpable abdication of a divinely assigned stewardship over creation.[125]↩︎
By the maker upon the made, at first; in time, by the made upon the maker. ↩︎
The fourth, fifth, and sixth derivatives of position with respect to time are commonly referred to as, respectively, snap, crackle, and pop. Yes, after the Rice Krispies mascots.[126]↩︎
x(t) = 0 + 0·t + (0/2)t² + (0/6)t³ + (0/24)t⁴ + (0/120)t⁵ + (1/720)t⁶ x(t) = t⁶/720; v(t) = t⁵/120 Circumference of the Earth ≈ 40,075 km Speed of light c ≈ 2.998e8 m/s t⁶/720 = 4.0075e7 m ⟹ t ≈ 55.38 s t⁵/120 = c ⟹ t ≈ 129.18 s ↩︎
"If we use, to achieve our purposes, a mechanical agency with whose operation we cannot efficiently interfere once we have started it, because the action is so fast and irrevocable that we have not the data to intervene before the action is complete, then..."[127]↩︎
From ibn, meaning 'son of'; the original referent is unknown. In the West, commonly referred to as Avicenna. ↩︎
Most thoroughly laid out in The Book of Salvation; a more succinct formulation can be found in Remarks and Admonitions: "Every totality having every one of its units as caused requires a cause external to its units. This is because either [1] it does not require a cause at all; hence it is necessary and not possible. But how could this be so when it is only necessitated by its units? [2] It requires a cause that is all its units; hence it is caused by itself. [...] [3] It requires a cause that is some of its units. But if every one of its units is caused, then some of its units are not more deserving of being the cause than some others. The reason is that the cause of the caused is more deserving of being the cause. Or [4] it requires a cause external to all its units. This is the remaining [truth]."[128][129]↩︎
"Thus, if there is a first cause, it is a cause of every existence and of the cause of the reality of every concrete existence."[129]↩︎
"Know that you came to perceive particular eclipses only because you fully comprehended its causes and everything concerning the heavens. When full comprehension takes place about all of the causes in things and their existence, there is a transference [of that full comprehension] from those to all of the effects."[130][131]↩︎
"Whatever necessarily exists through itself is pure good and pure perfection. In general, the good is that which everything desires, and that through which it is completed. Evil has no essence, but is either the absence of substance or of some state beneficial to the substance. Thus existence is goodness. Perfection of existence is goodness of existence, the existence that is untouched by absence — whether absence of substance or absence of something that belongs to substance — rather, it is perpetually actual. So it is pure good. That which, through itself, is [merely] possibly existent is not pure good. For its essence does not, through itself, have existence, so that its essence in itself suffers absence. What suffers absence in any way is not entirely free of evil and deficiency. Therefore there is no pure good apart from that which necessarily exists through itself."[132][131]↩︎
The 5th century of the hijrī calendar (1009–1106 AD; 401–500 AH). ↩︎
Elisions added to referent; other additions are from referent. ↩︎
Al-Ghazālī, The Incoherence of the Philosophers, trans. Michael E. Marmura, 2nd ed. (Provo, UT: Brigham Young University Press, 2000), Discussion 17, §§2–5, 166–167. ↩︎
Most famously in Ibn Rushd's[133]Incoherence of the Incoherence: "To deny the existence of efficient causes which are observed in sensible things is sophistry, and he who defends this doctrine either denies with his tongue what is present in his mind or is carried away by a sophistical doubt which occurs to him concerning this question. [...] Logic implies the existence of causes and effects, and knowledge of these effects can only be rendered perfect through knowledge of their causes. Denial of cause implies the denial of knowledge, and denial of knowledge implies that nothing in this world can be really known".[134]↩︎
It is slightly overdrawn to lay this entirely at the feet of al-Ghazālī. He is probably better understood as merely a seminal figure in an ongoing process than the immediate catalyst of the change (if such a thing could be allowed under Ash'arite metaphysics).
Beginning in the 700s, the Mu'tazili Islamic school of theology was formed. Mu'tazilis controversially believed the Qur'ān to be created, not coeternal, and took heavy influence from Greek rationalism, utilizing kalām, or the rational study of theology, to try to ground their beliefs.[135]Starting under the reign of al-Ma'mūn, Mu'tazilism gained increased prominence in the Abbasid empire, to the extent where in 833 the miḥna was established, where officials had to affirm a belief in Qur'ānic createdness under threat of persecution.[135][136]
The eccentricity of Mu'tazilism and the heavy-handedness under which it was enforced festered resentment among the ulamā, the Muslim religious scholar class.[136]The resulting backlash when al-Mutawakkil eventually reversed the miḥna in 851 saw the rise of the more traditional Ḥanbalī school, which rejected all kalām, and the reverse persecution of Mu'tazilis.[137]This also saw the creation of the Ash'arī school from an ex-Mu'tazilī, which sought to still use kalām but solely in the defence of traditional doctrine.[135]
With the rise of the Shī'a Buyids in 945, no Sunnī orthodoxy was enforced and Mu'tazilis and independent faylasūf thinkers, such as ibn Sīnā, were again permitted and sponsored, while the Ḥanbalī and Ash'arī factions competed among the ulamā.[138]When the Buyids fell to the Sunnī Seljuks in 1055, vizier al-Kundurī sided with the Ḥanbalīs and attempted to suppress all kalām, including the Ash'arīs.[138]However, upon the death of Ṭughril in 1063, al-Kundurī supported the infant son of Chaghrī Beg, Ṭughril's brother, over the elder son Alp Arslān, who with the help of Niẓām al-Mulk became the next sultan.[139]Upon himself becoming the vizier, Niẓām al-Mulk chose Ash'arism as the official state theology and sponsored the building of Niẓāmiyyas, a highly influential series of madrasas established as waqf whose endowment terms restricted instruction to the Islamic sciences.[137]He appointed al-Ghazālī the head of the most prominent, in Baghdad, in 1091.[138]↩︎
Besides preserving, spreading, and translating works from Greek, Persian, and Indian sources (itself a valuable work),[136]Islamic scholars made foundational contributions to the scientific corpus. Al-Khwārizmī's algebra,[135]the optical insights of al-Haytham,[138]and al-Battānī's astronomical discoveries[135]meaningfully furthered the Western world's canon of knowledge. ↩︎
"It is in this that Paul proposes to the Gentiles that they should contemplate God like the Sun in water or in a mirror. Why then as Christians should we take any less delight in its contemplation, since it is for us with true worship to honor God, to venerate him, to wonder at him? The more rightly we understand the nature and scope of what our God has founded, the more devoted the spirit in which that is done."[140]↩︎
"The consideration of the Vastness, Beauty, and Regular Motions, of the heavenly Bodies; the excellent Structure of Animals and Plants; besides a multitude of other Phænomena of Nature, and the Subserviency of most of these to Man; may justly induce him, as a Rational Creature, to Conclude, That this vast, beautiful, orderly, and (in a word) many ways admirable System of things, that we call the World, was fram'd by an Author supremely Powerful, Wise, and Good, can scarce be deny'd by an intelligent and unprejudic'd Considerer."[141][142]↩︎
"'Morally wrong' is the term which is the heir of the notion 'illicit,' or 'what there is an obligation not to do'; which belongs in a divine law theory or ethics. [...] [I]t is because 'morally wrong' is the heir of this concept, but an heir that is cut off from the family of concepts from which it sprang, that 'morally wrong' both goes beyond the mere factual description 'unjust' and seems to have no discernible content except a certain compelling force, which I should call purely psychological."[143]↩︎
"The capitalist economy rationalizes on the basis of strictly quantitative calculations and is oriented to the sought-after economic success in a systematic and dispassionate manner. [...] The pursuit of gain, in the region where it has become most completely unchained and stripped of its religious-ethical meaning, the United States, tends to be associated with purely competitive passions."[144]↩︎
"Life Expectancy, 1900–2000," Our World in Data. Data from Riley (2005); Zijdeman et al. (2015); HMD (2025); UN WPP (2024), with major processing by Our World in Data. ↩︎
Post-green revolution agricultural yields have made climate or agricultural shocks less likely to cascade into the anthropogenic failures that produce famine. Where famine persists in the modern era, its proximate causes are overwhelmingly associated with armed conflict.[145]↩︎
The lone dissenter, Justice Pierce Butler, filed no opinion, as was still common practice. The Court's only Catholic, his nomination four years earlier had drawn a vigorous anti-Catholic campaign from the Ku Klux Klan; his vote here, speculated to stem from his excessive popery.[146]↩︎
Justice Oliver Wendell Holmes, Jr., majority opinion in Buck v. Bell, 274 U.S. 200, 207 (1927). ↩︎
It should be noted that the factual basis for the case has also been subsequently disputed. It has been claimed that Carrie Buck, whom the case was around, was first committed as a cover-up of a familial rape, and had, based on school records, no signs of extreme mental deficiencies. Further, her counsel on the case, Irving Whitehead, was on the board of the institution Carrie was committed at, was friends with the doctor who wished to sterilize her and endorsed his previous sterilizations, and was old college friends with the state's attorney. Prior to the trial Whitehead never met with Carrie, and called no witnesses on her behalf.
This case thereafter stood as the basis for over 60,000 forced sterilizations in the US.[147]↩︎
"I do suggest, however, that the rules of instrumental rationality are singularly incapable of preventing such phenomena; that there is nothing in those rules which disqualifies the Holocaust-style methods of 'social-engineering' as improper or, indeed, the actions they served as irrational. I suggest, further, that the bureaucratic culture which prompts us to view society as an object of administration, as a collection of so many 'problems' to be solved, as 'nature' to be 'controlled', 'mastered' and 'improved' or 'remade', as a legitimate target for 'social engineering', and in general a garden to be designed and kept in the planned shape by force (the gardening posture divides vegetation into 'cultured plants' to be taken care of, and weeds to be exterminated), was the very atmosphere in which the idea of the Holocaust could be conceived, slowly yet consistently developed, and brought to its conclusion. And I also suggest that it was the spirit of instrumental rationality, and its modern, bureaucratic form of institutionalization, which had made the Holocaust-style solutions not only possible, but eminently 'reasonable' – and increased the probability of their choice."[148]↩︎
An emotivist or moral anti-realist structure is little comfort when confronted with calamity. ↩︎
At Nuremberg, Buck v. Bell was then used in defense to justify Nazi eugenic programs.[149]↩︎
See Intentio for a fuller treatment of this divergence in relation to social technology alongside a proposed solution.[101]↩︎
"...we had better be quite sure that the purpose put into the machine is the purpose which we really desire and not merely a colorful imitation of it."[127]↩︎
Throughout, references to 'AI', unless otherwise stated, should be read as referring to large language models or equivalent multimodal generative systems. ↩︎
Intentionally simplified, and probably of less use to those with an existing technical understanding of LLMs. Nevertheless, I have found the following useful in explaining to others, and so may still be of interest for its use as a pedagogical tool. ↩︎
Commonly referred to as 'pre-training'. The 'pre' here as against a subsequent task-specific 'fine-tuning', which has since broadened out into a myriad of different 'post-training' methods. ↩︎
Commonly referred to as 'post-training' or 'alignment'. ↩︎
Or more precisely 'tokens': the atomic chunks of text that models work on. ↩︎
Deliberately using 'model of the world' to avoid confusion with the distinct 'world model' AI approach. ↩︎
See Anthropic's interpretability research[150][151]on the structure of the model within the weights. ↩︎
This could be seen in older versions of GPT where, when asked something, the model would often continue the question rather than answer it (unless prompted in a way where answering seemed the more likely next part of the sequence). ↩︎
For a more extensive introduction still geared towards a general audience, see Andrej Karpathy's video series. ↩︎
See, for example, the claim that generating a 100-word email consumes a full bottle of water;[152]a figure later found to be orders of magnitude off.[153]↩︎
Related is Karen Hao's Empire of AI, which overstated relative data center water usage by over three orders of magnitude.[154]↩︎
"Even just going for a walk outside slightly wears out your sneakers. Typical athletic sneakers often last for 300-500 miles, and take 4 million ChatGPT prompts' worth of water to make.[155]This means that every mile you get out of your sneakers uses 8000 ChatGPT prompts' worth of water. Every single second you spend walking in sneakers uses enough water for 7 prompts.[156]At the end of an hour walk, you would have used enough water for 24,000 prompts."[157][158]↩︎
IEA, Energy and AI, World Energy Outlook Special Report (Paris: International Energy Agency, 2025), 56, 63, 93. ↩︎↩︎
This number relates to all data center emissions, of which AI, as of 2024, was thought to account for roughly 15% (though this percentage is likely to increase going forward).[63]↩︎
Given the most favorable numbers for the cement industry. The figure used, 1,470 Mt CO₂ in 2024, relates only to the emissions released in calcining limestone.[159]Taking into account the fuel used in cement kilns would meaningfully increase this number. ↩︎
The advent of hydraulic fracturing and horizontal drilling in shale plays has made energy supply increasingly elastic to demand over the last two decades in the US.[160]↩︎
See Aaron Swartz, widely held as a folk hero within the programming community. A co-author of RSS at age 14 and early architect of Creative Commons, then co-founder of Reddit, Swartz was arrested in 2011 for the bulk downloading of academic articles from JSTOR. Facing thirteen felony counts, he took his own life in January 2013, at the age of 26.[161]↩︎
"The limited scope of the copyright holder's statutory monopoly, like the limited copyright duration required by the Constitution, reflects a balance of competing claims upon the public interest: Creative work is to be encouraged and rewarded, but private motivation must ultimately serve the cause of promoting broad public availability of literature, music, and the other arts. The immediate effect of our copyright law is to secure a fair return for an 'author's' creative labor. But the ultimate aim is, by this incentive, to stimulate artistic creativity for the general public good. 'The sole interest of the United States and the primary object in conferring the monopoly,' this Court has said, 'lie in the general benefits derived by the public from the labors of authors.'[162]"[163]↩︎
"But while it is a moot question whether the origin of any kind of property is derived from nature at all, it would be singular to admit a natural and even an hereditary right to inventors. It is agreed by those who have seriously considered the subject, that no individual has, of natural right, a separate property in an acre of land, for instance. [...] Stable ownership is the gift of social law, and is given late in the progress of society. It would be curious then, if an idea, the fugitive fermentation of an individual brain, could, of natural right, be claimed in exclusive and stable property."[164]↩︎
"This right [copyright], as has been shown, does not exist at common law — it originated, if at all, under the acts of congress."[165]↩︎
These existed prior in case law from Folsom v. Marsh[166]in 1841, but were only explicitly codified under the Copyright Act of 1976. ↩︎
See Bartz v. Anthropic. Anthropic did settle in this case, but due to the retention of millions of pirated books for a permanent internal library. The claims of copyright infringement via training were rejected at summary judgment: "Regardless, the 'purpose and character' of using works to train LLMs was transformative — spectacularly so."[167]
Similarly, see Kadrey v. Meta where, again, even with the record viewed in the plaintiffs' favor, the copyright infringement claims regarding training were found so wanting that they were decided by summary judgment before reaching trial: "There is no serious question that Meta's use of the plaintiffs' books had a 'further purpose' and 'different character' than the books—that it was highly transformative."[168]
It should be noted that both these decisions are merely district court rulings, limiting their precedential weight. However, they do serve as good indications of broader judicial thinking on the topic. ↩︎
The implementation details on how this is set up are important. There is precedent in the space with the Audio Home Recording Act of 1992. Here, to safeguard artists from displacement caused by home copying, a levy of 2–3% (styled as royalties) was placed on 'digital audio recording devices' and 'digital audio recording media', with the proceeds to be distributed to affected parties.[169]However, due to how narrowly the category of 'digital audio recording device' was defined, devices such as MP3 players were held to fall outside the act,[170]greatly limiting its ultimate impact. ↩︎
To be clear, the argument here is strictly a legal one and against ideas that can induce passivity through belief that the ills currently attributed to AI already fall under some legal provision and so will be dealt with in due time. For a solution to some of these issues as they relate to intentional works of art, see Intentio.[101]↩︎
Commonly, 'temperature' is a parameter to tune how tokens are sampled, with lower levels leading to more determinism. However, in practice, even at temperature 0, there tends to still be some 'randomness' in output due to a lack of batch invariance in the GPU kernels.[171]↩︎
Conservatively. Had Waymo-equivalent autonomous vehicles ramped to even 5% of US vehicle-miles from their 2018 commercial introduction, ~7,200 lives would have been saved to date; at 80%, ~116,000.[172]↩︎
In the economic sense. Though not currently an official diagnosis, there does seem to be a growing issue related to 'AI mania' or 'AI psychosis' (or, more precisely, 'AI-associated delusions'[173]), from LLMs' tendency to reinforce delusional beliefs in vulnerable users.[174]↩︎
The stock has since cooled, but as of the last close at the time of writing (June 11, 2026; $3.88) still stands ~56% above the pre-announcement price (April 14, 2026; $2.49).[175]↩︎
The lower multiple comes despite higher margins and growth rate. Nvidia's fiscal year 2026 net margin stood at 55.6% against Cisco's 14.1% in fiscal year 2000, with revenue growing at a two-year CAGR of 88.3% against Cisco's 49.3%.[181]↩︎
There are some contrary measures here. For example, the S&P 500's cyclically adjusted price-to-earnings (CAPE) peaked at 44.20 in December 1999 and stands at 41.02 as of June 2026 (against a median of 16.6).[182]↩︎
Clayton M. Christensen, The Innovator's Dilemma: When New Technologies Cause Great Firms to Fail (Boston: Harvard Business School Press, 1997). ↩︎
Worryingly, there has been an increasing trend from the major hyperscalers to increase capex by tapping the bond market instead of utilizing free cash flows.[183]↩︎
In developed economies, energy infrastructure has an output elasticity of 0.054 and digital infrastructure an output elasticity of 0.070 (interestingly, even higher than the numbers for developing economies).[184]↩︎
Yes, for a period in Silicon Valley, a popular staple was a meal replacement drink named Soylent. An instrumentalization of food, which deliberately took its name from a literal instrumentalization of humanity. ↩︎
Alvaro S. Pereira, "The Opportunity of a Disaster: The Economic Impact of the 1755 Lisbon Earthquake," The Journal of Economic History 69, no. 2 (June 2009): 466–499. ↩︎
U.S. ten Brink et al., "Assessment of Tsunami Hazard to the U.S. Atlantic Margin,"Marine Geology 353 (2014): 45, citing J.T. Kozak, V.S. Moreira, and D.R. Oldroyd, Iconography of the 1755 Lisbon Earthquake (Prague: Academy of Sciences of the Czech Republic, 2005). ↩︎
Jean-Jacques Rousseau, "Letter to Voltaire on Providence," in The Discourses and Other Early Political Writings, ed. and trans. Victor Gourevitch, 2nd ed. (Cambridge: Cambridge University Press, 2019), 240–256. ↩︎
Immanuel Kant, "History and Natural Description of the Most Noteworthy Occurrences of the Earthquake that Struck a Large Part of the Earth at the End of the Year 1755," in Natural Science, ed. Eric Watkins, trans. Olaf Reinhardt (Cambridge: Cambridge University Press, 2012), 337–364. ↩︎
Jeremiah 2:19, The Holy Bible: A Translation from the Latin Vulgate in the Light of the Hebrew and Greek Originals, trans. Ronald Knox (London: Burns & Oates, 1955; repr. Baronius Press, 2012). ↩︎
Saint Irenaeus of Lyon, Against Heresies, Book IV, Chapter 37, §7, trans. Alexander Roberts and William Rambaut, in Ante-Nicene Fathers, vol. 1, ed. Alexander Roberts, James Donaldson, and A. Cleveland Coxe (Buffalo, NY: Christian Literature Publishing Co., 1885). ↩︎
Saint Augustine, Confessions, Book VII, Chapter 14, §20, trans. J.G. Pilkington, in Nicene and Post-Nicene Fathers, vol. 1, ed. Philip Schaff (Buffalo, NY: Christian Literature Publishing Co., 1887). ↩︎
Saint Thomas Aquinas, Summa TheologicaI, Q. 58, Art. 3; I, Q. 64, Art. 2, trans. Fathers of the English Dominican Province, 2nd rev. ed. (1920). ↩︎↩︎
Capitalization modified from referent; other additions are from referent. ↩︎
Blessed John Duns Scotus, Ordinatio II, d.7, nn.22, 59–60, trans. Peter L.P. Simpson (2023). ↩︎↩︎
Elisions added to referent; other additions are from referent. ↩︎
Ibn Sīnā, Remarks and Admonitions: Physics and Metaphysics, trans. Shams Constantine Inati (New York: Columbia University Press, 2014), Fourth Class, chs. 8 and 12, 122–123. ↩︎↩︎
Ibn Sīnā, The Book of Salvation. Based on Al-Najāt, ed. M.D.S. al-Kurdī (Cairo: Maṭbaʿat al-Saʿāda, 1938; repr. Tehran: Murtaẓavī, 1913). 'Metaphysics,' II.18–19 (al-Kurdī, 246–249), trans. Jon McGinnis and David C. Reisman, in Classical Arabic Philosophy: An Anthology of Sources (Indianapolis: Hackett Publishing, 2007), 219. 'Metaphysics' (al-Kurdī, 82.2–12), trans. Peter Adamson, in "From the Necessary Existent to God," Interpreting Avicenna: Critical Essays, ed. Peter S. Adamson (Cambridge: Cambridge University Press, 2013), 186. ↩︎↩︎
Ibn Rushd, The Incoherence of the Incoherence, trans. Simon Van Den Bergh, "E. J. W. Gibb Memorial" New Series XIX (Cambridge: E.J.W. Gibb Memorial Trust, 1954), 318–319. ↩︎
Marshall G. S. Hodgson, The Venture of Islam: Conscience and History in a World Civilization, vol. 1, The Classical Age of Islam (Chicago: University of Chicago Press, 1974), 414, 439, 476, 480. ↩︎↩︎↩︎↩︎↩︎
Dimitri Gutas, Greek Thought, Arabic Culture: The Graeco-Arabic Translation Movement in Baghdad and Early 'Abbasid Society (London: Routledge, 1998), 23–26, 95, 161, 192–93. ↩︎↩︎↩︎
George Makdisi, The Rise of Colleges: Institutions of Learning in Islam and the West (Edinburgh: Edinburgh University Press, 1981), 7, 75–78. ↩︎↩︎
Marshall G. S. Hodgson, The Venture of Islam: Conscience and History in a World Civilization, vol. 2, The Expansion of Islam in the Middle Periods (Chicago: University of Chicago Press, 1974), 35, 43, 47, 168, 171, 177, 180. ↩︎↩︎↩︎↩︎
A. C. S. Peacock, The Great Seljuk Empire (Edinburgh: Edinburgh University Press, 2015), 52. ↩︎
David J. Danelski, A Supreme Court Justice Is Appointed (New York: Random House, 1964), 92, 189–191. ↩︎
Paul A. Lombardo, Three Generations, No Imbeciles: Eugenics, the Supreme Court, and Buck v. Bell (Baltimore: Johns Hopkins University Press, 2008), introduction, chs. 5, 8, 10–11. ↩︎
Zygmunt Bauman, Modernity and the Holocaust (Ithaca, NY: Cornell University Press, 2001), ch. 1. ↩︎
Brandt Defense Document No. 53 (extract from Erich Ristov, Erbgesundheitsrecht [Law concerning hereditary health], Stuttgart-Berlin: Kohlhammer, 1935). ↩︎
US traffic deaths have averaged ≈ 39,300 per year over ≈ 3.2 trillion annual vehicle-miles ⟹ ~1.23 deaths per 100 million miles[185] Waymo, through its first 170 million driverless miles, recorded 92% fewer serious-or-fatal-injury crashes than human benchmarks on the same roads[186] Adoption ramps linearly from 2018 to a share s of vehicle-miles today (average s/2 across 8 years) Lives ≈ 3.2T × (s/2) × 1.23/100M × 0.92 × 8 s = 5% ⟹ ~900 per year ⟹ ~7,200 total s = 80% ⟹ ~14,500 per year ⟹ ~116,000 total ↩︎
This is, if anything, conservative. Cisco's fiscal year 2000 ended July 29, 2000, months after the March peak. Computing against the fiscal year 1999 diluted EPS of $0.29 gives an even greater multiple of ~276x.[177]↩︎
Cisco net income (fiscal year 2000) = $2,668M; revenue (fiscal years 1998, 2000) = $8,489M, $18,928M[177] Nvidia net income (fiscal year 2026) = $120,067M; revenue (fiscal years 2024, 2026) = $60,922M, $215,938M[179] Net margin: 2,668 / 18,928 ⟹ ~14.1%; 120,067 / 215,938 ⟹ ~55.6% CAGR = (end / start)^(1/n) − 1, over n = 2 years Cisco: (18,928 / 8,489)^(1/2) − 1 ⟹ ~49.3% Nvidia: (215,938 / 60,922)^(1/2) − 1 ⟹ ~88.3% ↩︎
Cross-posted from babbo.dev/articles/ai-risk. To experience the piece in its intended form, please visit there.
A Kishōtenketsu on AI Risk
Faces*, Madge Gill, 1920–1960. Newham Archives and Local Studies Library
I. The Will of Angels
A common topic of interest in 18th-century French salons was the theodicy:[3] given an omnipotent, omnibenevolent God, why is there evil in the world?[4] This excited a range of answers. There was the Panglossian theodicy, the pollyannaish view that, "all is for the best [in] [...] the best of all possible worlds".[5] Others subscribed to the Irenaean theodicy that evil is necessary for full human development.[6]
A particularly prominent theory though was one originally argued by Saint Augustine of Hippo. Pulling from the Neoplatonists, Augustine first contended that evil is not a thing in itself, but rather a privation of the good.[7] Then, in exercising free will, a great gift in itself, humans can move themselves away from the good.
However, there seems to be a slight hole in this notion. While the theodicy is adequate for explaining human-derived evil, it does not resolve the 'problem of evil' in regard to natural sources; if evil extends from human freedom, where do we place natural disasters, diseases, or independent animal suffering?[8]
To patch this issue, an addendum can be made to the theodicy in what is commonly referred to as the 'two falls theodicy'. In standard Catholic metaphysics, angels are aeviternal beings which apprehend 'intuitively' and whose wills, therefore, adhere immovably.[9][10] When an angel makes a choice, it persists forever. In the 'two falls theodicy', the non-human mediated ills of the world are downstream from these singular decisions of angels who reject the 'good'.[11]
A simple choice, made by a being of sufficient capacity, is enough to irrevocably distort the world. For most of human history, consequences were limited to our immediate positional and temporal surroundings. Today, actions engraved into the embeddings of systems have the capability of far outlasting their original context.[12]
Small changes to the starting state of a complex system can incite highly divergent outcomes. An object with a constant 'pop'[13] of 1 m/s⁶, assuming no other initial conditions, travels the circumference of the Earth in 56 seconds and achieves a velocity faster than the speed of light in 130.[14]
We stand now at an inflection point in history, where present choices can ramify forward without recall.[15]
The Supper (La Cena), Belkis Ayón, 1991. Tate Modern, London
II. Why the West Has (Not) Won;
or, The Incoherence of the Incoherence
Scientific development requires a specific epistemology. In order to proceed over time in scientific inquiry, one has to believe both that the world is knowable and that knowing the world is good.
Around 1027, ibn Sīnā[16] published The Book of Salvation. Within its second 'treatise', over two centuries before Aquinas' Five Ways, ibn Sīnā describes a proof of God arising from the contingent and yielding a necessary existent.[17] Given a necessary existent, then all else stems from it[18] and insofar as things stem from it they are knowable[19] and good.[20]
Less than 100 years later, al-Ghazālī, the 'mujaddid'[21] of the 11th century,[22] wrote the Incoherence of the Philosophers. Here al-Ghazālī argues for an Ash'arite occasionalism:
While the world may be good, it is not knowable.
Though not without dissent,[25] al-Ghazālī's approach proved ascendant. The madrasa system displaced the older patronage networks that sustained the Bayt al-Ḥikmah tradition. Natural philosophy made way for a curriculum which prioritized jurisprudence and Qur'ānic exegesis.[26] The Islamic world, which had preserved and expanded on the Western tradition in philosophy, math, and science[27] for half a millennium while Europe languished in darkness, became instead the recipient of European advancement.
The Latin West maintained the twin epistemologic pillars for slightly longer. Kepler presented astronomical inquiry as a way of knowing God.[28] A century later, Boyle's Christian Virtuoso defended experimental philosophy as a way to illuminate God's goodness.[29] In the General Scholium to the Principia, Newton held that the cosmic system "could only proceed from the counsel and dominion of an intelligent and powerful Being."[30]
However, with the Enlightenment, the idea that knowing the world is good began to fall. A possible theodicean response is to break down the premises. One cannot establish an 'ought' from an 'is',[31] so without an explicit moral ordering of the universe where is value derived? Scholastic vestiges remained, yet unmoored from any metaphysical anchor.[32]
In ethics, the good became restructured around the "two sovereign masters" of pain and pleasure;[33] efficiency the dominant signal in markets.[34] Without a telos, purpose came from what could be measured: "anything which does not conform to the standard of calculability and utility must be viewed with suspicion".[35] The value of discovery limited to its instrumentalization.
Modernity still has its triumphs. While the tractable is not identical to the good, the mode of orientation can be near enough to appear aligned in non-pathologic scenarios. In the 20th century alone, life expectancy doubled.[36] Famine has become a peripheral phenomenon primarily associated with war rather than agriculture.[37] Smallpox, the killer of over half a billion, has been eradicated.[38]
Its failure in sensing the inherent dignity of the person, the temptation towards the merely legible, though, curses the regime of modernity to eventually misalign from the good.
In Buck v. Bell, the United States Supreme Court, in an 8-1 decision,[39] upheld the forced sterilization of the 'unfit' to promote the general health and welfare of society:
In the Holocaust, perhaps the central tragedy of the modern age, modernity again showed its fruits. Though history is littered with stories of mass killings, the Holocaust was not a harkening back to a barbaric past. It is distinguished, not just in magnitude but also in kind, by a particular rationality. The genocide was taken as a means, and correct from within its own axiomatic structure.[42] Without an external understanding of the good, there is no end toward which to correctly direct one's rationality.[43][44]
Our current tech landscape has provided remarkable achievements, yet there still remains an absence at its core. As we pursue greater advancements, the divergence between the efficient and the good becomes increasingly pronounced.[45] While in an age of relative abundance, we are tightroping across a path where, with rising capability, a small deviation can quickly cascade into a fall.[46] Yet the system, having lasted hundreds of years, resists correction; corrigibility to the numinous remains strained.
The Rabbis (Rabinos), Leonora Carrington, 1960. Rose Art Museum, Brandeis University
III. Sed Contra
While the development of artificial intelligence carries grave risk and must be taken with the utmost care, many standard objections serve to obscure the genuine concern rather than clarify.[47] These critiques both inflate and diminish capability while positing issues orthogonal to any real capacity.
On Stochastic Parrots
A common critique[48] given to modern generative AI systems is that they are merely 'stochastic parrots'[49] or 'next token predictors'. The tricky part of this statement is that there is some level where this is correct.
A Brief Digression on the Nature of LLMs[50]
There are two main phases in the life-cycle of an LLM. The 'building' phase[51]and the 'usage' phase.[52]This building phase can then be split further into constructing out the 'foundation'[53]and then 'sculpting' it.[54]
In the initial foundation building phase, a large corpus of text is taken and strictly from how words[55] are used in relation to each other in the corpus, a high-dimensional 'raw' 'model of the world'[56] is formed.[57] Then, in the sculpting phase, this raw model of the world is shaped to be more amenable as a human assistant.[58] Finally, in the 'usage' phase your input goes into this sculpted model of the world and based on the structure within, emits a response.[59]
There is a sense that when prompting, an LLM is 'predicting' the next token in a sequence. However, emergent phenomena are the norm in complex systems, and when exhibited, generally make a description of parts to explain the whole not 'untrue' but misleading or non-helpful. Outside of highly specific scenarios, it is not useful to explain a computer as a series of transistors or further to describe any matter according to its fermionic makeup.
Most commonly, this type of argumentation seems to occur when one is trying to make a claim to separate human cognition from what occurs in LLMs. However, mereological observation does not constitute ontological proof. It is clear from even minor use that LLMs are not just stochastic parrots. Statements on what LLMs 'are' should be made directly, not laundered through a framing which falls apart on contact.
On the Environment
Despite apocalyptic hyperventilation to the contrary from media,[60][61] the environmental footprint for AI is fairly small. Obviously, as something that exists in the world, LLMs and their associated infrastructure have some environmental effect, but relative to their use this is negligible and possibly even negative when considering displacement costs.[62]
A similar story exists for emissions. Even in a 'lift-off' case with extremely strong AI uptake, limited local constraints, and low efficiency improvements in hardware or software, the IEA estimates data center emissions to only be at 475 Mt CO₂ by 2035[63][64] or less than ⅓ the output of the cement industry.[65]
This is not to say that data centers are immune to causing grid strain at a local level.[66] However, even here, energy supply is historically highly receptive to demand.[67] For those genuinely concerned about environmental degradation, the returns on attention are orders of magnitude higher in other areas.
On Intellectual Property
It has been a mainstay of internet culture, especially within tech circles, to be against intellectual property or believe it overly restrictive;[68] the operating assumption being that information should be free.[69] However, with the advent of LLMs, this notion has started to shift. There is a feeling that in AI's unmediated hoovering of data, some implicit boundary has been violated.
In the United States, copyright has always been something granted for the public benefit: "[t]o promote the Progress of Science and useful Arts".[70] Due to the inherent non-excludability of ideas, copyright exists as a way to incentivize their creation and discovery.[71]Copyright is merely a statutory concession for the good of the public and not a protection that flows from one's natural rights.[72][73]
To maximize the public welfare, even in the period of exclusivity provided, an author's rights are not absolute. The bounds on the exclusive rights of works were codified[74] in the Copyright Act of 1976 which established a four-factor test that utilized the purpose of use, the nature of the copyrighted work, the amount used, and the market impact on the copyrighted work to determine if a secondary work was a 'fair use' of the copyrighted material.[75]
With regards to AI, the use of copyrighted materials for training has been found to constitute fair use due to its highly transformative nature.[76] While whether the outputs of models themselves can constitute copyright infringement is under ongoing litigation,[77][78] copyright is a poor mechanism for addressing the displacement AI creates.
If we take AI displacement in specific industries to be bad, there are more direct and efficient ways to alleviate the harms. A Pigouvian tax on LLM providers, earmarked for affected industries, could compensate creators without requiring the increasingly strained application of a framework designed for a different issue.[79][80]
On Hallucination
One of the most tangible concerns people have of LLMs is related to having factual errors in outputs or 'hallucinations'. To clarify against what is a slightly common misconception, hallucinations do not come from some inherent 'randomness'[81] of LLMs, but rather through 'mistakes' in the models of the world being inferred into.[82]
The upside of this is that in using larger pre-training corpora and more robust post-training processes, successive model generations have significantly reduced hallucination rates while simultaneously becoming better calibrated in expressing uncertainty.[83]
A common mistake in judging digital output is comparison to an ideal form rather than existing alternatives. Compared to the counterfactual, existing deployment delays on autonomous vehicles have already cost thousands of lives.[84][85] Even in fields such as medical diagnosis with high potential consequences for inaccuracy, LLMs have been found, in certain circumstances, to exceed physician accuracy.[86]
This is not to say that hallucinations are not an issue, but rather that any sufficiently complex system will exhibit failure modes. When examining AI errors, it should be done in cognizance of their current rate of improvement and relative to the systems they stand to augment.
On Mania[87]
A final critique often levied against the AI industry as a whole is that it represents a speculative bubble whose eventual correction will inflict widespread economic harm. Stories of Allbirds, the shoe company, whose shares climbed over 600% after announcing a pivot to AI compute infrastructure,[88][89] make it hard to deny some mania in the markets.
However, it is perhaps clarifying here to view the AI industry in relation to its last major speculative episode: the dot-com bubble. In March of 2000, directly prior to the dot-com burst, information technology forward P/E stood at ~48.3x.[90] Today, due to massive earnings increases, the mark stands at only ~24.2x.[91] If we look at the 'shovel' play in both periods, Cisco had a P/E of ~222.4x, while today Nvidia still only stands at 41.8x.[92][93] Insofar as AI is a bubble, its 'irrational exuberance' still has room to grow.[94]
Even after any excess froth gets wrung out, the fallout is unlikely to follow the same patterns as some previous recent corrections. For established tech companies, AI is best thought of as a 'sustaining' rather than a 'disruptive' innovation.[95] Following from this, a majority of capex has come from free cash flow of highly profitable existing tech companies.[96] Compared to the dot-com crash, which was funded by retail speculation,[97] or 2008, which was built on leveraged consumer debt,[98] the current situation carries less structural risk. Further, in the case of a downturn, the majority of the build-out here is in compute and energy, which besides being highly fungible also historically have high positive externalities.[99]
Ten Virgins, Tove Jansson, 1953. Teuva Church, Finland
IV. Prohairesis
Despite their inadequacies, there is a reason why these anti-AI arguments persist. People have an intuitive perception that there is something hollow at the center of the modern world and so glom onto the first rationale with even a façade of logical coherence.
Those who make claims of ushering in a new age are of the same group that brought us Soylent,[100] engagement-optimized feeds,[101] and claims of massive AI job displacement.[102][103] Even when AI leaders make remarks pointing to more humanistic ends,[104] it is hard to take these outside the milieu they operate within: "[W]ho could endure the Gracchi railing at sedition?"[105]
It is well understood across a range of fields that negative signals carry more weight than positive signals.[106] In decision-making under uncertainty, losses carry twice the weight of equivalent gains.[107] A common response then is to treat the asymmetry as a bias to be corrected and so actively discount any negative stimuli.[108] However, occasionally, several hundred million years of evolutionary selection encodes valuable information; Taleb distributions are real and often non-obvious.[109]
So where do we go from here? One must walk away from St. Petersburg; silence the martingale. In acting, we immediately shape our soul before any external consequence. To be truly in zugzwang requires a fully staid environment; however, we can change the conditions under which we operate.
Technology qua technology is not a good in itself. Its pharmakonic nature is decided in relation to the telos on which it is produced. We currently exist in an axial moment where choices made today can have an outsized effect on the future. In proceeding, one must act with caution, humility, and an understanding that in building, the 'ends' must be aligned to and in respect of the fundamental dignity of the human person.
Thanks to my family for reviewing earlier drafts of this work.
For any comments on this piece, please message me at joseph@babbo.dev.
The classic golem narrative has no singular authoritative source. However, Berthold Auerbach's biographical novel of Spinoza is thought to contain one of its earliest literary renderings.[110]↩︎ ↩︎
Berthold Auerbach, Spinoza: A Novel, trans. E. Nicholson (New York: H. Holt and Company, 1882), 216–217. ↩︎ ↩︎
Coined by Leibniz in the eponymous Essais de Théodicée (1710).[111]↩︎
Especially after the 1755 Lisbon Earthquake, where on All Saints' Day an estimated 8.7-magnitude[112]earthquake struck Portugal, killing tens of thousands, destroying up to 48% of Portuguese GDP,[113]and generating waves felt as far as Brazil,[114]inspiring writings from Voltaire,[115]Rousseau,[116]and Kant.[117]↩︎
From Dr. Pangloss in Candide,[115]whom Voltaire introduces to caricature Leibniz's rosy position from the Théodicée.[111]↩︎
"Just in the same way is the heavenly kingdom honourable to those who have known the earthly one. But in proportion as it is more honourable, so much the more do we prize it; and if we have prized it more, we shall be the more glorious in the presence of God. The Lord has therefore endured all these things on our behalf, in order that we, having been instructed by means of them all, may be in all respects circumspect for the time to come, and that, having been rationally taught to love God, we may continue in His perfect love: for God has displayed long-suffering in the case of man's apostasy; while man has been instructed by means of it, as also the prophet says, 'Your own apostasy shall heal you;'[118]God thus determining all things beforehand for the bringing of man to perfection, for his edification, and for the revelation of His dispensations, that goodness may both be made apparent, and righteousness perfected, and that the Church may be fashioned after the image of His Son, and that man may finally be brought to maturity at some future time, becoming ripe through such privileges to see and comprehend God."[119]↩︎
This was required as, if evil itself were a substance, it would necessitate either origination in God, undermining omnibenevolence; or a Manichaean dualism: "And, because my soul dared not be displeased at my God, it would not allow anything to be Yours which displeased it. Hence it had gone into the opinion of two substances, and resisted not, but talked foolishly."[120]↩︎
Even if one presumes that non-human evil propagates from 'the Fall', this would not explain, in light of evolution, pre-fall animal suffering (or even the serpent in the traditional narrative) without positing some exotic causality. ↩︎
Another way to put this is that angels understand non-discursively; everything is grasped immediately and in full, not through a chain of syllogisms. As there is nothing left to further understand, there is no basis for a change of will.
"Such is the condition of the angels, because in the truths which they know naturally, they at once behold all things whatsoever that can be known in them."[121]
"Now the angel's apprehension differs from man's in this respect, that the angel by his intellect apprehends immovably [...] Consequently [...] the angel's will adheres fixedly and immovably."[121]↩︎
This view, however, is not entirely uncontested. Scotus argued non-discursive knowing is non-corollary to immovable will:
"[A]lthough the angel understand non-discursively what – according to them – man understands discursively, yet the intellect of man does not movably adhere to that which it reaches discursively; for it so holds with certitude (that is without doubting) the conclusion it discursively reaches as the angel holds it by seeing it in the principle non-discursively; therefore this immobility of the human intellect (that is, certitude) would have an equally immovable will just as that other [immovability] posited in the angel."[122][123]
but rather, there is still a permanency due to lack of grace from God to further move oneself back to the good; a divine dereliction:
"[F]or springing back meritoriously there is required another principle than the will, namely grace, which a bad angel cannot have of himself, – and God, according to the fact he deserted him, has disposed not to give it to him [...] [;] there not be another cause of their permanence in bad save divine dereliction."[123]
Though, it should be noted, the word 'dunce' originated to describe the followers of Duns Scotus.[124]↩︎
More recent formulations posit this as an angelic 'omission' (or 'archon abandonment'), in which natural evil is traced not to direct angelic action but to the culpable abdication of a divinely assigned stewardship over creation.[125]↩︎
By the maker upon the made, at first; in time, by the made upon the maker. ↩︎
The fourth, fifth, and sixth derivatives of position with respect to time are commonly referred to as, respectively, snap, crackle, and pop. Yes, after the Rice Krispies mascots.[126]↩︎
x(t) = 0 + 0·t + (0/2)t² + (0/6)t³ + (0/24)t⁴ + (0/120)t⁵ + (1/720)t⁶
x(t) = t⁶/720; v(t) = t⁵/120
Circumference of the Earth ≈ 40,075 km
Speed of light c ≈ 2.998e8 m/s
t⁶/720 = 4.0075e7 m ⟹ t ≈ 55.38 s
t⁵/120 = c ⟹ t ≈ 129.18 s ↩︎
"If we use, to achieve our purposes, a mechanical agency with whose operation we cannot efficiently interfere once we have started it, because the action is so fast and irrevocable that we have not the data to intervene before the action is complete, then..."[127]↩︎
From ibn, meaning 'son of'; the original referent is unknown. In the West, commonly referred to as Avicenna. ↩︎
Most thoroughly laid out in The Book of Salvation; a more succinct formulation can be found in Remarks and Admonitions:
"Every totality having every one of its units as caused requires a cause external to its units. This is because either [1] it does not require a cause at all; hence it is necessary and not possible. But how could this be so when it is only necessitated by its units? [2] It requires a cause that is all its units; hence it is caused by itself. [...] [3] It requires a cause that is some of its units. But if every one of its units is caused, then some of its units are not more deserving of being the cause than some others. The reason is that the cause of the caused is more deserving of being the cause. Or [4] it requires a cause external to all its units. This is the remaining [truth]."[128][129]↩︎
"Thus, if there is a first cause, it is a cause of every existence and of the cause of the reality of every concrete existence."[129]↩︎
"Know that you came to perceive particular eclipses only because you fully comprehended its causes and everything concerning the heavens. When full comprehension takes place about all of the causes in things and their existence, there is a transference [of that full comprehension] from those to all of the effects."[130][131]↩︎
"Whatever necessarily exists through itself is pure good and pure perfection. In general, the good is that which everything desires, and that through which it is completed. Evil has no essence, but is either the absence of substance or of some state beneficial to the substance. Thus existence is goodness. Perfection of existence is goodness of existence, the existence that is untouched by absence — whether absence of substance or absence of something that belongs to substance — rather, it is perpetually actual. So it is pure good. That which, through itself, is [merely] possibly existent is not pure good. For its essence does not, through itself, have existence, so that its essence in itself suffers absence. What suffers absence in any way is not entirely free of evil and deficiency. Therefore there is no pure good apart from that which necessarily exists through itself."[132][131]↩︎
Mohd Rosmizi and Salih Yucel, "The Mujaddid of His Age: Al-Ghazali and His Inner Spiritual Journey," UMRAN — International Journal of Islamic and Civilizational Studies 3, no. 2 (2016): 1–13. ↩︎
The 5th century of the hijrī calendar (1009–1106 AD; 401–500 AH). ↩︎
Elisions added to referent; other additions are from referent. ↩︎
Al-Ghazālī, The Incoherence of the Philosophers, trans. Michael E. Marmura, 2nd ed. (Provo, UT: Brigham Young University Press, 2000), Discussion 17, §§2–5, 166–167. ↩︎
Most famously in Ibn Rushd's[133]Incoherence of the Incoherence:
"To deny the existence of efficient causes which are observed in sensible things is sophistry, and he who defends this doctrine either denies with his tongue what is present in his mind or is carried away by a sophistical doubt which occurs to him concerning this question. [...] Logic implies the existence of causes and effects, and knowledge of these effects can only be rendered perfect through knowledge of their causes. Denial of cause implies the denial of knowledge, and denial of knowledge implies that nothing in this world can be really known".[134]↩︎
It is slightly overdrawn to lay this entirely at the feet of al-Ghazālī. He is probably better understood as merely a seminal figure in an ongoing process than the immediate catalyst of the change (if such a thing could be allowed under Ash'arite metaphysics).
Beginning in the 700s, the Mu'tazili Islamic school of theology was formed. Mu'tazilis controversially believed the Qur'ān to be created, not coeternal, and took heavy influence from Greek rationalism, utilizing kalām, or the rational study of theology, to try to ground their beliefs.[135]Starting under the reign of al-Ma'mūn, Mu'tazilism gained increased prominence in the Abbasid empire, to the extent where in 833 the miḥna was established, where officials had to affirm a belief in Qur'ānic createdness under threat of persecution.[135][136]
The eccentricity of Mu'tazilism and the heavy-handedness under which it was enforced festered resentment among the ulamā, the Muslim religious scholar class.[136]The resulting backlash when al-Mutawakkil eventually reversed the miḥna in 851 saw the rise of the more traditional Ḥanbalī school, which rejected all kalām, and the reverse persecution of Mu'tazilis.[137]This also saw the creation of the Ash'arī school from an ex-Mu'tazilī, which sought to still use kalām but solely in the defence of traditional doctrine.[135]
With the rise of the Shī'a Buyids in 945, no Sunnī orthodoxy was enforced and Mu'tazilis and independent faylasūf thinkers, such as ibn Sīnā, were again permitted and sponsored, while the Ḥanbalī and Ash'arī factions competed among the ulamā.[138]When the Buyids fell to the Sunnī Seljuks in 1055, vizier al-Kundurī sided with the Ḥanbalīs and attempted to suppress all kalām, including the Ash'arīs.[138]However, upon the death of Ṭughril in 1063, al-Kundurī supported the infant son of Chaghrī Beg, Ṭughril's brother, over the elder son Alp Arslān, who with the help of Niẓām al-Mulk became the next sultan.[139]Upon himself becoming the vizier, Niẓām al-Mulk chose Ash'arism as the official state theology and sponsored the building of Niẓāmiyyas, a highly influential series of madrasas established as waqf whose endowment terms restricted instruction to the Islamic sciences.[137]He appointed al-Ghazālī the head of the most prominent, in Baghdad, in 1091.[138]↩︎
Besides preserving, spreading, and translating works from Greek, Persian, and Indian sources (itself a valuable work),[136]Islamic scholars made foundational contributions to the scientific corpus. Al-Khwārizmī's algebra,[135]the optical insights of al-Haytham,[138]and al-Battānī's astronomical discoveries[135]meaningfully furthered the Western world's canon of knowledge. ↩︎
"It is in this that Paul proposes to the Gentiles that they should contemplate God like the Sun in water or in a mirror. Why then as Christians should we take any less delight in its contemplation, since it is for us with true worship to honor God, to venerate him, to wonder at him? The more rightly we understand the nature and scope of what our God has founded, the more devoted the spirit in which that is done."[140]↩︎
"The consideration of the Vastness, Beauty, and Regular Motions, of the heavenly Bodies; the excellent Structure of Animals and Plants; besides a multitude of other Phænomena of Nature, and the Subserviency of most of these to Man; may justly induce him, as a Rational Creature, to Conclude, That this vast, beautiful, orderly, and (in a word) many ways admirable System of things, that we call the World, was fram'd by an Author supremely Powerful, Wise, and Good, can scarce be deny'd by an intelligent and unprejudic'd Considerer."[141][142]↩︎
Isaac Newton, Principia: The Mathematical Principles of Natural Philosophy, trans. Andrew Motte (1729), General Scholium. ↩︎
David Hume, A Treatise of Human Nature, Book III, Part I, Section I (1739). ↩︎
"'Morally wrong' is the term which is the heir of the notion 'illicit,' or 'what there is an obligation not to do'; which belongs in a divine law theory or ethics. [...] [I]t is because 'morally wrong' is the heir of this concept, but an heir that is cut off from the family of concepts from which it sprang, that 'morally wrong' both goes beyond the mere factual description 'unjust' and seems to have no discernible content except a certain compelling force, which I should call purely psychological."[143]↩︎
Jeremy Bentham, An Introduction to the Principles of Morals and Legislation (1789), ch. I, §1. ↩︎
"The capitalist economy rationalizes on the basis of strictly quantitative calculations and is oriented to the sought-after economic success in a systematic and dispassionate manner. [...] The pursuit of gain, in the region where it has become most completely unchained and stripped of its religious-ethical meaning, the United States, tends to be associated with purely competitive passions."[144]↩︎
Max Horkheimer and Theodor W. Adorno, Dialectic of Enlightenment: Philosophical Fragments, trans. Edmund Jephcott (Stanford: Stanford University Press, 2002), 3. ↩︎
"Life Expectancy, 1900–2000," Our World in Data. Data from Riley (2005); Zijdeman et al. (2015); HMD (2025); UN WPP (2024), with major processing by Our World in Data. ↩︎
Post-green revolution agricultural yields have made climate or agricultural shocks less likely to cascade into the anthropogenic failures that produce famine. Where famine persists in the modern era, its proximate causes are overwhelmingly associated with armed conflict.[145]↩︎
F. Fenner et al., Smallpox and its Eradication (Geneva: World Health Organization, 1988). ↩︎
The lone dissenter, Justice Pierce Butler, filed no opinion, as was still common practice. The Court's only Catholic, his nomination four years earlier had drawn a vigorous anti-Catholic campaign from the Ku Klux Klan; his vote here, speculated to stem from his excessive popery.[146]↩︎
Justice Oliver Wendell Holmes, Jr., majority opinion in Buck v. Bell, 274 U.S. 200, 207 (1927). ↩︎
It should be noted that the factual basis for the case has also been subsequently disputed. It has been claimed that Carrie Buck, whom the case was around, was first committed as a cover-up of a familial rape, and had, based on school records, no signs of extreme mental deficiencies. Further, her counsel on the case, Irving Whitehead, was on the board of the institution Carrie was committed at, was friends with the doctor who wished to sterilize her and endorsed his previous sterilizations, and was old college friends with the state's attorney. Prior to the trial Whitehead never met with Carrie, and called no witnesses on her behalf.
This case thereafter stood as the basis for over 60,000 forced sterilizations in the US.[147]↩︎
"I do suggest, however, that the rules of instrumental rationality are singularly incapable of preventing such phenomena; that there is nothing in those rules which disqualifies the Holocaust-style methods of 'social-engineering' as improper or, indeed, the actions they served as irrational. I suggest, further, that the bureaucratic culture which prompts us to view society as an object of administration, as a collection of so many 'problems' to be solved, as 'nature' to be 'controlled', 'mastered' and 'improved' or 'remade', as a legitimate target for 'social engineering', and in general a garden to be designed and kept in the planned shape by force (the gardening posture divides vegetation into 'cultured plants' to be taken care of, and weeds to be exterminated), was the very atmosphere in which the idea of the Holocaust could be conceived, slowly yet consistently developed, and brought to its conclusion. And I also suggest that it was the spirit of instrumental rationality, and its modern, bureaucratic form of institutionalization, which had made the Holocaust-style solutions not only possible, but eminently 'reasonable' – and increased the probability of their choice."[148]↩︎
An emotivist or moral anti-realist structure is little comfort when confronted with calamity. ↩︎
At Nuremberg, Buck v. Bell was then used in defense to justify Nazi eugenic programs.[149]↩︎
See Intentio for a fuller treatment of this divergence in relation to social technology alongside a proposed solution.[101]↩︎
"...we had better be quite sure that the purpose put into the machine is the purpose which we really desire and not merely a colorful imitation of it."[127]↩︎
Throughout, references to 'AI', unless otherwise stated, should be read as referring to large language models or equivalent multimodal generative systems. ↩︎
Though rarer to see these days. ↩︎
Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell, "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜," in Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (2021): 610–623. ↩︎
Intentionally simplified, and probably of less use to those with an existing technical understanding of LLMs. Nevertheless, I have found the following useful in explaining to others, and so may still be of interest for its use as a pedagogical tool. ↩︎
Commonly referred to as 'training'. ↩︎
Commonly referred to as 'inference'. ↩︎
Commonly referred to as 'pre-training'. The 'pre' here as against a subsequent task-specific 'fine-tuning', which has since broadened out into a myriad of different 'post-training' methods. ↩︎
Commonly referred to as 'post-training' or 'alignment'. ↩︎
Or more precisely 'tokens': the atomic chunks of text that models work on. ↩︎
Deliberately using 'model of the world' to avoid confusion with the distinct 'world model' AI approach. ↩︎
See Anthropic's interpretability research[150][151]on the structure of the model within the weights. ↩︎
This could be seen in older versions of GPT where, when asked something, the model would often continue the question rather than answer it (unless prompted in a way where answering seemed the more likely next part of the sequence). ↩︎
For a more extensive introduction still geared towards a general audience, see Andrej Karpathy's video series. ↩︎
See, for example, the claim that generating a 100-word email consumes a full bottle of water;[152]a figure later found to be orders of magnitude off.[153]↩︎
Related is Karen Hao's Empire of AI, which overstated relative data center water usage by over three orders of magnitude.[154]↩︎
"Even just going for a walk outside slightly wears out your sneakers. Typical athletic sneakers often last for 300-500 miles, and take 4 million ChatGPT prompts' worth of water to make.[155]This means that every mile you get out of your sneakers uses 8000 ChatGPT prompts' worth of water. Every single second you spend walking in sneakers uses enough water for 7 prompts.[156]At the end of an hour walk, you would have used enough water for 24,000 prompts."[157][158]↩︎
IEA, Energy and AI, World Energy Outlook Special Report (Paris: International Energy Agency, 2025), 56, 63, 93. ↩︎ ↩︎
This number relates to all data center emissions, of which AI, as of 2024, was thought to account for roughly 15% (though this percentage is likely to increase going forward).[63]↩︎
Given the most favorable numbers for the cement industry. The figure used, 1,470 Mt CO₂ in 2024, relates only to the emissions released in calcining limestone.[159]Taking into account the fuel used in cement kilns would meaningfully increase this number. ↩︎
Catherina Gioino, "Nearly 50,000 Lake Tahoe Residents Have to Find a New Power Source After Their Energy Source Looks to Redirect Lines to Data Centers," Fortune, May 12, 2026. ↩︎
The advent of hydraulic fracturing and horizontal drilling in shale plays has made energy supply increasingly elastic to demand over the last two decades in the US.[160]↩︎
See Aaron Swartz, widely held as a folk hero within the programming community. A co-author of RSS at age 14 and early architect of Creative Commons, then co-founder of Reddit, Swartz was arrested in 2011 for the bulk downloading of academic articles from JSTOR. Facing thirteen felony counts, he took his own life in January 2013, at the age of 26.[161]↩︎
Stewart Brand, remarks at the first Hackers Conference (November 1984), in "'Keep Designing': How the Information Economy Is Being Created and Shaped by the Hacker Ethic," Whole Earth Review, no. 46 (May 1985): 49. ↩︎
U.S. Const. art. I, § 8, cl. 8. ↩︎
"The limited scope of the copyright holder's statutory monopoly, like the limited copyright duration required by the Constitution, reflects a balance of competing claims upon the public interest: Creative work is to be encouraged and rewarded, but private motivation must ultimately serve the cause of promoting broad public availability of literature, music, and the other arts. The immediate effect of our copyright law is to secure a fair return for an 'author's' creative labor. But the ultimate aim is, by this incentive, to stimulate artistic creativity for the general public good. 'The sole interest of the United States and the primary object in conferring the monopoly,' this Court has said, 'lie in the general benefits derived by the public from the labors of authors.'[162]"[163]↩︎
"But while it is a moot question whether the origin of any kind of property is derived from nature at all, it would be singular to admit a natural and even an hereditary right to inventors. It is agreed by those who have seriously considered the subject, that no individual has, of natural right, a separate property in an acre of land, for instance. [...] Stable ownership is the gift of social law, and is given late in the progress of society. It would be curious then, if an idea, the fugitive fermentation of an individual brain, could, of natural right, be claimed in exclusive and stable property."[164]↩︎
"This right [copyright], as has been shown, does not exist at common law — it originated, if at all, under the acts of congress."[165]↩︎
These existed prior in case law from Folsom v. Marsh[166]in 1841, but were only explicitly codified under the Copyright Act of 1976. ↩︎
17 U.S.C. § 107. ↩︎
See Bartz v. Anthropic. Anthropic did settle in this case, but due to the retention of millions of pirated books for a permanent internal library. The claims of copyright infringement via training were rejected at summary judgment: "Regardless, the 'purpose and character' of using works to train LLMs was transformative — spectacularly so."[167]
Similarly, see Kadrey v. Meta where, again, even with the record viewed in the plaintiffs' favor, the copyright infringement claims regarding training were found so wanting that they were decided by summary judgment before reaching trial: "There is no serious question that Meta's use of the plaintiffs' books had a 'further purpose' and 'different character' than the books—that it was highly transformative."[168]
It should be noted that both these decisions are merely district court rulings, limiting their precedential weight. However, they do serve as good indications of broader judicial thinking on the topic. ↩︎
Complaint, The New York Times Co. v. Microsoft Corp., No. 1:23-cv-11195 (S.D.N.Y. Dec. 27, 2023), ECF No. 1. ↩︎
Complaint, Disney Enterprises, Inc. v. Midjourney, Inc., No. 2:25-cv-05275 (C.D. Cal. June 11, 2025), ECF No. 1. ↩︎
The implementation details on how this is set up are important. There is precedent in the space with the Audio Home Recording Act of 1992. Here, to safeguard artists from displacement caused by home copying, a levy of 2–3% (styled as royalties) was placed on 'digital audio recording devices' and 'digital audio recording media', with the proceeds to be distributed to affected parties.[169]However, due to how narrowly the category of 'digital audio recording device' was defined, devices such as MP3 players were held to fall outside the act,[170]greatly limiting its ultimate impact. ↩︎
To be clear, the argument here is strictly a legal one and against ideas that can induce passivity through belief that the ills currently attributed to AI already fall under some legal provision and so will be dealt with in due time. For a solution to some of these issues as they relate to intentional works of art, see Intentio.[101]↩︎
Commonly, 'temperature' is a parameter to tune how tokens are sampled, with lower levels leading to more determinism. However, in practice, even at temperature 0, there tends to still be some 'randomness' in output due to a lack of batch invariance in the GPU kernels.[171]↩︎
Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala, and Edwin Zhang, "Why Language Models Hallucinate," arXiv:2509.04664 (September 2025). ↩︎
"AA-Omniscience: Knowledge and Hallucination Benchmark," Artificial Analysis. ↩︎
Conservatively. Had Waymo-equivalent autonomous vehicles ramped to even 5% of US vehicle-miles from their 2018 commercial introduction, ~7,200 lives would have been saved to date; at 80%, ~116,000.[172]↩︎
Nidhi Kalra and David G. Groves, The Enemy of Good: Estimating the Cost of Waiting for Nearly Perfect Automated Vehicles, RR-2150-RC (Santa Monica, CA: RAND Corporation, 2017). ↩︎
Ethan Goh et al., "Large Language Model Influence on Diagnostic Reasoning: A Randomized Clinical Trial," JAMA Network Open 7, no. 10 (2024): e2440969. ↩︎
In the economic sense. Though not currently an official diagnosis, there does seem to be a growing issue related to 'AI mania' or 'AI psychosis' (or, more precisely, 'AI-associated delusions'[173]), from LLMs' tendency to reinforce delusional beliefs in vulnerable users.[174]↩︎
Allison Morrow, "Allbirds Shares Soar on a Very 2026 Pivot to AI," CNN Business, April 15, 2026. ↩︎
The stock has since cooled, but as of the last close at the time of writing (June 11, 2026; $3.88) still stands ~56% above the pre-announcement price (April 14, 2026; $2.49).[175]↩︎
Edward Yardeni and Joseph Abbott, Predicting the Markets Topical Study #4: S&P 500 Earnings, Valuation, and the Pandemic (YRI Press, 2020), 28. ↩︎
Yardeni Research, "On Technology, Semiconductors & Fusion," Morning Briefing, May 14, 2026. ↩︎
Cisco closing price (March 27, 2000) = $80.06[176]
Cisco diluted EPS (fiscal year 2000) = $0.36[177]
Nvidia closing price (June 11, 2026) = $204.87[178]
Nvidia diluted EPS (fiscal year 2026) = $4.90[179]
Cisco: 80.06 / 0.36 ⟹ ~222.4x[180]
Nvidia: 204.87 / 4.90 ⟹ ~41.8x ↩︎
The lower multiple comes despite higher margins and growth rate. Nvidia's fiscal year 2026 net margin stood at 55.6% against Cisco's 14.1% in fiscal year 2000, with revenue growing at a two-year CAGR of 88.3% against Cisco's 49.3%.[181]↩︎
There are some contrary measures here. For example, the S&P 500's cyclically adjusted price-to-earnings (CAPE) peaked at 44.20 in December 1999 and stands at 41.02 as of June 2026 (against a median of 16.6).[182]↩︎
Clayton M. Christensen, The Innovator's Dilemma: When New Technologies Cause Great Firms to Fail (Boston: Harvard Business School Press, 1997). ↩︎
Worryingly, there has been an increasing trend from the major hyperscalers to increase capex by tapping the bond market instead of utilizing free cash flows.[183]↩︎
Eli Ofek and Matthew Richardson, "DotCom Mania: The Rise and Fall of Internet Stock Prices," The Journal of Finance 58, no. 3 (2003): 1113–1137. ↩︎
Atif Mian and Amir Sufi, "Household Leverage and the Recession of 2007 to 2009," IMF Economic Review 58, no. 1 (2010): 74–117. ↩︎
In developed economies, energy infrastructure has an output elasticity of 0.054 and digital infrastructure an output elasticity of 0.070 (interestingly, even higher than the numbers for developing economies).[184]↩︎
Yes, for a period in Silicon Valley, a popular staple was a meal replacement drink named Soylent. An instrumentalization of food, which deliberately took its name from a literal instrumentalization of humanity. ↩︎
Joseph Babbo, Intentio: Intentionality in an Age of Slop; or, A Manifesto for the Intentional Internet, Babbo.dev, 2026. ↩︎ ↩︎ ↩︎
Ross Andersen, "Does Sam Altman Know What He's Creating?," The Atlantic, July 24, 2023. ↩︎
To be clear, a group I count myself among. ↩︎
Dario Amodei, "Machines of Loving Grace," October 2024. ↩︎
Juvenal, Satires II, l. 24, trans. G.G. Ramsay, Loeb Classical Library (London: William Heinemann; New York: G.P. Putnam's Sons, 1918). ↩︎
Roy F. Baumeister et al., "Bad Is Stronger Than Good," Review of General Psychology 5, no. 4 (2001): 323–370. ↩︎
Amos Tversky and Daniel Kahneman, "Advances in Prospect Theory: Cumulative Representation of Uncertainty," Journal of Risk and Uncertainty 5, no. 4 (1992): 297–323. ↩︎
Steven D. Levitt, "Heads or Tails: The Impact of a Coin Toss on Major Life Decisions and Subsequent Happiness," The Review of Economic Studies 88, no. 1 (2021): 378–405. ↩︎
Reality seems to have a peculiar preference towards non-ergodicity. ↩︎
Edan Dekel and David Gantt Gurley, "How the Golem Came to Prague," The Jewish Quarterly Review 103, no. 2 (Spring 2013): 246. ↩︎
G. W. Leibniz, Theodicy: Essays on the Goodness of God, the Freedom of Man and the Origin of Evil, trans. E. M. Huggard (London: Routledge & Kegan Paul, 1951). ↩︎ ↩︎
Arch C. Johnston, "Seismic Moment Assessment of Earthquakes in Stable Continental Regions—III. New Madrid 1811–1812, Charleston 1886 and Lisbon 1755," Geophysical Journal International 126, no. 2 (1996): 314–344. ↩︎
Alvaro S. Pereira, "The Opportunity of a Disaster: The Economic Impact of the 1755 Lisbon Earthquake," The Journal of Economic History 69, no. 2 (June 2009): 466–499. ↩︎
U.S. ten Brink et al., "Assessment of Tsunami Hazard to the U.S. Atlantic Margin," Marine Geology 353 (2014): 45, citing J.T. Kozak, V.S. Moreira, and D.R. Oldroyd, Iconography of the 1755 Lisbon Earthquake (Prague: Academy of Sciences of the Czech Republic, 2005). ↩︎
Voltaire, Candide (1759), trans. unknown. ↩︎ ↩︎
Jean-Jacques Rousseau, "Letter to Voltaire on Providence," in The Discourses and Other Early Political Writings, ed. and trans. Victor Gourevitch, 2nd ed. (Cambridge: Cambridge University Press, 2019), 240–256. ↩︎
Immanuel Kant, "History and Natural Description of the Most Noteworthy Occurrences of the Earthquake that Struck a Large Part of the Earth at the End of the Year 1755," in Natural Science, ed. Eric Watkins, trans. Olaf Reinhardt (Cambridge: Cambridge University Press, 2012), 337–364. ↩︎
Jeremiah 2:19, The Holy Bible: A Translation from the Latin Vulgate in the Light of the Hebrew and Greek Originals, trans. Ronald Knox (London: Burns & Oates, 1955; repr. Baronius Press, 2012). ↩︎
Saint Irenaeus of Lyon, Against Heresies, Book IV, Chapter 37, §7, trans. Alexander Roberts and William Rambaut, in Ante-Nicene Fathers, vol. 1, ed. Alexander Roberts, James Donaldson, and A. Cleveland Coxe (Buffalo, NY: Christian Literature Publishing Co., 1885). ↩︎
Saint Augustine, Confessions, Book VII, Chapter 14, §20, trans. J.G. Pilkington, in Nicene and Post-Nicene Fathers, vol. 1, ed. Philip Schaff (Buffalo, NY: Christian Literature Publishing Co., 1887). ↩︎
Saint Thomas Aquinas, Summa Theologica I, Q. 58, Art. 3; I, Q. 64, Art. 2, trans. Fathers of the English Dominican Province, 2nd rev. ed. (1920). ↩︎ ↩︎
Capitalization modified from referent; other additions are from referent. ↩︎
Blessed John Duns Scotus, Ordinatio II, d.7, nn.22, 59–60, trans. Peter L.P. Simpson (2023). ↩︎ ↩︎
Oxford English Dictionary, "dunce (n.)," March 2026. ↩︎
Brian Cutter and Philip Swenson, "The Omission Theodicy," forthcoming in Oxford Studies in Philosophy of Religion. ↩︎
Peter M. Thompson, "Snap, Crackle, and Pop," Systems Technology, Inc., Hawthorne, CA. ↩︎
Norbert Wiener, "Some Moral and Technical Consequences of Automation," Science 131, no. 3410 (May 6, 1960): 1358. ↩︎ ↩︎
Elisions added to referent; other additions are from referent. ↩︎
Ibn Sīnā, Remarks and Admonitions: Physics and Metaphysics, trans. Shams Constantine Inati (New York: Columbia University Press, 2014), Fourth Class, chs. 8 and 12, 122–123. ↩︎ ↩︎
Additions are from referent. ↩︎
Ibn Sīnā, The Book of Salvation. Based on Al-Najāt, ed. M.D.S. al-Kurdī (Cairo: Maṭbaʿat al-Saʿāda, 1938; repr. Tehran: Murtaẓavī, 1913).
'Metaphysics,' II.18–19 (al-Kurdī, 246–249), trans. Jon McGinnis and David C. Reisman, in Classical Arabic Philosophy: An Anthology of Sources (Indianapolis: Hackett Publishing, 2007), 219.
'Metaphysics' (al-Kurdī, 82.2–12), trans. Peter Adamson, in "From the Necessary Existent to God," Interpreting Avicenna: Critical Essays, ed. Peter S. Adamson (Cambridge: Cambridge University Press, 2013), 186. ↩︎ ↩︎
Additions are from referent. ↩︎
Commonly referred to as Averroes in the West. ↩︎
Ibn Rushd, The Incoherence of the Incoherence, trans. Simon Van Den Bergh, "E. J. W. Gibb Memorial" New Series XIX (Cambridge: E.J.W. Gibb Memorial Trust, 1954), 318–319. ↩︎
Marshall G. S. Hodgson, The Venture of Islam: Conscience and History in a World Civilization, vol. 1, The Classical Age of Islam (Chicago: University of Chicago Press, 1974), 414, 439, 476, 480. ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
Dimitri Gutas, Greek Thought, Arabic Culture: The Graeco-Arabic Translation Movement in Baghdad and Early 'Abbasid Society (London: Routledge, 1998), 23–26, 95, 161, 192–93. ↩︎ ↩︎ ↩︎
George Makdisi, The Rise of Colleges: Institutions of Learning in Islam and the West (Edinburgh: Edinburgh University Press, 1981), 7, 75–78. ↩︎ ↩︎
Marshall G. S. Hodgson, The Venture of Islam: Conscience and History in a World Civilization, vol. 2, The Expansion of Islam in the Middle Periods (Chicago: University of Chicago Press, 1974), 35, 43, 47, 168, 171, 177, 180. ↩︎ ↩︎ ↩︎ ↩︎
A. C. S. Peacock, The Great Seljuk Empire (Edinburgh: Edinburgh University Press, 2015), 52. ↩︎
Johannes Kepler, Mysterium Cosmographicum: The Secret of the Universe, trans. A.M. Duncan, intro. and commentary E.J. Aiton (New York: Abaris Books, 1981), 53. ↩︎
Medial s (ſ) replaced with modern s to enhance readability. ↩︎
Robert Boyle, The Christian Virtuoso: Shewing, That by being addicted to Experimental Philosophy, a Man is rather Assisted, than Indisposed, to be a Good Christian (London: In the Savoy, by Edw. Jones, 1690), 14. ↩︎
G. E. M. Anscombe, "Modern Moral Philosophy," Philosophy 33, no. 124 (January 1958): 15. ↩︎
Max Weber, The Protestant Ethic and the Spirit of Capitalism, trans. Stephen Kalberg (New York: Roxbury, 2001), 35, 124. ↩︎
Sergio Tezanos-Vázquez, "Why Do Famines Still Occur in the 21st Century? A Review on the Causes of Extreme Food Insecurity," Journal of Economic Surveys 39 (2025): 1433-1461. ↩︎
David J. Danelski, A Supreme Court Justice Is Appointed (New York: Random House, 1964), 92, 189–191. ↩︎
Paul A. Lombardo, Three Generations, No Imbeciles: Eugenics, the Supreme Court, and Buck v. Bell (Baltimore: Johns Hopkins University Press, 2008), introduction, chs. 5, 8, 10–11. ↩︎
Zygmunt Bauman, Modernity and the Holocaust (Ithaca, NY: Cornell University Press, 2001), ch. 1. ↩︎
Brandt Defense Document No. 53 (extract from Erich Ristov, Erbgesundheitsrecht [Law concerning hereditary health], Stuttgart-Berlin: Kohlhammer, 1935). ↩︎
Adly Templeton et al., "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet," Transformer Circuits Thread, May 2024. ↩︎
Wes Gurnee et al., "When Models Manipulate Manifolds: The Geometry of a Counting Task," Transformer Circuits Thread, October 2025. ↩︎
Pranshu Verma and Shelly Tan, "A Bottle of Water per Email: The Hidden Environmental Costs of Using AI Chatbots," The Washington Post, September 18, 2024. ↩︎
Andy Masley, "I Think I Figured Out Exactly How the 'AI Used a Bottle of Water per Prompt' Miscalculation Happened," May 25, 2026. ↩︎
Andy Masley, "Empire of AI Is Wildly Misleading About AI Water Use," November 16, 2025. ↩︎
"How to Tell If It's Time for New Sneakers," John's Run/Walk Shop. ↩︎
Emily Cronkleton, "Average Walking Speed: Pace, and Comparisons by Age and Sex," Healthline, updated July 1, 2025. ↩︎
Hyperlinks in referent converted to references. ↩︎
Andy Masley, "Using ChatGPT Is Not Bad for the Environment - A Cheat Sheet," April 28, 2025. ↩︎
Pierre Friedlingstein et al., "Global Carbon Budget 2025," Global Carbon Project (2025), 'Fossil Emissions by Category'. ↩︎
Richard G. Newell, Brian C. Prest, and Ashley B. Vissing, "Trophy Hunting versus Manufacturing Energy: The Price Responsiveness of Shale Gas," Journal of the Association of Environmental and Resource Economists 6, no. 2 (2019): 391–431. ↩︎
Larissa MacFarquhar, "Requiem for a Dream," The New Yorker, March 11, 2013. ↩︎
Chief Justice Charles Evans Hughes, opinion of the Court in Fox Film Corp. v. Doyal, 286 U.S. 123, 127 (1932). ↩︎
Justice Potter Stewart, majority opinion in Twentieth Century Music Corp. v. Aiken, 422 U.S. 151, 156 (1975). ↩︎
Thomas Jefferson to Isaac McPherson, August 13, 1813, in The Founders' Constitution, ed. Philip B. Kurland and Ralph Lerner (Chicago: University of Chicago Press, 1987), vol. 3, art. 1, § 8, cl. 8, doc. 12. ↩︎
Justice John McLean, majority opinion in Wheaton v. Peters, 33 U.S. (8 Pet.) 591, 663 (1834). ↩︎
Justice Joseph Story, Folsom v. Marsh, 9 F. Cas. 342, 348 (C.C.D. Mass. 1841) (No. 4,901). ↩︎
Judge William Alsup, Order on Fair Use, Bartz v. Anthropic PBC, No. 3:24-cv-05417-WHA (N.D. Cal. June 23, 2025), ECF No. 231. ↩︎
Judge Vince Chhabria, Order Denying Plaintiffs' Motion for Partial Summary Judgment and Granting Meta's Cross-Motion for Partial Summary Judgment, Kadrey v. Meta Platforms, Inc., No. 3:23-cv-03417-VC (N.D. Cal. June 25, 2025), ECF No. 598. ↩︎
Audio Home Recording Act of 1992, Pub. L. No. 102-563, 106 Stat. 4237 (codified at 17 U.S.C. §§ 1001–1010). ↩︎
Recording Industry Ass'n of America v. Diamond Multimedia Systems, Inc., 180 F.3d 1072 (9th Cir. 1999). ↩︎
Horace He and Thinking Machines Lab, "Defeating Nondeterminism in LLM Inference," Thinking Machines Lab: Connectionism, September 2025. ↩︎
US traffic deaths have averaged ≈ 39,300 per year over ≈ 3.2 trillion annual vehicle-miles ⟹ ~1.23 deaths per 100 million miles[185]
Waymo, through its first 170 million driverless miles, recorded 92% fewer serious-or-fatal-injury crashes than human benchmarks on the same roads[186]
Adoption ramps linearly from 2018 to a share s of vehicle-miles today (average s/2 across 8 years)
Lives ≈ 3.2T × (s/2) × 1.23/100M × 0.92 × 8
s = 5% ⟹ ~900 per year ⟹ ~7,200 total
s = 80% ⟹ ~14,500 per year ⟹ ~116,000 total ↩︎
Hamilton Morrin et al., "Artificial Intelligence-Associated Delusions and Large Language Models: Risks, Mechanisms of Delusion Co-creation, and Safeguarding Strategies," The Lancet Psychiatry (March 5, 2026). ↩︎
Søren Dinesen Østergaard, "Emotion Contagion Through Interaction With Generative Artificial Intelligence Chatbots May Contribute to Development and Maintenance of Mania," Acta Neuropsychiatrica 37, e79 (2025): 1–3. ↩︎
"Allbirds, Inc. (BIRD) Stock Price History, April 14–June 11, 2026," Yahoo Finance. ↩︎
"Cisco Systems, Inc. (CSCO) Stock Price History, March 2000," Yahoo Finance. ↩︎
Cisco Systems, Inc., Form 10-K, fiscal year ended July 29, 2000, 'Consolidated Statements of Operations'. ↩︎ ↩︎ ↩︎
"NVIDIA Corporation (NVDA) Stock Price History, June 11, 2026," Yahoo Finance. ↩︎
NVIDIA Corporation, Form 10-K, fiscal year ended January 25, 2026, 'Consolidated Statements of Income', 51. ↩︎ ↩︎
This is, if anything, conservative. Cisco's fiscal year 2000 ended July 29, 2000, months after the March peak. Computing against the fiscal year 1999 diluted EPS of $0.29 gives an even greater multiple of ~276x.[177]↩︎
Cisco net income (fiscal year 2000) = $2,668M; revenue (fiscal years 1998, 2000) = $8,489M, $18,928M[177]
Nvidia net income (fiscal year 2026) = $120,067M; revenue (fiscal years 2024, 2026) = $60,922M, $215,938M[179]
Net margin: 2,668 / 18,928 ⟹ ~14.1%; 120,067 / 215,938 ⟹ ~55.6%
CAGR = (end / start)^(1/n) − 1, over n = 2 years
Cisco: (18,928 / 8,489)^(1/2) − 1 ⟹ ~49.3%
Nvidia: (215,938 / 60,922)^(1/2) − 1 ⟹ ~88.3% ↩︎
Robert J. Shiller, "U.S. Stock Markets 1871–Present and CAPE Ratio" (ie_data.xls), shillerdata.com, June 2026. ↩︎
Hugh Leask, "How the AI Debt Binge Shattered Hyperscalers' 'Unspoken Contract' with Investors," CNBC, February 23, 2026. ↩︎
Vivien Foster, Nisan Gorgulu, Dhruv Jain, Stéphane Straub, and Maria Vagliasindi, "The Impact of Infrastructure on Development Outcomes: A Meta-Analysis," World Bank Policy Research Working Paper 10350 (March 2023), Table A.12. ↩︎
Early Estimate of Motor Vehicle Traffic Fatalities and Fatality Rate in 2025, DOT HS 813 800 (Washington, DC: National Highway Traffic Safety Administration, April 2026). ↩︎
"Waymo Safety Impact," Waymo, data through December 2025. ↩︎