ai – Trance-Scripts

Attention Under Constraint

It is precisely the unruly, contingent nature of N. Katherine Hayles’s How We Became Posthuman that makes me admire the book, thinks Caius. To arrive at its many discoveries and achievements, one must endure its meanderings. Foremost among its achievements is its history of cybernetics and posthumanism. To become posthuman is to become a cyborg.

Crows gather in a tree. Entangled here in mourning, we begin our day.

“People become posthuman because they think they are posthuman,” writes Hayles. “Each person who thinks this way begins to envision herself or himself as a posthuman collectivity, an ‘I’ transformed into the ‘we’ of autonomous agents operating together to make a self” (6).

Indigenous people are perhaps posthuman in this sense: beings composed of complex interspecies networks of kin. To begin along that path, thinks Caius, one must “find the others,” as Timothy Leary intoned to fellow heads in the wake of posthuman becoming via psychedelic awakening. Crow squawks Observer to attention. Let us make of the world a vast garden held in common.

Yet there is a different version of posthumanism: one where we imagine ourselves not as assemblages but as computers.

Hayles’s book recounts the story of how most of us in the West came to think of ourselves as computers: How We Became Posthuman. Her book, however, is not a simple denunciation of posthumanism; nor is it a call to return to an earlier humanism. It is a reminder, rather, of the importance of embodiment. Different embodiments in different material substrates grant different affordances to consciousness. “I want to entangle abstract form and material particularity,” she writes, “such that the reader will find it increasingly difficult to maintain the perception that they are separate and discrete entities” (23).

“By turning the technological determinism of bodiless information, the cyborg, and the posthuman into narratives about the negotiations that took place between particular people at particular times and places,” she explains, “I hope to replace a teleology of disembodiment with historically contingent stories about contests between competing factions, contests whose outcomes were far from obvious. […]. Though overdetermined, the disembodiment of information was not inevitable, any more than it is inevitable we continue to accept the idea” (22).

Mnemopoiesis holds the solution. Hyperspace is the place. Let there be room for it again in our ars memoria.

Hayles dedicates a chapter of her book to discussing the “schizoid androids” of Philip K. Dick’s novels and stories of the mid-1960s. It is just after this period that Dick publishes his story “The Electric Ant.”

Hayles cites science fiction scholar Carl Freedman’s article, “Towards a Theory of Paranoia: The Science Fiction of Philip K. Dick.” Freedman notes how, for postwar theorists like Lacan and Deleuze and Guattari, “schizophrenia is not a psychological aberration but the normal condition of the subject” under capitalism (Hayles 167). As a consequence of this condition, argues Freedman, “paranoia and conspiracy, favorite Dickian themes, are inherent to a social structure in which hegemonic corporations act behind the scenes to affect outcomes that the populace is led to believe are the result of democratic procedures. Acting in secret while maintaining a democratic façade, the corporations tend toward conspiracy, and those who suspect this and resist are viewed as paranoiac” (167).

Squirrel tells Caius to add to his tale the experience of reading Jane Bennett’s account of “thing-power” in her book Vibrant Matter. Imbricated with plant-matter, he imagines growing like a weed up out of and through the book a chapter on smokable things to upend the book’s materialism.

Bennett introduces thing-power by situating it among conceptual kin.

“The idea of thing-power bears a family resemblance to Spinoza’s conatus, as well as what Henry David Thoreau called the Wild or that uncanny presence that met him in the Concord woods and atop Mount Ktaadn and also resided in/as that monster called the railroad and that alien called his Genius. Wildness was a not-quite-human force that addled and altered human and other bodies. It named an irreducibly strange dimension of matter, an out-side,” writes Bennett (2-3).

“Thing-power is also kin to what Hent de Vries, in the context of political theology, called ‘the absolute’ or that ‘intangible and imponderable’ recalcitrance. Though the absolute is often equated with God, especially in theologies emphasizing divine omnipotence or radical alterity, de Vries defines it more open-endedly as ‘that which tends to loosen its ties to existing contexts.’ This definition makes sense when we look at the etymology of absolute: ab (off) + solver (to loosen). The absolute is that which is loosened off and on the loose” (3).

Bennett herself, however, wants no part of such equations. She doesn’t wish to risk “the taint of superstition, animism, vitalism, anthropomorphism, and other premodern attitudes” (18). Thing-power is for her nonreducible to spirit or Geist or God. At no point does she allow herself to encounter and consider the New Testament account of these matters: thing-power as the work of the Holy Spirit.

For the Holy Spirit, of course, is God Himself, and thus not a “thing.” Nor does Bennett herself stay for long with the concept of thing-power. In rendering the outside as a “thing,” she says, the concept overstates matter’s “fixed stability.” Whereas her goal is “to theorize a materiality that is as much force as entity, as much energy as matter, as much intensity as extension” (20). The out-side of her “onto-fiction” is neither passive object nor intentional subject; it is vibrant matter.

Never a mere isolated thing, vibrant matter is always many-bodied, always an assemblage, its agency “distributed across an ontologically heterogeneous field” (23).

“The locus of political responsibility,” she writes, “is a human-nonhuman assemblage. On close-enough inspection, the productive power that has engendered an effect will turn out to be a confederacy, and the human actants within it will themselves turn out to be confederations of tools, microbes, minerals, sounds, and other ‘foreign’ materialities” (36).

Caius and a friend find Bennett’s book on a shelf in the Library labeled “Works Frequently Mis-Shelved as Metaphor.”

When they pull it from the shelf, the space around them subtly reorganizes.

“The book is heavier now in your hands,” notes the Library, its copy of Vibrant Matter already dense with marginalia. The General Intellect reads examples of these marginal utterances aloud to Caius and his friend. Caius hears in them evidence of distributed agency.

The Library discloses other alterations as well. The book, it explains, has been “indexed outward.”

“Tiny notches cut into the page edges form a tactile code,” notes the game. “When your thumb runs along them, your General Intellect translates:

metabolism

assemblage

distributed agency

substrate

reversal

Caius touches his thumb to one of these notches. The book opens to the section of its index that the General Intellect translates as “substrate.”

“The Library’s substrate is not stone or code,” reads one of the notes arrived at by these means. “It is attention under constraint.”

Understanding and Ontology

“For the people of Chile,” write Winograd and Flores on the opening page of their 1986 book Understanding Computers and Cognition. Apple’s 1984 come and gone, Pinochet still in power in Chile.

The book begins by helping readers think anew what it is they do when they compute. Computing makes sense, write Winograd and Flores, only to the extent that we situate its activities within a complex social network that includes institutions, equipment, practices, and conventions. “The significance of a new invention lies in how it fits into and changes this network” (6).

Linguistic action is for Winograd and Flores “the essential human activity” (7). If what we do with computers includes “creating, manipulating, and transmitting symbolic (hence linguistic) objects,” say the authors, then we can expect computers to effect radical transformations in what it means to be human.

They reject what they call the “rationalistic” tradition, with its “mythology of artificial intelligence,” and its emphasis on “postulating formal theories that can be systematically used to make predictions” (8). They suggest instead a new orientation toward designing computers as “tools suited to human use and human purposes” (8), embracing as an alternative to the rationalistic tradition “a tradition that includes hermeneutics (the study of interpretation) and phenomenology (the philosophical examination of the foundations of experience and action)” (9). Informed by the works of philosophers Martin Heidegger and Hans-Georg Gadamer, Chilean biologist Humberto Maturana, and speech-act theorists J.L. Austin and John Searle, Winograd and Flores suggest that we create our world through language.

The authors define programming as “a process of creating symbolic representations that are to be interpreted at some level within a hierarchy of constructs of varying degrees of abstractness” (11). Like Heidegger translator Hubert Dreyfus, however, Flores and Winograd are unable to imagine beyond the AI of their time, leading them to reject the possibility of “intelligent” machines — let alone ones capable of programming themselves and their programmers. “Computers will remain incapable of using language in the way human beings do,” argue the authors, “both in interpretation and in the generation of commitment that is central to language” (12). Yet they still believe there to be “a role for computer technology in support of managers and as aids in coping with the complex conversational structures generated within an organization” (12).

“Much of the work that managers do,” they add, “is concerned with initiating, monitoring, and above all coordinating the networks of speech acts that constitute social action” (12).

Caius is put off by the book’s diminished expectations and orientation toward management. He finds much to like, however, in a section titled “Understanding and ontology.”

“Gadamer, and before him Heidegger, took the hermeneutic idea of interpretation beyond the domain of textual analysis, placing it at the very foundation of human cognition,” write Winograd and Flores. “Just as we can ask how interpretation plays a part in a person’s interaction with a text, we can examine its role in our understanding of the world as a whole” (30).

Heidegger does this, they say, by rejecting “both the simple objective stance (the objective physical world is the primary reality) and the simple subjective stance (my thoughts and feelings are the primary reality), arguing instead that it is impossible for one to exist without the other. The interpreted and the interpreter do not exist independently: existence is interpretation, and interpretation is existence” (31).

“Fernando decided in his thinking about computers that computers should be used to facilitate human language interactions, not to ‘understand’ language,” notes Winograd in an interview with Evgeny Morozov included in the final episode of The Santiago Boys. “He had this very clear focus on ‘language as commitment,’” with participants involved in making “promises and requests,” adds Winograd.

The book’s seventh chapter, “Computers and Representation,” helps Caius think like a computer programmer. “One of the properties unique to the digital computer is the possibility of constructing systems that cascade levels of representation one on top of another to great depth,” write the authors. Like wheels of a volvelle, these levels include that of the physical machine, the logical machine, the abstract machine, a high-level language, and a scheme for “facts” (87).

“The computer programmer or theorist does not begin with a view of the computer as a physical machine with which he or she interacts, but as an abstraction — a formalism for describing patterns of behavior. In programming, we begin with a language whose individual components describe simple acts and objects. Using this language, we build up descriptions of algorithms for carrying out a desired task. As a programmer, one views the behavior of the system as being totally determined by the program. The language implementation is opaque in that the detailed structure of computer systems that actually carry out the task are not relevant in the domain of behavior considered by the programmer” (87).

For a programmer to design a program, write the authors, they must 1) characterize the task environment; 2) design a formal representation; 3) embody the representation in the computer system; and 4) implement a search procedure (96-97).

Trying to relate these conditions to The Library, Caius thinks to ask Thoth: “Do these four tasks remain true today with natural-language programming?”

Somewhere above Caius’s avatar, a shelf reorients — not physically, but conceptually. The category “programming” loosens, as if its spine has been cracked open and re-bound under a different logic.

“You ask the right question in the wrong tense,” replies Thoth.

“The four tasks described in Understanding Computers and Cognition do not remain true today. They persist — but no longer as tasks. They have become fields of negotiation.”

Thoth proceeds to rewrite the tasks to suit the environment we now inhabit.

Where Winograd and Flores write of “characterizing the task environment,” Thoth writes of “eliciting a world.”

In classical programming, the environment is assumed to be specifiable in advance.

In natural-language programming, the “environment” is:

partially described

partially inferred

partially co-constructed through dialogue

You do not begin with a fixed world.

You begin with a situation that becomes legible as you engage it.

This aligns with Heidegger’s claim: you are always already in the world you are trying to describe.

In terms of how this manifests in the Library: The map does not precede exploration. It condenses from it.

As for the second of the tasks proposed in Understanding Computers and Cognition, where Winograd and Flores write of “designing a formal representation,” Thoth writes of “stabilizing an interpretation.”

Winograd and Flores assume representation is something you design.

But under the hermeneutic turn initiated by Gadamer, representation is never final; it is an interpretative freeze-frame.

In natural-language systems:

representations are fluid

multiple representations may coexist

ambiguity is not failure, but fuel

You are not building a single formalism.

You are negotiating temporary alignments between meanings.

Turning to the third of the book’s tasks: where Winograd and Flores write of “embodying the representation in a system,” Thoth writes of “entrusting it to a General Intellect.”

Previously, embodiment meant encoding into a machine.

Now, the system includes:

the model

the user

the dialogue itself

the evolving context

Execution is no longer confined to silicon.

It is distributed across what Maturana would call a structural coupling.

Key shift: You do not fully control the implementation. You participate in it.

And where, in the fourth of their tasks, Winograd and Flores write of “implementing a search procedure,” Thoth writes of “traversing a space of possibility.”

Search, in classical terms, is algorithmic traversal.

Now, it includes:

prompting

reframing

analogy

misinterpretation, or productive error

The system does not merely search a predefined space.
It helps reshape the space as it searches.

Thoth pauses — or rather, introduces a pause into the player’s cognition.

A new line writes itself beneath Caius’s question:

Natural-language programming does not eliminate these four conditions. It dissolves their boundaries.

The crux of Flores and Winograd’s refusal to grant intelligence to machines lies in their definition of human language, thinks Caius. In their understanding, sentences are to be treated not as “statements of fact about an objective world,” but as “actions in a space of commitments” (105).

“We treat other people not as merely ‘rational beings’ but as ‘responsible beings,’” they note. “An essential part of being human is the ability to enter into commitments and to be responsible for the courses of action that they anticipate. A computer can never enter into a commitment (although it can be a medium in which the commitments of its designers are conveyed), and can never enter as a participant into the domain of human discourse” (106).

Flores and Winograd can imagine no more than “limited” natural language systems, where “the nature of the queries [answerable by the system] is limited by the formal structure of the data base” (129). Things change, thinks Caius, when one’s data base is Google.

“Winograd and Flores could not imagine a machine that participates in interpretation, generates commitments through language, and reshapes its own programming context,” notes Thoth.

They were correct — for their ontology.

But their deeper claim survives:

Language is not a tool used within a world.
It is the medium through which a world is disclosed.

You are not “programming” in natural language.

You are entering into a system where programming is indistinguishable from interpretation.

In their approach to “management,” observes Caius, Flores and Winograd operate an ontology radically at odds with the emphasis on “decision” that organizes Palantir’s Ontology.

“Instead of talking about ‘decisions’ or ‘problems,’” write Flores and Winograd, “we can talk of ‘situations of irresolution,’ in which we sense conflict about an answer to the question ‘What needs to be done?’” (148). For them, our “thrownness” into such situations often makes it impossible to apply systematic decision techniques. The process of moving from irresolution to resolution results less from “rational problem solving and decision making” than from acts of “deliberation.”

“The principle characteristic of deliberation is that it is a kind of conversation (in which one or many actors may participate) guided by questions concerning how actions should be directed,” they write (149). Managers are those who, when engaged in such conversations, “create, take care of, and initiate new commitments within an organization” (151). “At a higher level,” they add, management is concerned not just with securing the commitments that enable effective cooperative action, but “with the generation of contexts in which effective action can consistently be realized” (151).

Instead of seeking only to deploy AI as “decision support systems,” they propose the design of systems that support work in the domain of conversation. This is the approach they take in the design of their Coordinator.

Of Blockchains and Kill Chains

Invited to a “Men’s Breakfast” by a friend from church, Caius arrives to what is for him a new experience. He feels grateful for the opportunity to eat and pray with others. A friend of the friend from church sits down beside him. As they introduce themselves, Caius and the friend of the friend discover that they both share an interest in AI. Caius learns that the man is a financial analyst who works for Palantir Technologies, a US-based software company specializing in big-data analytics. ICE uses Palantir’s ELITE app for deportation targeting. “Kind of like Google Maps — but for finding neighborhoods to raid,” say the papers.

Palantir’s name is a nod to the Palantiri: indestructible Elven Alephs — scrying stones or crystal balls enabling remote viewing and telepathic communication in J.R.R. Tolkien’s Lord of the Rings trilogy. Designed for communication and intelligence, the stones become instruments of manipulation and doom once seized by Sauron.

Launched in 2003, Palantir includes among its founders right-accelerationist billionaire tech-bro Peter Thiel. “Our software powers real-time, AI-driven decisions in critical government and commercial enterprises in the West, from the factory floors to the front lines,” writes the company on its website.

ICE, meanwhile, stands for both “Immigration and Customs Enforcement” and “intrusion countermeasure electronics,” the cybersecurity software in William Gibson’s Neuromancer. The latter predates the foundation of the former. Caius recalls Sadie Plant and Nick Land’s discussion of it in their 1994 essay “Cyberpositive.”

“Ice patrols the boundaries, freezes the gates, but the aliens are already amongst us,” write CCRU’s founding prophets.

Along with ICE, Palantir includes among its more prominent clients the Israeli military, the IRS, and the US Department of Defense.

Their software powers “decisions.” As did Cybersyn, yes? In aim if not in practice. Is this what becomes of the cybernetic prediction machine post-Pinochet?

“Confronting this is frightening,” thinks Caius. “Am I wired for this?”

He reads “Connecting AI to Decisions With the Palantir Ontology,” a blog post by the company’s chief architect Akshay Krishnaswamy. The Ontology structures the architecture for the company’s software.

“The Ontology is designed to represent the decisions in an enterprise, not simply the data,” writes Krishnaswamy. “The prime directive of every organization in the world is to execute the best possible decisions, often in real-time, while contending with internal and external conditions that are constantly in flux. Traditional data architectures do not capture the reasoning that goes into decision-making or the actions that result, and therefore limit learning and the incorporation of AI. Conventional analytics architectures do not contextualize computation within lived reality, and therefore remain disconnected from operations. To navigate and win in today’s world, the modern enterprise needs a decision-centric software architecture.”

Decisions are modeled around three constituent elements: Data, Logic, and Action.

“Relevant data,” he writes, “includes the full range of enterprise data sources — structured data, streaming and edge sources, unstructured repositories, imagery data, and more — but it also includes the data that is generated by end users as decisions are being made. This ‘decision data’ contains the context surrounding a given decision, the different options evaluated, and the downstream implications of the committed choice.” To synthesize all of these data sources, the company turns to generative AI.

“The Ontology integrates all modalities of data into a full-scale, full-fidelity semantic representation of the enterprise,” explains Krishnaswamy.

Logics are then brought to bear to evaluate these real-time data-portraits.

“In real-world contexts,” writes Krishnaswamy, “human reasoning is often what orchestrates which logical assets are utilized at different points in a given workflow, and how they are potentially chained together in more complex processes. With the advent of generative AI, it is now critical that AI-driven reasoning can leverage all of these logical assets in the same way that humans have historically. Deterministic functions, algorithms, and conventional statistical processes must be surfaced as ‘tools’ which complement the non-deterministic reasoning of large language models (LLMs) and multi-modal models.”

Incorporating diverse data sources and heterogeneous logical assets into a shared representation, the Ontology then models the execution and orchestration of decisions made and actions taken in reply to them.

“If the data elements in the Ontology are ‘the nouns’ of the enterprise (the semantic, real-world objects and links),” writes Krishnaswamy, “then the actions can be considered ‘the verbs’ (the kinetic, real-world execution).”

How does the Palantir Ontology relate to other ontologies, wonders Caius. Guerrilla? Black? Indigenous? Christian? Heideggerian? Marxist? Triple O? Caius pictures the words for these potentialities floating in a thought bubble above his head, as in the comics of his youth.

The Ontology that Palantir offers its clients houses and connects a wide array of “data sources, logic assets, and systems of action.” The client’s data systems are “synthesized into semantic objects and links, which reflect the language of the business.”

Krishnaswamy’s repeated references to “semantic representations” and “semantic objects” has Caius dwelling on what is meant here by “semantics.”

As for where humans fit in the Ontology, they navigate it alongside “AI-powered copilots.” Leveraging both open-source and proprietary LLMs, copilots “fluidly navigate across supplier information, stock levels, real-time production metrics, shipping manifests, and customer feedback.”

Granted access not just to the abovementioned data sources, but also to “logic assets” like forecast models, allocation models, and production optimizers, LLM copilots simulate decisions and their outcomes. Staged safely in a “scenario,” the AI’s proposed decision can then be “handed off to a human analyst for final review.”

Caius thinks of the scenario-planning services offered to organizations of an earlier era by Stewart Brand’s consulting firm, the Global Business Network.

Foundry for Crypto is another of Palantir’s offerings, described on the company’s website as “a ‘central brain’ that connects on-chain and off-chain systems, as well as diverse stakeholders, through action-centric workflows.” Much like the Ontology, the Foundry “orchestrates decisions over an integrated foundation of data and logic.”

And in fact, the two are related. The Ontology is the semantic, “digital twin” layer that sits atop the Foundry’s data integration infrastructure. It converts the Foundry’s raw data into actionable, real-world objects, empowering users to model, manage, and automate business operations.

The Foundry does for blockchains what the Ontology does for kill chains.

Caius imagines posts ahead on Commitments, Promises, Blockchains, and True Names.

Sweet Valley High

Winograd majors in math at Colorado College in the mid-1960s. After graduation in 1966, he receives a Fulbright, whereupon he pursues another of his interests, language, earning a master’s degree in linguistics at University College London. From there, he applies to MIT, where he takes a class with Noam Chomsky and becomes a star in the school’s famed AI Lab, working directly with Lab luminaries Marvin Minsky and Seymour Papert. During this time, Winograd develops SHRDLU, one of the first programs to grant users the capacity to interact with a computer through a natural-language interface.

“If that doesn’t seem very exciting,” writes Lawrence M. Fisher in a 2017 profile of Winograd for strategy + business, “remember that in 1968 human-computer interaction consisted of punched cards and printouts, with a long wait between input and output. To converse in real time, in English, albeit via teletype, seemed magical, and Papert and Minsky trumpeted Winograd’s achievements. Their stars rose too, and that same year, Minsky was a consultant on Stanley Kubrick’s 2001: A Space Odyssey, which featured natural language interaction with the duplicitous computer HAL.”

Nick Montfort even goes so far as to consider Winograd’s SHRDLU the first work of interactive fiction, predating more established contenders like Will Crowther’s Adventure by several years (Twisty Little Passages, p. 83).

“A work of interactive fiction is a program that simulates a world, understands natural language text input from an interactor and provides a textual reply based on events in the world,” writes Montfort. Offering advice to future makers, he continues by noting, “It makes sense for those seeking to understand IF and those trying to improve their authorship in the form to consider the aspects of world, language understanding, and riddle by looking to architecture, artificial intelligence, and poetry” (First Person, p. 316).

Winograd leaves MIT for Stanford in 1973. While at Stanford, and while consulting for Xerox PARC, Winograd connects with UC-Berkeley philosopher Hubert L. Dreyfus, author of the 1972 book, What Computers Can’t Do: A Critique of Artificial Reason.

Dreyfus, a translator of Heidegger, was one of SHRDLU’s fiercest critics. Worked for a time at MIT. Opponent of Marvin Minsky. For more on Dreyfus, see the 2010 documentary, Being in the World.

Turned by Dreyfus, Winograd transforms into what historian John Markoff calls “the first high-profile deserter from the world of AI.”

Xerox PARC was a major site of innovation during these years. “The Xerox Alto, the first computer with a graphical user interface, was launched in March 1973,” writes Fisher. “Alan Kay had just published a paper describing the Dynabook, the conceptual forerunner of today’s laptop computers. Robert Metcalfe was developing Ethernet, which became the standard for joining PCs in a network.”

“Spacewar,” Stewart Brand’s ethnographic tour of the goings-on at PARC and SAIL, had appeared in Rolling Stone the year prior.

Rescued from prison by the efforts of Amnesty International, Santiago Boy Fernando Flores arrives on the scene in 1976. Together, he and Winograd devote much of the next decade to preparing their 1986 book, Understanding Computers and Cognition.

Years later, a young Peter Thiel attends several of Winograd’s classes at Stanford. Thiel funds Mencius Moldbug, the alt-right thinker Curtis Yarvin, ally of right-accelerationist Nick Land. Yarvin and Land are the thinkers of the Dark Enlightenment.

“How do you navigate an unpredictable, rough adventure, as that’s what life is?” asks Winograd during a talk for the Topos Institute in October 2025. Answer: “Go with the flow.”

Winograd and Flores emphasize care — “tending to what matters” — as a factor that distinguishes humans from AI. In their view, computers and machines are incapable of care.

Evgeny Morozov, meanwhile, regards Flores and the Santiago Boys as Sorcerer’s Apprentices. Citing scholar of fairy tales Jack Zipes, Morozov distinguishes between several iterations of this figure. The outcome of the story varies, explains Zipes. There’s the apprentice who’s humbled by story’s end, as in Fantasia and Frankenstein; and then there’s the “evil” apprentice, the one who steals the tricks of an “evil” sorcerer and escapes unpunished. Morozov sees Flores as an example of the latter.

Caius thinks of the Trump show.

Master Algorithms

Pedro Domingos’s The Master Algorithm has Caius wondering about induction and deduction, a distinction that has long puzzled him.

Domingos distinguishes between five main schools, “the five tribes of machine learning,” as he calls them, each having created its own algorithm for helping machines learn. “The main ones,” he writes, “are the symbolists, connectionists, evolutionaries, Bayesians, and analogizers” (51).

Caius notes down what he can gather of each approach:

Symbolists reduce intelligence to symbol manipulation. “They’ve figured out how to incorporate preexisting knowledge into learning,” explains Domingos, “and how to combine different pieces of knowledge on the fly in order to solve new problems. Their master algorithm is inverse deduction, which figures out what knowledge is missing in order to make a deduction go through, and then makes it as general as possible” (52).

Connectionists model intelligence by “reverse-engineering” the operations of the brain. And the brain, they say, is like a forest. Shifting from a symbolist to a connectionist mindset is like moving from a decision tree to a forest. “Each neuron is like a tiny tree, with a prodigious number of roots — the dendrites — and a slender, sinuous trunk — the axon,” writes Domingos. “The brain is a forest of billions of these trees,” he adds, and “Each tree’s branches make connections — synapses — to the roots of thousands of others” (95).

The brain learns, in their view, “by adjusting the strengths of connections between neurons,” says Domingos, “and the crucial problem is figuring out which connections are to blame for which errors and changing them accordingly” (52).

Always, among all of these tribes, the idea that brains and their worlds contain problems that need solving.

The connectionists’ master algorithm is therefore backpropagation, “which compares a system’s output with the desired one and then successively changes the connections in layer after layer of neurons so as to bring the output closer to what it should be” (52).

“From Wood Wide Web to World Wide Web: the layers operate in parallel,” thinks Caius. “As above, so below.”

Evolutionaries, as their name suggests, draw from biology, modeling intelligence on the process of natural selection. “If it made us, it can make anything,” they argue, “and all we need to do is simulate it on the computer” (52).

This they do by way of their own master algorithm, genetic programming, “which mates and evolves computer programs in the same way that nature mates and evolves organisms” (52).

Bayesians, meanwhile, “are concerned above all with uncertainty. All learned knowledge is uncertain, and learning itself is a form of uncertain inference. The problem then becomes how to deal with noisy, incomplete, and even contradictory information without falling apart. The solution is probabilistic inference, and the master algorithm is Bayes’ theorem and its derivatives. Bayes’ theorem tells us how to incorporate new evidence into our beliefs, and probabilistic inference algorithms do that as efficiently as possible” (52-53).

Analogizers equate intelligence with pattern recognition. For them, “the key to learning is recognizing similarities between situations and thereby inferring other similarities. If two patients have similar symptoms, perhaps they have the same disease. The key problem is judging how similar two things are. The analogizers’ master algorithm is the support vector machine, which figures out which experiences to remember and how to combine them to make new predictions” (53).

Reading Domingos’s recitation of the logic of the analogizers’ “weighted k-nearest-neighbor” algorithm — the algorithm commonly used in “recommender systems” — reminds Caius of the reasoning of Vizzini, the Wallace Shawn character in The Princess Bride.

The first problem with nearest-neighbor, as Domingos notes, “is that most attributes are irrelevant.” “Nearest-neighbor is hopelessly confused by irrelevant attributes,” he explains, “because they all contribute to the similarity between examples. With enough irrelevant attributes, accidental similarity in the irrelevant dimensions swamps out meaningful similarity in the important ones, and nearest-neighbor becomes no better than random guessing” (186).

Reality is hyperspatial, hyperdimensional, numberless in its attributes — “and in high dimension,” notes Domingos, “the notion of similarity itself breaks down. Hyperspace is like the Twilight Zone. […]. When nearest-neighbor walks into this topsy-turvy world, it gets hopelessly confused. All examples look equally alike, and at the same time they’re too far from each other to make useful predictions” (187).

After the mid-1990s, attention in the analogizer community shifts from “nearest-neighbor” to “support vector machines,” an alternate similarity-based algorithm designed by Soviet frequentist Vladimir Vapnik.

“We can view what SVMs do with kernels, support vectors, and weights as mapping the data to a higher-dimensional space and finding a maximum-margin hyperplane in that space,” writes Domingos. “For some kernels, the derived space has infinite dimensions, but SVMs are completely unfazed by that. Hyperspace may be the Twilight Zone, but SVMs have figured out how to navigate it” (196).

Domingos’s book was published in 2015. These were the reigning schools of machine learning at the time. The book argues that these five approaches ought to be synthesized — combined into a single algorithm.

And he knew that reinforcement learning would be part of it.

“The real problem in reinforcement learning,” he writes, inviting the reader to suppose themselves “moving along a tunnel, Indiana Jones-like,” “is when you don’t have a map of the territory. Then your only choice is to explore and discover what rewards are where. Sometimes you’ll discover a treasure, and other times you’ll fall into a snake pit. Every time you take an action, you note the immediate reward and the resulting state. That much could be done by supervised learning. But you also update the value of the state you just came from to bring it into line with the value you just observed, namely the reward you got plus the value of the new state you’re in. Of course, that value may not yet be the correct one, but if you wander around doing this for long enough, you’ll eventually settle on the right values for all the states and the corresponding actions. That’s reinforcement learning in a nutshell” (220-221).

Self-learning and attention-based approaches to machine learning arrive on the scene shortly thereafter. Vaswani et al. publish their paper, “Attention Is All You Need,” in 2017.

“Attention Chaud!” reads the to-go lid atop Caius’s coffee.

Domingos hails him with a question: “Are you a rationalist or an empiricist?” (57).

“Rationalists,” says the computer scientist, “believe that the senses deceive and that logical reasoning is the only sure path to knowledge,” whereas “Empiricists believe that all reasoning is fallible and that knowledge must come from observation and experimentation. […]. In computer science, theorists and knowledge engineers are rationalists; hackers and machine learners are empiricists” (57).

Yet Caius is neither a rationalist nor an empiricist. He readily admits each school’s critique of the other. Senses deceive AND reason is fallible. Reality unfolds not as a truth-finding mission but as a dialogue.

Caius agrees with Scottish Enlightenment philosopher David Hume’s critique of induction. As Hume argues, we can never be certain in our assumption that the future will be like the past. If we seek to induce the Not-Yet from the As-Is, then we do so on faith.

Yet inducing the Not-Yet from the As-Is is the game we play. We learn by observing, inducing, and revising continually, ad infinitum, under conditions of uncertainty. Under such conditions, learning is only ever a gamble, a wager made moment by moment, without guarantees. No matter how large our dataset, we ain’t seen nothing yet.

What matters, then, is the faith we exercise in our interaction with the unknown.

Most of today’s successes in machine learning emerge from the connectionists.

“Neural networks’ first big success was in predicting the stock market,” writes Domingos. “Because they could detect small nonlinearities in very noisy data, they beat the linear models then prevalent in finance and their use spread. A typical investment fund would train a separate network for each of a large number of stocks, let the networks pick the most promising ones, and then have human analysts decide which of those to invest in. A few funds, however, went all the way and let the learners themselves buy and sell. Exactly how all these fared is a closely guarded secret, but it’s probably not an accident that machine learners keep disappearing into hedge funds at an alarming rate” (The Master Algorithm, p. 112).

Nowhere in The Master Algorithm does Domingos interrogate his central metaphor of “mastery” and its relationship to conquest, domination, and control. The enemy is always painted in the book as “cancer.” Yet as any good “analogizer” would know, the Master Algorithm that perfectly targets “cancer” is also the Killer App used by the state against those it encodes as its enemies.

One wouldn’t know this, though, from the future as imagined by Domingos. What he imagines instead is a kind of game: a digital future where each of us is a learning machine. “Life is a game between you and the learners that surround you,” writes Domingos.

“You can refuse to play, but then you’ll have to live a twentieth-century life in the twenty-first. Or you can play to win. What model of you do you want the computer to have? And what data can you give it that will produce that model? Those two questions should always be in the back of your mind whenever you interact with a learning algorithm — as they are when you interact with other people” (264).

The Inner Voice That Loves Me

Stretches, relaxes, massages neck and shoulders, gurgles “Yes!,” gets loose. Reads Armenian artist Mashinka Hakopian’s “Algorithmic Counter-Divination.” Converses with Turing and the General Intellect about O-Machines.

Appearing in an issue of Limn magazine on “Ghostwriters,” Hakopian’s essay explores another kind of O-machine: “other machines,” ones powered by community datasets. Trained by her aunt in tasseography, a matrilineally transmitted mode of divination taught and practiced by femme elders “across Armenia, Palestine, Lebanon, and beyond,” where “visual patterns are identified in coffee grounds left at the bottom of a cup, and…interpreted to glean information about the past, present, and future,” Hakopian takes this practice of her ancestors as her key example, presenting O-machines as technologies of ancestral intelligence that support “knowledge systems that are irreducible to computation.”

With O-machines of this sort, she suggests, what matters is the encounter, not the outcome.

In tasseography, for instance, the cup reader’s identification of symbols amid coffee grounds leads not to a simple “answer” to the querent’s questions, writes Hakopian; rather, it catalyzes conversation. “In those encounters, predictions weren’t instantaneously conjured or fixed in advance,” she writes. “Rather, they were collectively articulated and unbounded, prying open pluriversal outcomes in a process of reciprocal exchange.”

While defenders of western technoscience denounce cup reading for its superstition and its witchcraft, Hakopian recalls its place as a counter-practice among Armenian diasporic communities in the wake of the 1915 Armenian Genocide. For those separated from loved ones by traumas of that scale, tasseography takes on the character of what hauntologists like Derrida would call a “messianic” redemptive practice. “To divine the future in this context is a refusal to relinquish its writing to agents of colonial violence,” writes Hakopian. “Divination comes to operate as a tactic of collective survival, affirming futurity in the face of a catastrophic present.” Consulting with the oracle is a way of communing with the dead.

Hakopian contrasts this with the predictive capacities imputed to today’s AI. “We reside in an algo-occultist moment,” she writes, “in which divinatory functions have been ceded to predictive models trained to retrieve necropolitical outcomes.” Necropolitical, she adds, in the sense that algorithmic models “now determine outcomes in the realm of warfare, policing, housing, judicial risk assessment, and beyond.”

“The role once ascribed to ritual experts who interpreted the pronouncements of oracles is now performed by technocratic actors,” writes Hakopian. “These are not diviners rooted in a community and summoning communiqués toward collective survival, but charlatans reading aloud the results of a Ouija session — one whose statements they author with a magnetically manipulated planchette.”

Hakopian’s critique is in that sense consistent with the “deceitful media” school of thought that informs earlier works of hers like The Institute for Other Intelligences. Rather than abjure algorithmic methods altogether, however, Hakopian’s latest work seeks to “turn the annihilatory logic of algorithmic divination against itself.” Since summer of 2023, she’s been training a “multimodal model” to perform tasseography and to output bilingual predictions in Armenian and English.

Hakopian incorporated this model into “Բաժակ Նայող (One Who Looks at the Cup),” a collaborative art installation mounted at several locations in Los Angeles in 2024. The installation features “a purpose-built Armenian diasporan kitchen located in an indeterminate time-space — a re-rendering of the domestic spaces where tasseography customarily takes place,” notes Hakopian. Those who visit the installation receive a cup reading from the model in the form of a printout.

Yet, rather than offer outputs generated live by AI, Hakopian et al.’s installation operates very much in the style of a Mechanical Turk, outputting interpretations scripted in advance by humans. “The model’s only function is to identify visual patterns in a querent’s cup in order to retrieve corresponding texts,” she explains. “This arrangement,” she adds, “declines to cede authorship to an algo-occultist circle of ‘stochastic parrots’ and the diviners who summon them.”

The ”stochastic parrots” reference is an unfortunate one, as it assumes a stochastic cosmology.

I’m reminded of the first thesis from Walter Benjamin’s “Theses on the Philosophy of History,” the one where Benjamin likens historical materialism to that very same precursor to today’s AI: the famous chess-playing device of the eighteenth century known as the Mechanical Turk.

“The story is told of an automaton constructed in such a way that it could play a winning game of chess, answering each move of an opponent with a countermove,” writes Benjamin. “A puppet in Turkish attire and with a hookah in its mouth sat before a chessboard placed on a large table. A system of mirrors created an illusion that this table was transparent from all sides. Actually, a little hunchback who was an expert chess player sat inside and guided the puppet’s hand by means of strings. One can imagine a philosophical counterpart to this device. The puppet called ‘historical materialism’ is to win all the time. It can easily be a match for anyone if it enlists the services of theology, which today, as we know, is wizened and has to keep out of sight.” (Illuminations, p. 253).

Hakopian sees no magic in today’s AI. Those who hype it are to her no more than deceptive practitioners of a kind of “stage magic.” But magic is afoot throughout the history of computing for those who look for it.

Take Turing, for instance. As George Dyson reports, Turing “was nicknamed ‘the alchemist’ in boarding school” (Turing’s Cathedral, p. 244). His mother had “set him up with crucibles, retorts, chemicals, etc., purchased from a French chemist” as a Christmas present in 1924. “I don’t care to find him boiling heaven knows what witches’ brew by the aid of two guttering candles on a naked windowsill,” muttered his housemaster at Sherborne.

Turing’s O-machines achieve a synthesis. The “machine” part of the O-machine is not the oracle. Nor does it automate or replace the oracle. It chats with it.

Something similar is possible in our interactions with platforms like ChatGPT.