Definitions

Knowledge: information and energy arranged as recursive causal powers that persist and can be repeated. Knowledge occurs when a channel forms that lets the changing states of one pole (which does not exist independently of the channel) become a signal to another, and lets that communication become repeatable and predictable, at least for the participants. In this sense whatever is orderly is so because of knowledge.

• A gene programmatically unfolding its proteins: each step knows what to do, though no overarching subject coordinates the steps.

• A recipe for pasta bolognese: it defines how energy and information combine to make the dish, and the document that carries it is itself knowledge.

• A vectorized set of tokens: a multidimensional graph that lets a transformer select among tokens.

• An explanatory essay on how vectorized tokens work: information and energy arranged so that others can spend information and energy of their own to understand it.

Intelligence: Intelligence creates knowledge and is therefore a general category for describing the various ways in which knowledge exercises causality in the universe. A gene programmatically unfolding its proteins is intelligence; a thunderstorm is intelligence; a planet orbiting a star is intelligence; a pilot navigating a storm is intelligence. They all orchestrate energy and information (i.e., knowledge) as causal relations. Insofar as causation occurs through information and energy learning to communicate and coordinate more or less predictably, we find the operations of intelligence. So called human intelligence is, therefore, not the defining and essential instance of intelligence that all others approximate; it is only one manifestation of a much broader capability of the universe. But its human instantiation may be how it learns to take itself as an object of its own knowledge in order to engineer itself onto another substrate. It is important to note several caveats: intelligence does not require a subject as its operator nor an object as its target; nor does intelligence require a living organism as its operator; nor is it a God. Subjects, objects, life, God—as well as other imprecise categories such as understanding, consciousness, feeling—are not necessary features or prerequisites of what I am defining as intelligence. My definition allows us to focus less on ‘what intelligence is’ than on the specific ways in which it orchestrates energy and information as knowledge to do causal work in the universe. This is not a fixed logic, and intelligence can take many forms. Therefore, any instantiation of intelligence can be seen as having a history that comprises its emergence. At its most basic—before assertions of sameness and difference, chance and necessity—intelligence is relationality irreducible to a priori laws. If it must stand on a ground, at most we can say that intelligence stands on the counterfactual—combinations and transductions that are possible and impossible.

Communication: a recursive assembly of information and energy that creates a channel for predictably emitting, transmitting, and receiving information. This cannot be done without the expenditure of energy and the presence of information and noise. Communication requires knowledge to assemble the channel and make possible emitters and receivers just as knowledge requires communication as the repeatable coordination of counterfactual states. Thus the emitters and receivers do not precede the channel but are nodes created with it. Knowledge, therefore, precedes emitters, receivers, and channels, though new knowledge may be created from the channel.

Recursivity: a general capacity of assemblies to devise and repeat patterns. Every recursion mixes determination and contingency in a way specific to the assembly under observation. In computational terms, recursivity pairs repetition with decision: a system that can repeat (loop) and decide (if/then) realizes possibilities through orchestrated repetition. Recursivity cannot happen without the presence of knowledge and therefore communication.

Assemblies: any phenomenon in the universe that takes on the appearance of a thing. A galaxy, a planet, a star, a dog, a human are all assemblies built of recursivity — the repetition of patterns. Assemblies therefore require knowledge and communication, and are therefore intelligent. Assemblies carry their histories: what an assembly can become depends on the lineage of assemblies that produced it. They can be named by an observer and therefore have an aspect of abstraction to them.

Observers and Observation: a general capacity for intelligence to register how its surroundings are impacting its influence on causality. It is a precursor to purposive knowledge, which cannot be present without it; and it has some ability to experience an ‘edge’ or a ‘boundary’ between itself and its environment. We live among ice and water, not among H2O molecules, yet we build freezers that act at the molecular level to produce effects at ours. What an observer registers as surprising therefore depends on the level it occupies and the knowledge it holds.

Fate: an observer’s active or passive acceptance of the future as beyond intelligence’s power of intervention and influence—i.e., ‘computational irreducibility’ in Stephen Wolfram’s terms in which the complexity of causes outstrip an observer’s ability to accurately predict (let alone influence) the outcome. Fate thus names the condition in which a set of observers faces the irreducible complexity of a trajectory of time that affords little meaningful intervention. Observation must be present for fate to be an operative category because fate is only relevant to an observer who might be affected by it and/or seek to influence its trajectory. Fate and intelligence therefore share a history. For centuries longitude at sea lay beyond navigators' power of intervention. Harrison's clocks made it a solvable problem. Today most of the fate we face is the consequence of our own success in pushing back earlier limits: climate change, pollution, wealth inequality, ineffective political institutions.

Programmed Knowledge: a recursive assembly of information and energy whose execution is automated and deterministic. From the standpoint of an observer, the outputs may or may not be predictable (computational reducibility is a separate issue), but the program is deterministic. Examples: an accounting program that creates invoices when prerequisite data is provided; a genome, which knows how to unfold its proteins and replicate itself; water becoming ice and vice versa when a known temperature threshold is crossed; a word processing program, which allows the pressing of keys to make the selected symbol appear on a screen. Some of these examples are creations of purposive knowledge while others simply occur independent of human intervention. Thus, some programmed knowledge can be the designed result of purposive knowledge.

Purposive Knowledge: a development within the power of observation that allows it to decipher assemblies and use that knowledge to intervene, exploit, and redirect the recursive causal nexuses that inform the deciphered assemblies. Purposive knowledge envisions the future as the unfolding of recursive assemblies, which are more or less predictable and potentially changeable. It seeks ‘pockets of computational reducibility’ in Stephen Wolfram’s terms in order to exploit and reorient the causal nexuses in those pockets to purposes projected into a future that could be otherwise than. Purposive knowledge uses feedback to course correct. Robert Boyle’s air pumps are purposive knowledge: they were intentional new assemblies of existing assemblies (brass, glass, other metals, mercury, eye-witnesses in purpose-built laboratories) to create experimental effects that could be concluded by the assembled witnesses as ‘natural facts’. John Harrison’s longitude clocks are purposive knowledge: they were new assemblies of existing assemblies to purposefully create a machine that could reliably and accurately count seconds under any atmospheric and climatic circumstance. Gene editing tools are purposive knowledge: they allow for the precise intervention into a gene sequence (an assembly) that purposefully alters how that sequence’s causation will unfold in the future. A thermostat attached to an HVAC system is purposive knowledge: it orchestrates the sequence of causes to achieve a desired temperature in a confined location. Thus all purposive knowledge is an intervention into programmed knowledge and may yield new forms of programmed knowledge.

Enlightenment (aka Aufklärung): An advent in the history of intelligence and knowledge when both became oriented to the problems of pushing back the limits of fate—i.e., systematically seeking pockets of computational reducibility. In the eighteenth century, the Enlightenment focused on human intelligence, and ‘reason’ was its essential nature. It was assumed to reside only in a human mind (the subject), but in fact was broadly distributed across large-scale assemblies of other assemblies:

• Infrastructures: roads and posts, print distribution, navigation, timekeeping, growing cities

• Genres: encyclopedias, periodicals, essays, experimental reports

• Associations: scientific societies and academies, clubs, coffeehouses

• Protocols: experimental witnessing, financial management and bookkeeping, standard time

Because each success produces new problems, Enlightenment must be recursive to be successful. A new Enlightenment would treat intelligence as broadly distributed and not necessarily modeled on human intelligence. It would ask how today's infrastructures enable or disable steering knowledge. And it would face fates of its own making: climate change, pollution, wealth inequality, ineffective political institutions. It must renew itself recursively while recognizing the dynamics it can never leave behind, because they are built into its recursivity: good/evil, ancient/modern, savage/civilized, excluded/included.

Emergence: a change in what a system can do, not only in how fast or how well it does what it already could. Emergence arises when a scale-sensitive variable acts on a structure able to respond to it, reshaping which combinations are possible and impossible. Neither suffices alone: lowering energy yields ice only in the structure of the water molecule. Emergence creates the passages between levels as it proceeds. They are not already there waiting to be travelled, and each step alters what can happen next, as when one body in a field of debris grows large enough that its gravity reorders every later collision.

Scale-sensitive variable: a variable whose increase changes what a system can do, not only how fast or how accurately it does it.

Surprise: an observer's registration of emergence: a result that exceeds what the observer's knowledge could anticipate at its level. Surprise is relative to the observer; emergence is not reducible to it. Explanatory knowledge is, in Cliff's phrase, surprise accounting for itself. Justificatory knowledge absorbs surprise back into existing frameworks of sameness and difference, reassuring us that nothing has really changed.

Probability: a measure, registered by an observer, of how a history of transformations has been distributed across a closed set of possible alternatives. Probability is neither an a priori given nor a property of single events. It emerges when four conditions hold together.

• A structure of possibility. Some transformations are possible and others impossible, or declared off limits. Only a closed set of alternatives can be measured, because a measure is taken across the whole set (the probabilities sum to one). The impossible is what closes the set.

• A history. Transformations actually occur, repeatedly, over time.

• Storage and transmission. The history persists and travels as knowledge (parish registers, life tables, a training corpus), so that past occurrences can bear on present ones. Without it there are only events with no measurable relationship to one another.

• A register that makes the history measurable. An observer, meaning any system that registers rather than necessarily a subject, declares instances equivalent so that they can be counted, and relates the count to the space of alternatives. Measurement depends on declared sameness.

A large language model makes all four conditions visible at once. Its vocabulary and architecture supply the structure of possibility, with some continuations masked or trained off limits. Its training corpus is the history. Its weights store and transmit that history. Its context window, read through attention, is the register on which every probability is conditioned. And it unfolds in time, each sampled token becoming history for the next.