Glossary

abiogenesis The emergence of life from non-life. See also: life.

Acoela An ancient order of small marine worms that diverged from other animals more than 550 million years ago; among the earliest bilaterians, they have no gut, or circulatory or respiratory systems.

agent An entity who/that does or acts; in a directed interaction between two entities, the active one may be referred to as the agent, while the recipient of the action is the patient. See also: moral agent, patient.

animism The belief (held by many traditional societies) that objects, places, and creatures possess spirits, and thus that all things are alive and interconnected. See also: vitalism.

anthropic principle The principle that the fundamental laws of the observed universe must be compatible with the existence of the conscious, intelligent observer positing those laws. See also: fine-tuning (cosmology).

anthropocentrism The belief that humans are the most important entities in the universe; viewing or understanding all other entities only in terms of their value to, or in comparison with, humanity. See also: anthropomorphism.

anthropomorphism The (often inappropriate) attribution of human characteristics, behaviors, or emotions to nonhuman entities, such as other animals, objects, or phenomena. See also: anthropocentrism.

Artificial General Intelligence (AGI) AI capable of a wide, human-like range of tasks and behaviors; the term was popularized when models capable only of single tasks started to be called AI. See also: Artificial Narrow Intelligence.

Artificial Intelligence (AI) A term coined in 1955 by John McCarthy and colleagues for “the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.” See also: cybernetics, Good Old-Fashioned Artificial Intelligence.

Artificial Narrow Intelligence (ANI) An AI system designed only to do one task or a narrow range of tasks; examples include character recognition, face detection, and product review sentiment analysis. See also: Artificial Intelligence.

Artificial Superintelligence (ASI) A general AI or AGI with broadly superhuman performance (though there is no consensus on what that constitutes); such a superintelligence would presumably be capable of recursive self-improvement, potentially creating an intelligence explosion. See also: Artificial General Intelligence, intelligence explosion.

attention A cognitive process wherein certain perceptual information is allocated preferential mental processing, enabling the formation of a coherent view of the most important aspects of environment, often guided by factors such as novelty, salience, relevance, and task demands; also, a family of machine learning techniques designed to selectively concentrate neural net processing in a given layer only on certain activations or relationships in the previous layer. See also: Transformer.

attention layer The kind of neural net layer that powers the Transformer architecture, implementing content-based pattern-matching and selective attention over a context window of fixed size. See also: attention, Transformer.

autocatalysis A process whereby a reaction is catalyzed by one of its own products, causing the reaction to speed up exponentially until limited by the availability of a reactant. See also: autocatalytic set.

autocatalytic set As conceived by biochemist Stuart Kauffman, a network of interacting molecules where each molecule can be catalytically produced from other molecules in the set, thereby creating a self-sustaining and self-replicating system; I use this term more generally for any set of entities, including whole organisms or code instructions, with the same functional properties (mutual catalysis and collective self-replication). See also: abiogenesis, autocatalysis.

backward causality When the effect of an event seems to paradoxically precede its own cause. See also: entensional.

bff A minimalistic computer simulation demonstrating the emergence of functional replicators out of random noise; based on a variant of the Brainfuck programming language. See also: abiogenesis, Brainfuck.

blindsight A neurological phenomenon where, following removal or destruction of the visual cortex, people can still respond to visual stimuli, yet claim to be blind; discoverers Nicholas Humphrey and Lawrence Weiskrantz considered blindsight “seeing without consciousness,” but perhaps it simply reflects lack of connectivity between the brain regions processing vision and generating language.

Brainfuck A minimalistic, esoteric programming language closely resembling a Turing Machine, devised in 1993 by Swiss physics student Urban Müller. See also: bff.

calculus ratiocinator A hypothetical (and unrealizable) universal logical calculation framework described in the writings of Gottfried Wilhelm Leibniz, wherein the manipulation of symbolic propositions could be used to prove the truth or falsity of any statement; similar in spirit to Good Old-Fashioned AI. See also: characteristica universalis.

cellular automaton A grid of cells, each in a specific state, that evolves over time based on a set of rules determined by the states of neighboring cells; developed by John von Neumann and collaborators from the 1940s onward. See also: neural cellular automaton.

characteristica universalis An (unrealizable) universal symbolic language imagined by Gottfried Wilhelm Leibniz to represent all knowledge and ideas in a clear and unambiguous way, enabling logical calculation and reasoning. See also: calculus ratiocinator.

choice blindness The phenomenon studied in cognitive science where individuals fail to notice mismatches between their intended choice and the outcome they’re presented with, demonstrating the pliability of our preferences and memories; the term was first coined by a team of Swedish researchers led by cognitive scientist Petter Johansson. See also: interpreter, the.

comb jelly A phylum of ancient marine invertebrates, the Ctenophora, dating back to the Ediacaran, characterized by biradial symmetry (radial symmetry like a jellyfish, but sometimes including a pair of tentacles) and a fringe of cilia (tiny hair-like structures) that move in a way reminiscent of the teeth of a comb; they are delicate, near-transparent, and often bioluminescent.

competence without comprehension A concept introduced by philosopher Daniel Dennett to describe systems or entities that can perform tasks effectively without understanding how or why they are doing so; perhaps more properly interpreted as an inability to generalize a skill beyond a narrow domain, or to explain it using language.

complexification The process whereby things become more complex over time; I argue that this characterizes the evolution of life, intelligence, and technology, and occurs because existing dynamically stable components can sometimes combine (or “compose”) to form larger dynamically stable wholes, which can themselves serve as components for further composition. See also: symbiogenesis.

composition (functional) When one or more functions are composed (i.e. combined) to create another function. See also: recursion (computing).

computational neuroscience The study of nervous systems and brains as computational or information-processing systems. See also: functionalism, NeuroAI.

computronium A substance that computes with massive parallelism; the term was originally coined in the 1980s by physicist and computer scientist Norman Margolus to describe a hypothetical material optimized at the atomic or molecular level to perform computation as efficiently as the laws of physics allow, though I use it more broadly to describe the organized, computational state of matter characterizing life. See also: massively parallel.

cybernetics A field of study founded by mathematician Norbert Wiener and colleagues in the 1940s focusing on feedback, prediction, control, and learning as exhibited by living systems and some machines; the intellectual predecessor of modern machine learning. See also: Artificial Intelligence, machine learning.

dark room problem A challenge in predictive coding and active inference theories of the brain, highlighting the paradox that if organisms seek to minimize prediction errors, they might prefer unchanging, low-stimulation environments (like a dark room), which contradicts observed behavior where organisms actively seek novel and stimulating experiences. See also: predictive brain hypothesis.

Dress, The A viral photograph that became an internet sensation in 2015, making people question whether “The Dress” was blue and black or white and gold, due to an optical illusion related to color perception; the leading theory is that one’s inference depends on whether one is a night owl, which affects statistical priors on lighting.

dualism A philosophical and religious concept, often associated with René Descartes, positing that the mind and body have two fundamentally different natures, one immaterial (the mind) and the other physical (the body).

dynamic instability When a small variation in initial conditions can cause wildly diverging outcomes, often associated with the butterfly effect. See also: dynamic stability.

dynamic kinetic stability (DKS) A concept introduced by the chemist Addy Pross to describe systems that persist through time not by being static, but through continuous dynamic change and turnover, such as the self-replication of molecules or the metabolic activity of living organisms. See also: dynamic stability.

dynamic stability In this book, used to describe systems that cycle through a loop of states (in dynamical systems theory, a stable limit cycle), thus preserving their identity through time even though they are in constant flux; typical of living systems. See also: dynamic instability.

Effective Altruism (EA) A utilitarian philosophical and social movement that advocates using evidence and reason to determine the most effective ways to improve the world and directing resources—time, money, or skills—toward actions that achieve the greatest positive impact; Peter Singer, Toby Ord, and William MacAskill are among its philosophical standard-bearers. See also: utilitarianism.

efficient cause Per Aristotle, an agent or force that produces an effect or change in another object or being. See also: final cause.

élan vital A philosophical term introduced by French philosopher Henri Bergson in his 1907 work Creative Evolution, referring to a vital force or impulse inherent in all living organisms, driving growth, change, and the creative evolution of life. See also: psychic pneuma, vitalism.

embedding A learned representation of data where objects (e.g., words or images) are mapped to continuous, low-dimensional vector spaces, capturing their semantic or structural relationships and enabling generalization for tasks like classification, clustering, or translation. See also: Word2Vec.

endosymbiosis A symbiotic relationship in which one organism comes to live inside another organism, conferring mutual benefit and potentially resulting in a new form of life; most famously associated with the work of Lynn Margulis, who showed that mitochondria and chloroplasts, today eukaryotic organelles, were originally free-living bacteria that were engulfed by ancestral host cells related to archaea. See also: symbiogenesis, symbiosis.

entensional A term coined by cognitive scientist Terrence Deacon referring to systems (including simple and engineered ones) that operate according to physical laws, but that are goal-directed or purposive. See also: purposive, teleology.

existential risk (X-risk) A potential event or process that could either cause the extinction of humanity or irreversibly curtail its long-term potential, encompassing threats such as nuclear war, advanced artificial intelligence misalignment, pandemics, or catastrophic climate change.

felicific calculus An approach proposed (though never fully specified) by philosopher Jeremy Bentham for quantifying happiness or pleasure to evaluate the moral rightness of actions, based on factors such as intensity, duration, certainty, proximity, and extent of the resulting pleasure or pain. See also: utilitarianism.

few-shot learning A method in machine learning where an AI system learns to classify or interpret new data from only a small number of examples; this mimics human ability, which requires fewer instances than traditional machine learning to understand something new. See also: machine learning, one-shot learning.

final cause Per Aristotle, the purpose or goal for which a thing exists or an action is undertaken, being the “end” or “telos” towards which that thing strives. See also: efficient cause, teleology.

fine-tuning (cosmology) The puzzle posed by the fact that certain fundamental physical and cosmological constants appear to have precisely the values necessary to support the existence of life in the universe. See also: anthropic principle.

fine-tuning (machine learning) When a pretrained model, such as a language model, is further trained or “tuned” on a more specific task or dataset, thereby enhancing its precision and applicability to a particular problem or domain. See also: machine learning.

functionalism A computational position held by Alan Turing and John von Neumann emphasizing that what something “is” is defined by what it does, which can be independent of implementation, hence multiply realizable; the corresponding philosophical perspective, championed by Hillary Putnam and others, conceives mental states in terms of their role in a system or their behavioral effects, thus opening the possibility that different kinds of systems, including non-biological ones, could manifest the same types of mental states. See also: computational neuroscience, multiple realizability.

Gaia hypothesis Posits that the Earth is homeostatic, functioning as a self-regulating system in which living organisms interact with their inorganic environment to maintain conditions conducive to life, such as stable temperatures, atmospheric composition, and ocean salinity; proposed by James Lovelock in 1972 and later developed with Lynn Margulis.

generalization (machine learning) A model’s ability to perform well on unseen data by capturing the underlying patterns in the training data rather than merely memorizing it, ensuring it can make accurate predictions or decisions for new inputs not encountered during training. See also: perceptual invariance.

Good Old-Fashioned Artificial Intelligence (GOFAI) A term coined by philosopher John Haugeland in 1985 for the classical (and largely ineffective) approach to AI that emphasizes high-level symbolic representations and rule-based manipulation to model intelligent behavior; this paradigm relies on explicitly programmed knowledge and logical reasoning, contrasting with cybernetics and its modern successor, machine learning. See also: Artificial Intelligence, cybernetics, machine learning.

grandmother cell A hypothetical neuron that activates in response to very specific complex stimuli, such as a person’s own grandmother, postulated (with tongue in cheek) by Jerry Lettvin in 1969; while neurons can indeed respond to highly specific stimuli, they do so in populations, making it unlikely that complex concepts are represented by unique neurons.

hallucination (machine learning) A generated output by a model (such as a large language model) that is factually incorrect, nonsensical, or fabricated, despite appearing plausible or confident, often arising due to limitations in the model’s training data or inference mechanisms.

Homo economicus A theoretical individual proposed in economic theory who acts consistently rational and narrowly self-interested, always making decisions that maximize utility given the information available. See also: utilitarianism.

homuncular fallacy Mistakenly attributing selfhood or subjectivity only to a particular part of the brain. See also: homunculus.

homunculus A term derived from Latin meaning “little man,” often used in discussions of philosophy and neuroscience to describe the misleading idea of a small being inside the brain who interprets sensory information and controls our actions.

Hydra A genus of small, simple aquatic animals of the phylum Cnidaria, of special biological interest for their distributed nerve net, regenerative capabilities, and immortality.

Industrial Revolution A period from the late eighteenth to mid-nineteenth century marked by significant developments in agricultural, manufacturing, mining, transport, and technology sectors (most of which were enabled by fossil fuels and the steam engine) which dramatically altered the socio-economic and cultural conditions in Britain, and later spread to Western Europe and North America, often cited as marking the transition to modern economic growth; some economists posit subsequent Industrial Revolutions based on electrification and information technology. See also: Major Evolutionary Transition.

inference An integral part of both human reasoning and AI, inference involves making logical deductions or predictions based on existing information or data, such as interpreting the contents of an image or the meaning of a word in a sentence. See also: machine learning.

inpainting An image restoration technique where missing or corrupted parts of images are filled in with plausible content, based on information and context from the surrounding areas and from similar images. See also: masking.

intelligence The ability to model, predict, and influence one’s future; it can evolve in relation to other intelligences to create a larger symbiotic intelligence.

intelligence explosion An event hypothesized by mathematician I. J. Good in which an artificial intelligence system becomes capable of autonomously improving its own design, leading to rapid, unforeseen advancements in technology and potentially surpassing human intellectual capacity; as I use the term, intelligence explosions are a broader evolutionary phenomenon driven by positive feedback in mutual theory-of-mind modeling, and include the rapid growth in human brain size over the past several million years. See also: existential risk, theory of mind.

intentionality level The depth of representation a person, system, or model has about the self and others; intentionality levels range from zero, representing inanimate objects, to one, representing only immediate experience, to higher levels characterized by higher-order theory of mind, including self-reflection and understanding others’ minds. See also: theory of mind.

interpreter, the (neuroscience) Refers to the brain’s left hemisphere mechanism (for most right-handed people) that creates explanations and narratives to make sense of our experiences and actions, even when it lacks complete information about their true causes; proposed by neuroscientist Michael Gazzaniga to explain observations from split-brain patients, and consistent with many other findings, including choice blindness. See also: choice blindness.

is/ought dichotomy A distinction proposed by philosopher David Hume between descriptive statements (is: what is the case) and prescriptive or normative statements (ought: what ought to be the case), to argue that values or morality cannot be deduced from factual knowledge alone.

joint probability distribution A mathematical description of the likelihood of two or more events occurring together, often represented in the form of a table or matrix, which can provide important insights into the dependencies and interactions between different variables in a system.

Krebs cycle A key metabolic pathway discovered by biochemist Hans Krebs, the Krebs cycle (also known as the citric acid cycle or TCA cycle) involves the oxidation of acetyl-CoA into carbon dioxide and the transfer of energy to produce NADH, FADH2, and ATP, which cells use for energy. See also: reverse Krebs cycle.

latent variable A variable that is not directly observed but inferred through a mathematical model from other variables that are observed; latent variables are often used in machine learning to explain hidden or underlying structures in complex data. See also: embedding.

lateral inhibition A ubiquitous neurological principle first presented in a 1957 paper by Keffer Hartline and Floyd Ratliff in the context of the visual systems of horseshoe crabs, where an excited neuron reduces the activity of its immediate neighbors, leading to a contrast enhancement that aids perception of edges and spatial differences.

learning The process whereby an organism or artificial intelligence system acquires knowledge, skill, understanding, or the ability to perceive nuance through experience, study, experiment, instruction, or error minimization, ultimately resulting in the modification of behavior or the ability to predict outcomes based on patterns or inferences. See also: machine learning.

life A self-modifying, computational state of matter arising from selection for dynamic stability; it evolves through the symbiotic composition of simpler dynamically stable entities.

logical proposition A statement that has a distinct truth value—either true or false—which can form the basic building block of logical arguments and reasoning. See also: calculus ratiocinator.

Longtermism An ethical standpoint emphasizing the long-term future and advocating for decisions and actions that consider their potential impact on many generations to come; most controversially, its adherents may use utilitarian principles that weigh equally the interests of all people, including (potentially vast numbers of) hypothetical future people. See also: utilitarianism.

machine learning (ML) An approach to building intelligent machines that provides systems with the ability to learn and improve from experience without being explicitly programmed, involving the construction and study of algorithms that can make predictions or decisions based on patterns in data; in the tradition of cybernetics, and often contrasted with Good Old-Fashioned Artificial Intelligence. See also: Artificial Intelligence, Good Old-Fashioned Artificial Intelligence.

Major Evolutionary Transition (MET) A key point in the evolution of life that results in a significant increase in complexity through symbiotic interdependence, such as the shift from single-celled to multicellular organisms, or from asexual to sexual reproduction; proposed by biologists John Maynard Smith and Eörs Szathmáry. See also: symbiogenesis.

Mary’s Room (or Mary the Super-Scientist) A philosophical thought experiment proposed by Frank Jackson, where Mary, a scientist who knows everything about the physical science of color but has only ever experienced black and white, sees color for the first time, raising questions about the nature of subjective experiences or qualia. See also: qualia.

masking (machine learning) An unsupervised training technique in which parts of the input are hidden, and the model’s task is to reconstruct these parts as accurately as possible; often used to pretrain large language models, as well as image and video autoencoders.

massively parallel In computer science, the use of a large number of processors (or computers) to simultaneously perform coordinated computations; such parallelism is essential for running large neural networks in any reasonable amount of time, and is ubiquitous in neuroscience and more broadly in biological computation. See also: computronium, neural cellular automaton.

McCulloch-Pitts neuron A simplified model neuron where each neuron receives one or more Boolean input signals and produces a Boolean output, thus working like a logic gate; proposed by Warren McCulloch and Walter Pitts in 1943.

mechanical philosophy A worldview originating in the seventeenth century, most notably advocated by philosopher René Descartes, positing that natural phenomena, including living bodies, can be explained like machines, with their functions arising purely from their physical structures and movements, without needing to invoke unobservable forces or principles.

model A mathematical or computational representation of a system or phenomenon that allows its future (or otherwise unobserved) behavior to be predicted. See also: machine learning.

moral agent An individual or entity capable of making ethical decisions and being held accountable for them. See also: agent, moral patient.

moral patient An individual or entity whose well-being and interests are considered in ethical decision-making processes and moral actions. See also: moral agent, patient.

multifractal A mathematical object characterized by a non-integer, or fractal, dimension that exhibits characteristic structure at every scale; but unlike a fractal, this structure may vary at different scales.

multilayer perceptron (MLP) A type of artificial neural network first proposed by Frank Rosenblatt in the 1950s, consisting of multiple layers of interconnected neurons where each layer’s output serves as input for the next layer, enabling the model to learn and represent complex functions and concepts. See also: neural network.

multiple realizability A concept in philosophy of mind and computational neuroscience that posits the same mental state, process, or function can be realized with different physical substrates—such as the human brain, a computer, or an alien physiology—given the correct functional organization.

neural cellular automaton (NCA) A computational model combining cellular automata, neural nets, and Turing’s concept of morphogenesis, where each cell evaluates a (shared) neural net whose inputs and outputs are local values, akin to morphogens; the concept was developed by Alex Mordvintsev in 2020. See also: cellular automaton.

neural network (artificial) A massively parallel machine learning model inspired by the human brain’s interconnected neurons; when implemented with multiple layers, each layer of neurons runs in parallel but the layers themselves run sequentially, with each layer feeding its outputs to the next layer as inputs. See also: machine learning, neural network (biological).

neural network (biological) Interconnected neurons in an animal’s nervous system.

NeuroAI A hybrid research field combining neuroscience with artificial intelligence, striving to imitate the human neural system in order to enhance AI’s adaptability, functionality, and learning efficiency; the field overlaps significantly with computational neuroscience. See also: computational neuroscience.

one-hot coding A method used in machine learning where a population of neurons, typically an output layer, is trained to represent a categorical variable such that only one neuron is “on” (has value near one) while the others are all “off” (have values near zero); also sometimes referred to as “winner-takes-all.” See also: lateral inhibition.

one-shot learning The ability to learn a general pattern, category, or other regularity from a single example or exposure. See also: machine learning.

Palomilla A wheeled robot that implemented light-seeking behavior (its name means “moth” in Spanish), designed by Norbert Wiener in the late 1940s to demonstrate negative feedback and goal-directed behavior, foundational concepts in cybernetics. See also: cybernetics.

patient (philosophy) An entity that is acted upon or affected by an agent’s actions. See also: agent, moral patient.

perceptual invariance A concept in cognitive science describing how humans and other animals recognize objects, sounds, or other stimuli as being the same across various different viewing angles, distances, lighting conditions, or other potentially complicating factors, suggesting powerful computational processing capabilities in the brain.

phase transition A sudden change in a system’s properties, such as density, magnetization, or conductance, often as a result of changing external conditions, like temperature or pressure; the transition from water to ice is a common example.

philosophical zombie A hypothetical being that is indistinguishable from a human in behavior and physical appearance, but lacks conscious experience; popularized by philosopher David Chalmers to question whether consciousness can be reduced to purely physical processes. See also: qualia.

Pirahã A people of the Amazonas state of Brazil known for unique characteristics of their language, including the absence of numbers, absence of recursion, and possibly the smallest inventory of phonemes observed in any human language; notably studied by linguist and former missionary Daniel Everett. See also: recursion (linguistics).

predictive brain hypothesis A neuroscientific theory suggesting that the brain is a predictor that constantly models the future based on past experiences; these models are continuously tested against sensory information to minimize prediction errors.

pretraining An approach to machine learning in which a model is initially trained on a large, general dataset (usually using unsupervised learning) before being fine-tuned on a smaller, specific dataset. See also: machine learning.

psychic pneuma According to ancient Greek philosophers and René Descartes, an elusive substance believed to be the vital force or breath of life that penetrates and harnesses inanimate matter, often associated with the human soul and intellect. See also: dualism, élan vital.

psychon A hypothetical unit of cognitive activity or fundamental element of thought; hypothesized by Warren McCulloch and Walter Pitts in 1943 to consist of a single logical proposition, i.e. one bit of information. See also: logical proposition.

purposive (or purposeful) Functions or behaviors that are goal-directed, therefore suggestive of agency or intentionality. See also: backward causality, entensional.

qualia The subjective or qualitative properties of experiences, such as the redness of a rose or the taste of sugar, which philosopher Frank Jackson and others have argued cannot be fully understood through scientific or computational analysis; hence the supposed “hard problem” in consciousness studies.

recursion (computing) A method in computing where a function, in order to accomplish a task, calls itself with modified inputs, enabling problems to be broken down into smaller, similar problems until reaching a base case that can be solved directly; frequently used in algorithms and data structures. See also: composition.

recursion (linguistics) A linguistic phenomenon in which a grammatical rule can be applied to the result of its own application, allowing for constructions such as nested clauses and indefinitely extended sentences; introduced by linguist Noam Chomsky in his theory of generative grammar. See also: Pirahã.

reinforcement learning (RL) A machine learning method where an AI system learns to make decisions by receiving rewards or penalties for its actions; often used in coaching robots or recommender systems, and inspired originally by Pavlovian or reward-based learning in lab animals. See also: machine learning.

relational quantum mechanics (RQM) An interpretation of quantum mechanics proposed by Carlo Rovelli in 1996 suggesting that the state of a quantum system is relative to the observer, meaning that different observers can have different descriptions of the same system, and these descriptions are valid within their own frames of reference; according to RQM, there is no objective, absolute state of a system independent of all observers.

representation learning An approach to machine learning that allows a system to automatically discover and learn useful representations from raw data for improved prediction or inference without explicit feature engineering; fundamental to deep learning. See also: machine learning.

reverse Krebs cycle A biochemical process believed to have played a key role in the origin of life; it performs the reverse operations of the conventional (forward) Krebs cycle by consuming carbon dioxide and hydrogen to generate organic compounds, a pathway mainly found today in ancient bacteria and archaea. See also: Krebs cycle.

RNA world An influential hypothesis suggesting an early stage of evolution where RNA molecules acted both as genetic material and as catalysts for chemical reactions, potentially predating DNA and proteins.

robustness The ability of a purposive system to keep working in the face of environmental changes, challenges, disruptions, or uncertainties.

rubber hand illusion An experimental phenomenon in which a person comes to perceive a fake rubber hand as part of their own body, typically through having the real and fake hands simultaneously stroked in the same way, exemplifying the flexibility and malleability of body perception; first reported by Matthew Botvinick and Jonathan Cohen in 1998.

saccade A rapid movement made by the eye when changing its fixation point in a scene, usually occurring approximately two or three times per second.

Sally-Anne Test A psychological test popularized by Simon Baron-Cohen, Alan Leslie, and Uta Frith, used to gauge a subject’s capacity for understanding that others have viewpoints and knowledge distinct from their own; often employed to assess Theory of Mind deficits, especially in children with autism. See also: theory of mind. See also: theory of mind.

scaling law The way performance or some other property of a system varies as a function (typically a power-law) of some parameter; for example, metabolic rate varies as an approximate power-law of animal size, and task performance of a neural network often varies as a power-law of the dataset size or model size.

semantics (linguistics) The aspect of language concerning meaning and interpretation, as opposed to properties like grammar and syntax.

sequence modeling A machine learning method in which historical data inputs, such as time series or text data, are used to predict future data points or complete missing ones. See also: machine learning.

singularity A hypothesized future point in time when advances in technology (particularly advancements in artificial intelligence) enter a self-improving feedback loop so extreme that we are unable to make any prediction about the course of history beyond this point; popularized by futurist Ray Kurzweil.

social brain hypothesis A hypothesis proposed by British anthropologist Robin Dunbar asserting that the complex social interactions within primate groups, including humans, have driven the evolution of larger brains to handle these complexities. See also: intentionality level, theory of mind.

social intelligence hypothesis A hypothesis proposed by neuroscientist Nicholas Humphrey in 1976 suggesting that the driving force behind the evolution of high intelligence among hominins is the complexity of social interactions, requiring skills such as empathy, deception, cooperation, and competition.

sphexish A term coined by cognitive scientist Douglas Hofstadter to describe seemingly intelligent behavior that is actually mechanistic or deterministic, based usually on a fixed set of responses or rules; named after the Sphex wasp, which entomologist Jean-Henri Fabre believed (probably incorrectly) to exhibit such behavior.

strange loop A concept put forth by cognitive scientist Douglas Hofstadter referring to a situation in which, whenever one steps upwards or downwards through the layers of some hierarchical system, one unexpectedly finds oneself back at the starting point; Hofstadter believes this kind of self-referentiality is fundamental to consciousness.

swing (rowing crew) A term borrowed from rowing referring to a state of perfect harmony and synchronicity in a crew; used metaphorically (or not) to describe the way multiple predictive agents can coalesce into a single larger agent.

symbiogenesis An evolutionary process proposed by biologist Konstantin Mereschkowski and further developed by Lynn Margulis, whereby symbiosis can lead to new life forms through the long-term union of separate but interdependent organisms. See also: symbiosis.

symbiosis A long-term interaction between two different species (or other entities) in which both benefit, or one benefits and the other neither benefits nor is harmed. See also: symbiogenesis.

teleology A philosophical concept suggesting that certain phenomena, systems or actions are directed toward ends or goals; while often maligned because it seems to imply intelligent design, Wiener, Rosenbleuth and Bigelow wrote an essay in 1943 explaining how negative feedback systems in general have a teleological character, and experiments like bff show that purposive behavior can arise without any intelligent designer. See also: bff, cybernetics.

Theory of Everything (ToE) A comprehensive theoretical framework that would include and reconcile all known physical phenomena and laws, including the four fundamental forces—gravity, electromagnetism, the weak nuclear force, and the strong nuclear force—into a single coherent model of the universe.

theory of mind (ToM) The ability to attribute mental states like beliefs, intents, desires, emotions, knowledge, etc., to oneself and others and to understand that others have beliefs, desires, emotions, and knowledge that are different from one’s own. See also: intelligence explosion.

“thing in itself” (Ding an sich) The underlying reality of an object, independent of any observer’s experience or perception of it; Immanuel Kant coined the term to make a distinction with the “phenomenon,” the object as it appears to us.

Three Laws of Robotics A set of rules devised by science fiction author Isaac Asimov to govern AI behavior, specifying that a robot 1) may not injure a human being, 2) must obey orders given by humans except where such orders would conflict with the first law, and 3) must protect its own existence as long as such protection does not conflict with the first or second law.

Transformer A popular AI model first introduced in the 2017 paper “Attention is All You Need” by Vaswani et al. that uses an “attention” mechanism to weight the importance of different pieces of input data; Transformers power most modern LLMs, including Gemini, Claude, and ChatGPT.

umwelt A term introduced by biologist Jakob von Uexküll referring to the perceptual world in which an organism exists and interacts, uniquely filtered and shaped by the organism’s specific sensory, motor, and cognitive capabilities and experiences.

universals Abstract concepts, properties, or features, such as redness or roundness, thought to exist independently of particular instances or objects, a concept explored extensively by philosophers such as Plato and Aristotle.

utilitarianism A philosophy developed by Jeremy Bentham and John Stuart Mill that emphasizes the greatest happiness for the greatest number; the ethical value of an action is determined by its “utility,” or contribution to overall wellbeing, which is presumed to be representable as a number. See also: utility.

utility A concept central to economics and game theory representing the total satisfaction or fulfilment that a person derives from consuming a good or service; utilitarian philosophers such as Jeremy Bentham posit that actions should be evaluated by their contribution to overall utility, or “the greatest good for the greatest number.” See also: utilitarianism.

vitalism The belief that life and its phenomena are due to a vital principle distinct from physical or chemical forces; no longer considered reputable by most scholars, though there are a few modern proponents, including philosopher Jane Bennett. See also: animism, élan vital.

von Neumann architecture A computer system architecture proposed by mathematician and physicist John von Neumann which still forms the basis of most modern computers, where stored-program memory and data share the same memory space, and this memory is connected to a central processing unit by a bus.

von Neumann replicator A theoretical self-replicating machine proposed by mathematician and physicist John von Neumann, which can replicate itself using raw materials from its environment, thereby exhibiting exponential population growth given a suitable medium and an energy source.

Wiener sausage (statistical physics) A mathematical construct representing the volume swept out by a ball of fixed radius as it moves along the trajectory of a Wiener process (a random walk or Brownian motion) in space; it is used to study problems involving diffusion and encounter probabilities, and provides insights into how a random path explores the environment over time.

Word2Vec An unsupervised machine learning technique that creates word embeddings, which are vector representations of words based on the “company they keep” (i.e. the statistics of the surrounding words); developed by Tomáš Mikolov at Google Research in 2013. See also: embedding.