[{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/ai/","section":"Tags","summary":"","title":"AI","type":"tags"},{"content":" Multi-agent demos are persuasive: one agent proposes, another criticises, a third summarises. The theatre of disagreement suggests independent thought. Sometimes the disagreement is real enough to improve results. Often it is correlated sampling with role-play labels. This essay offers criteria for independence that do not depend on marketing diagrams.\nWhat independence would mean # For agents \\(A_1, \\ldots, A_n\\) producing judgments \\(J_i\\), statistical independence would require:\n\\[ P(J_1, \\ldots, J_n) = \\prod_{i=1}^{n} P(J_i) \\]In deployed systems this almost never holds. Agents share:\nBase model weights Overlapping context windows Similar system prompts The same tool outputs Shared structure induces positive dependence. Role prompts (“you are a critic”) change the style of \\(J_i\\) without guaranteeing independent evidence.\nKinds of diversity # Mechanism What it diversifies Failure mode Role prompts Rhetorical stance Same model, same priors Temperature / sampling Surface form Still one distribution Different models Architecture / training Vendor correlation remains Private tools / data Evidence base Integration errors Human subgroups Values and expertise Cost and coordination Meaningful independence tracks private information and distinct error patterns, not the number of chat bubbles.\nWhen multi-agent helps # Empirical multi-agent setups can help when:\nDecomposition maps to real interfaces (retrieve, code, verify) Adversarial roles catch classes of errors the solo prompt misses External tools give agents different observations They help less when the task is a single closed-book judgment and every agent is the same model with a costume.\nflowchart LR subgraph correlated [Correlated ensemble] M[Shared model] --\u003e R1[Role: pro] M --\u003e R2[Role: con] M --\u003e R3[Role: judge] end subgraph stronger [Stronger separation] M1[Model A + data X] --\u003e J1[Judgment] M2[Model B + data Y] --\u003e J2[Judgment] J1 --\u003e AGG[Aggregator] J2 --\u003e AGG end A practical test # Before trusting a “debate,” ask:\nIf I reshuffle role labels, does the conclusion change? If I ablate all but one agent, how much quality drops? Do agents have non-overlapping evidence? Is the aggregator able to prefer minority views when they are better calibrated? If quality barely drops under ablation, the system was a single thinker with extra latency.\nClosing position # Multiple AI agents can improve workflows through specialisation and tooling. Independent thought is a stronger claim and requires independence of information and errors, not only concurrency. Design for explicit diversity of evidence; do not mistake dialog format for epistemology.\nNotes and references # Demonstration essay. Multi-agent research is active; treat architectural recommendations as provisional and validate on your task distribution.\n","date":"5 May 2026","externalUrl":null,"permalink":"/investigations/can-multiple-ai-agents-think-independently/","section":"Investigations","summary":"Independence is a property of information flow and incentives, not of process count. Multi-agent systems often share models, prompts, and training—correlation in disguise.","title":"Can Multiple AI Agents Think Independently?","type":"investigations"},{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/epistemology/","section":"Tags","summary":"","title":"Epistemology","type":"tags"},{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/multi-agent/","section":"Tags","summary":"","title":"Multi-Agent","type":"tags"},{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/systems/","section":"Tags","summary":"","title":"Systems","type":"tags"},{"content":"","date":"5 May 2026","externalUrl":null,"permalink":"/tags/","section":"Tags","summary":"","title":"Tags","type":"tags"},{"content":"","date":"2 April 2026","externalUrl":null,"permalink":"/tags/algebra/","section":"Tags","summary":"","title":"Algebra","type":"tags"},{"content":"","date":"2 April 2026","externalUrl":null,"permalink":"/tags/algorithms/","section":"Tags","summary":"","title":"Algorithms","type":"tags"},{"content":"","date":"2 April 2026","externalUrl":null,"permalink":"/tags/compilers/","section":"Tags","summary":"","title":"Compilers","type":"tags"},{"content":"","date":"2 April 2026","externalUrl":null,"permalink":"/tags/decision-trees/","section":"Tags","summary":"","title":"Decision Trees","type":"tags"},{"content":"","date":"2 April 2026","externalUrl":null,"permalink":"/tags/ml/","section":"Tags","summary":"","title":"ML","type":"tags"},{"content":" Algebra looks continuous and clean; decision trees look discrete and procedural. In practice, much of computational mathematics is a translation between the two: an expression defines a function; evaluating it under real machine constraints introduces cases, branches, and control flow. This investigation makes that translation explicit.\nThe quiet branching inside formulas # Consider a piecewise definition:\n\\[ f(x) = \\begin{cases} x^2 \u0026 \\text{if } x \\ge 0 \\\\ -x \u0026 \\text{if } x \u003c 0 \\end{cases} \\]On paper this is algebra. In a program it is already a decision tree with two leaves. Compilers, CAS systems, and automatic differentiation frameworks spend enormous effort managing such cases—singularities, domains, and numeric guards—often invisible in the textbook formula.\nExpressions as DAGs; evaluation as trees # An arithmetic expression is naturally a directed acyclic graph of operations. Evaluation order and short-circuiting turn parts of that graph into a tree of decisions:\n(x ≥ 0)? / \\ square neg | | x x The same idea scales. Gaussian elimination is a sequence of pivot decisions. Simplex methods branch on entering variables. Piecewise-polynomial models in statistics are forests of local algebras.\nA minimal example # def f(x: float) -\u0026gt; float: # Algebra as control flow if x \u0026gt;= 0: return x * x return -x def f_as_tree(x: float) -\u0026gt; float: \u0026#34;\u0026#34;\u0026#34;Same function, annotated as an explicit tree walk.\u0026#34;\u0026#34;\u0026#34; node = (\u0026#34;cmp\u0026#34;, \u0026#34;x\u0026gt;=0\u0026#34;, (\u0026#34;op\u0026#34;, \u0026#34;square\u0026#34;, \u0026#34;x\u0026#34;), (\u0026#34;op\u0026#34;, \u0026#34;neg\u0026#34;, \u0026#34;x\u0026#34;)) kind, *rest = node if kind == \u0026#34;cmp\u0026#34;: _, true_branch, false_branch = rest branch = true_branch if x \u0026gt;= 0 else false_branch return eval_leaf(branch, x) raise ValueError(\u0026#34;unknown node\u0026#34;) def eval_leaf(node, x): _, op, _ = node return x * x if op == \u0026#34;square\u0026#34; else -x The algebra did not disappear; it was scheduled.\nWhy the translation matters # Verification # Proofs about floating-point programs must reason about branches. Interval arithmetic and abstract interpretation are, in part, algebra over sets with join operations at merge points—decision structure again.\nMachine learning # Gradient-boosted decision trees and mixture-of-experts models make the dualism obvious: each leaf holds a simple algebraic model; the tree allocates inputs to leaves. Neural networks hide the branching in continuous gates, but routing and sparsity research keeps rediscovering discrete decisions.\nHuman understanding # Experts often think in cases (“if the matrix is singular…”, “if the market is in regime B…”). Writing only the closed form can conceal the real algorithm.\nflowchart TD A[Symbolic expression] --\u003e B{Domain / guards} B --\u003e|case 1| C[Local algebra] B --\u003e|case 2| D[Local algebra] C --\u003e E[Numeric evaluation] D --\u003e E E --\u003e F[Result / proof obligation] A small formal view # Let \\(\\mathcal{E}\\) be a set of expressions and \\(\\mathcal{T}\\) a set of decision trees whose leaves are expressions in a weaker fragment (for example, polynomials). A lowering map \\(L: \\mathcal{E} \\to \\mathcal{T}\\) is correct when for all inputs \\(x\\) in the intended domain:\n\\[ [\\![e]\\!](x) = [\\![L(e)]\\!](x) \\]where \\([\\![\\cdot]\\!]\\) denotes denotational evaluation. Compiler IRs, autodiff traces, and symbolic simplifiers are engineering approximations to such maps.\nClosing position # Treating algebra and decision trees as enemies misreads computation. Algebra specifies; trees schedule and guard. Seeing both at once is a practical skill for anyone who writes numerical code, designs learning systems, or tries to prove programs correct.\nNotes and references # Demonstration essay for layout and technical features (math, code, Mermaid). For production research, replace schematic claims with citations to compiler and ML literature appropriate to the specific lowering you study.\n","date":"2 April 2026","externalUrl":null,"permalink":"/investigations/when-algebra-becomes-a-decision-tree/","section":"Investigations","summary":"Symbolic structure and branching control flow are two faces of the same computational coin. Seeing algebra as a tree of cases clarifies both classical algorithms and modern learning systems.","title":"When Algebra Becomes a Decision Tree","type":"investigations"},{"content":" Generative models routinely produce outputs that feel new: unfamiliar images, surprising sentences, code that solves a problem the author has not written before. Feeling new is not the same as being novel in a philosophically or scientifically useful sense. This essay separates kinds of novelty and asks which, if any, current systems can claim.\nWhy the question is slippery # “Genuine novelty” packs together several questions:\nIs the output statistically rare under the training distribution? Is it useful or valued by some community of practice? Does it introduce a new conceptual frame, not only a new instance? Does the system intend novelty, or only sample from a learned measure? Conflating these produces both hype and dismissiveness. A model that recombines motifs in a high-dimensional space can score well on (1) and sometimes (2) while failing (3) and (4).\nThe interesting claim is not that machines surprise us—they do—but that surprise is the same kind of thing as theoretical invention. That second claim needs work.\nCombinatorial novelty # Most generative success is combinatorial: sampling mixtures of patterns that were present, in fragments, in training data. That is not trivial. Chess and go programs, protein models, and large language models all exploit structure that human designers cannot enumerate by hand.\nA simple formal sketch: let \\(\\mathcal{X}\\) be a space of artefacts and \\(P_{\\mathrm{data}}\\) an empirical distribution over training examples. A generator \\(G\\) induces \\(P_G\\). One weak novelty score is:\n\\[ N_{\\mathrm{rare}}(x) = -\\log P_{\\mathrm{data}}(x) \\]for \\(x \\sim P_G\\), subject to a quality constraint \\(Q(x) \\ge \\tau\\). High \\(N_{\\mathrm{rare}}\\) with adequate quality is surprising recombination, not proof of a new theory of the domain.\nConceptual novelty # Conceptual novelty is stricter. It requires a change in the representational vocabulary—new primitives, new invariants, or a reorganisation of what counts as a problem. Historical examples often cited in this spirit (with the usual historiographic caveats) include non-Euclidean geometry or the idea of a programmable stored-program computer: not merely new instances, but new kinds.\nCurrent foundation models can discuss conceptual shifts and can help humans search for them. Whether they originate such shifts without a human steering loop remains an open empirical and definitional question. This essay does not pretend that the literature has settled it.\nA working taxonomy # Kind What changes Typical evidence Surface variation Local features Paraphrase, style transfer Combinatorial novelty Arrangement of known parts High rarity + quality filters Problem reformulation How the task is framed New benchmarks, new constraints Conceptual invention Primitives or invariants New theory, new mathematics Most product demos live in the first two rows. Research progress sometimes reaches the third. The fourth is rare for humans and not established as a routine property of present systems.\nImplications for evaluation # If we evaluate only with preference models and short-horizon human ratings, we will over-count surface novelty. Stronger tests might include:\nTransfer under distribution shift that punishes memorised templates Compression tests: does the artefact simplify a domain for later work? Human–AI process studies: who proposes the conceptual move, and when? def rarity_score(model_logprob_fn, x, quality, tau): \u0026#34;\u0026#34;\u0026#34;Illustrative only — not a production metric.\u0026#34;\u0026#34;\u0026#34; if quality(x) \u0026lt; tau: return None return -model_logprob_fn(x) Closing position # AI systems can create outputs that are new as instances and sometimes new as combinations. Claims of genuine conceptual invention should be treated as extraordinary and supported by domain-specific argument, not by awe at fluent generation. The useful research program is to sharpen the taxonomy and build evaluations that track which kind of novelty we actually got.1\nNotes and references # This piece is a layout demonstration for Noetium. Empirical claims about model capabilities should be checked against current literature; capability frontiers move quickly, and secondary summaries lag. For KaTeX notation reference, see the KaTeX documentation.\n“Genuine” here is a term of art in this essay’s taxonomy, not a metaphysical claim that novelty is an ontic primitive.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\n","date":"12 March 2026","externalUrl":null,"permalink":"/investigations/can-ai-create-genuine-novelty/","section":"Investigations","summary":"Novelty is not a single property. Distinguishing combinatorial surprise from conceptual invention changes what we can fairly claim about generative models.","title":"Can AI Create Genuine Novelty?","type":"investigations"},{"content":"","date":"12 March 2026","externalUrl":null,"permalink":"/tags/creativity/","section":"Tags","summary":"","title":"Creativity","type":"tags"},{"content":"","date":"12 March 2026","externalUrl":null,"permalink":"/tags/novelty/","section":"Tags","summary":"","title":"Novelty","type":"tags"},{"content":"","date":"12 March 2026","externalUrl":null,"permalink":"/tags/philosophy-of-mind/","section":"Tags","summary":"","title":"Philosophy of Mind","type":"tags"},{"content":"","date":"20 February 2026","externalUrl":null,"permalink":"/tags/automation/","section":"Tags","summary":"","title":"Automation","type":"tags"},{"content":"","date":"20 February 2026","externalUrl":null,"permalink":"/tags/biology/","section":"Tags","summary":"","title":"Biology","type":"tags"},{"content":"","date":"20 February 2026","externalUrl":null,"permalink":"/tags/infrastructure/","section":"Tags","summary":"","title":"Infrastructure","type":"tags"},{"content":"","date":"20 February 2026","externalUrl":null,"permalink":"/tags/labs/","section":"Tags","summary":"","title":"Labs","type":"tags"},{"content":"","date":"20 February 2026","externalUrl":null,"permalink":"/tags/robotics/","section":"Tags","summary":"","title":"Robotics","type":"tags"},{"content":"Laboratory automation is real: liquid handlers, plate readers, high-throughput sequencing, and cloud labs already change who can run which protocols. The question is not whether any steps can be automated, but whether biology’s wet core will become as programmable as software. This essay argues for structural resistance—not impossibility of progress, but asymmetric difficulty.\nWhat automation has already won # Where inputs are standardised and readouts are digital, automation thrives:\nSequencing pipelines with stable library preps Plate-based assays with fixed volumes and timings Imaging with controlled optics and barcodes These domains still need humans for exception handling, but the happy path is machine-runnable. That success tempts a generalisation: the rest of the lab is only a few robots away.\nThree sources of resistance # 1. Physical contingency # Biological materials vary. A “standard” cell line drifts; a reagent lot changes; humidity alters evaporation in a 384-well plate. Software failures are reproducible from logs; many wet failures are not, or not cheaply.\nSoftware analogue Wet-lab reality Deterministic CPU Variable living systems Bit-identical copies Batch effects Free rollback Consumed samples Stack traces Ambiguous phenotypes 2. Tacit skill and perception # Pipetting technique, colony picking, “how viscous does this look?”, and knowing when a gel “looks wrong” remain partly embodied. Vision systems and force sensors improve, but the long tail of lab craft is expensive to encode.\n3. Open-ended protocols # Discovery science changes the protocol mid-course. Automation prefers closed workflows. When the hypothesis shifts, the robot’s program is suddenly the wrong abstraction. Cloud labs help with scheduled, pre-specified work; they help less when the experiment is a conversation with nature.\nFull automation is easiest when you already know the answer’s shape. Biology often does not.\nPartial automation is still transformative # Resistance is not a counsel of despair. High leverage exists in:\nInstrument APIs and data models that make partial pipelines composable Exception-aware orchestration—humans as supervisors of fleets, not pipetters of every well Standardisation where science allows without pretending all biology is ELISA The economic picture may resemble aviation more than pure software: high automation with mandatory human judgment at critical junctures.\nA caution on forecasts # Claims that “self-driving labs will close the loop on all of biology” should be treated as speculation unless scoped to a protocol class with clear success metrics. Historical automation in chemistry and manufacturing offers analogies, but cells are not parts bins.\nClosing position # Wet-lab biology will keep absorbing automation at the edges and in standardised cores. The centre of gravity—messy samples, shifting questions, living variance—will continue to demand human responsibility. Designing tools that respect that fact is more useful than promising a fully unmanned discovery engine.\nNotes and references # Layout demonstration. Empirical rates of automation adoption vary by subfield; readers should consult domain surveys rather than treat this taxonomy as data.\n","date":"20 February 2026","externalUrl":null,"permalink":"/investigations/why-wet-lab-biology-may-resist-automation/","section":"Investigations","summary":"Software ate many workflows; cells and reagents do not compile. Physical contingency, tacit skill, and open-ended protocols set hard limits on end-to-end automation.","title":"Why Wet-Lab Biology May Resist Automation","type":"investigations"},{"content":"","date":"15 January 2026","externalUrl":null,"permalink":"/tags/drm/","section":"Tags","summary":"","title":"DRM","type":"tags"},{"content":"","date":"15 January 2026","externalUrl":null,"permalink":"/tags/obfuscation/","section":"Tags","summary":"","title":"Obfuscation","type":"tags"},{"content":"","date":"15 January 2026","externalUrl":null,"permalink":"/tags/security/","section":"Tags","summary":"","title":"Security","type":"tags"},{"content":"","date":"15 January 2026","externalUrl":null,"permalink":"/tags/software/","section":"Tags","summary":"","title":"Software","type":"tags"},{"content":"","date":"15 January 2026","externalUrl":null,"permalink":"/tags/threat-models/","section":"Tags","summary":"","title":"Threat Models","type":"tags"},{"content":" Software vendors ship “protected” binaries: packers, virtualisation obfuscators, anti-debug checks, white-box crypto sketches. Marketing language sometimes implies that the program’s secrets are safe. Under standard threat models, that implication is false. This essay separates cost imposition from confidentiality.\nThreat model first # Define the adversary before the tool:\nRemote attacker — no code execution on the client; network-facing surface only Malicious client — full control of the machine that runs the binary Insiders / supply chain — access before shipping Executable protection almost always targets (2). For (2), the attacker has CPU, memory, and patience. Anything the CPU must decrypt to execute, the attacker can eventually observe.\nWhat protection can do # Honest goals include:\nDelay reverse engineering enough to outlast a business need (short-lived keys, seasonal content) Deter casual copying and script-kiddie patches Detect some tampering and refuse to run (integrity checks) Bind execution to environment signals (licenses, secure elements)—with caveats These are economic and operational goods. They are real. They are not mathematical secrecy for general program logic.\nWhat protection cannot do # If \\(P\\) is a program that the client must run, and \\(S\\) is a secret embedded such that correct execution reveals \\(S\\)’s functional effect, then a sufficiently patient adversary with a debugger can recover behaviour equivalent to \\(S\\). Formally, white-box attackers model the implementation as an oracle they can inspect.\nObfuscation may hide how; it does not remove that the behaviour is available.\nClaim Status under malicious client “Logic cannot be recovered” Generally false “Recovery is expensive” Often true “Secrets never in RAM” False if CPU uses them “DRM is cryptographically solved” False in general Design implications # Put real secrets on servers or hardware you control, not only in client binaries. Use protection as friction, not as a root of trust. Measure against skilled reverse engineers, not only against strings. Plan for failure: assume client-side checks will be bypassed. Good: server holds master key; client holds capability token Bad: master key encrypted in .rodata with fixed XOR \u0026#34;for safety\u0026#34; Closing position # Executable protection is a legitimate engineering tool for raising analysis cost. It becomes harmful when it substitutes for sound architecture. The sentence to keep is simple: you cannot hide from the computer that must run your code.\nNotes and references # Demonstration essay. Specific commercial packers and attacks change; treat product claims skeptically and consult current reverse-engineering literature for concrete techniques.\n","date":"15 January 2026","externalUrl":null,"permalink":"/investigations/what-executable-protection-can-and-cannot-protect/","section":"Investigations","summary":"Obfuscation, packing, and anti-tamper raise the cost of analysis. They do not create cryptographic confidentiality of logic on a machine the adversary runs.","title":"What Executable Protection Can—and Cannot—Protect","type":"investigations"},{"content":"Noetium publishes original essays developed through sustained human–AI inquiry. The brand mark and tagline—A new element of thought—name an editorial ambition: treat ideas as materials that can be refined, tested, and recast into durable prose.\nThis is a static editorial publication, not a chatbot, not a transcript archive, and not a scientific institution.\nEditorial method # Each investigation roughly follows the same path:\nA human begins with a question, intuition, or objection—something that resists a neat answer. The idea is tested through sustained interaction with AI systems: drafting, counter-argument, re-framing, calculation, and search for counterexamples. Weak explanations are challenged. Comfortable narratives are pushed until they break or hold. Evidence is examined. Sources, calculations, and known results are checked where possible. Speculation is labeled as such. The surviving result is rewritten as a coherent standalone essay—argument first, process second. Raw conversations are not published by default. The product is the finished intellectual result, not the chat log. AI output is never treated as automatically correct. The human remains responsible for judgment, structure, and final claims.\nWhat we publish # Subjects include artificial intelligence, philosophy, mathematics, probability, markets, programming, infrastructure, biology, music, and society. The unifying trait is not topic but method: long-form thought that has been pressure-tested.\nEssays may include mathematics, code, diagrams, tables, and footnotes when those tools serve clarity.\nStandards of claim # Noetium is not a peer-reviewed journal and does not claim institutional authority. We distinguish, as carefully as we can, among:\nEstablished evidence — results with clear sources or well-known status in a field Inference — conclusions that follow from premises we state Speculation — open hypotheses, thought experiments, and exploratory models Where a factual claim is uncertain, the essay should say so rather than invent a citation.\nWhat this site is not # Not a chat product or AI assistant interface Not a dump of unedited model outputs Not a substitute for primary literature or professional advice Not a social network Contact and reuse # For corrections, licensing, or correspondence, use the channels listed when available on the live site. Unless noted otherwise, site design and configuration are project-owned; essay rights remain with their authors.\nEssays are developed through human–AI inquiry and edited for accuracy, clarity, and coherence.\n","externalUrl":null,"permalink":"/about/","section":"Noetium","summary":"Noetium publishes original essays developed through sustained human–AI inquiry. The brand mark and tagline—A new element of thought—name an editorial ambition: treat ideas as materials that can be refined, tested, and recast into durable prose.\n","title":"About","type":"page"},{"content":"","externalUrl":null,"permalink":"/archives/","section":"Noetium","summary":"","title":"Archives","type":"page"},{"content":"Essays on programming languages, systems design, and the limits of software protection.\n","externalUrl":null,"permalink":"/topics/code-and-systems/","section":"Topics","summary":"Essays on programming languages, systems design, and the limits of software protection.\n","title":"Code \u0026 Systems","type":"topics"},{"content":"Polished essays derived from long investigations. Each piece stands alone; raw conversations are not published by default.\n","externalUrl":null,"permalink":"/investigations/","section":"Investigations","summary":"Polished essays derived from long investigations. Each piece stands alone; raw conversations are not published by default.\n","title":"Investigations","type":"investigations"},{"content":"Essays on artificial intelligence, cognition, and the boundary between recombination and invention.\n","externalUrl":null,"permalink":"/topics/mind-and-machine/","section":"Topics","summary":"Essays on artificial intelligence, cognition, and the boundary between recombination and invention.\n","title":"Mind \u0026 Machine","type":"topics"},{"content":"","externalUrl":null,"permalink":"/","section":"Noetium","summary":"","title":"Noetium","type":"page"},{"content":"Noetium is a static website. By default it does not load third-party analytics, advertising, or comment widgets.\nWhat we collect # Server or host logs may be retained by the hosting provider (for example request paths, IP addresses, user agents). That is outside this site’s application code and depends on where the site is deployed. No first-party accounts are required to read essays. Newsletter signup is a placeholder until a provider is configured. When it is connected, this page will name the provider and describe what is stored. Cookies and tracking # Until analytics or similar services are explicitly enabled, Noetium does not set tracking cookies and does not show a cookie consent banner.\nThe theme may use localStorage in the browser to remember appearance preference (light/dark). That preference stays on your device.\nThird parties # If the live deployment later enables search analytics, a newsletter service, or similar tools, they will be documented here before use. External links in essays lead to other sites with their own policies.\nContact # For privacy questions related to this publication, use the contact method listed on the About page when available.\n","externalUrl":null,"permalink":"/privacy/","section":"Noetium","summary":"Noetium is a static website. By default it does not load third-party analytics, advertising, or comment widgets.\nWhat we collect # Server or host logs may be retained by the hosting provider (for example request paths, IP addresses, user agents). That is outside this site’s application code and depends on where the site is deployed. No first-party accounts are required to read essays. Newsletter signup is a placeholder until a provider is configured. When it is connected, this page will name the provider and describe what is stored. Cookies and tracking # Until analytics or similar services are explicitly enabled, Noetium does not set tracking cookies and does not show a cookie consent banner.\n","title":"Privacy","type":"page"},{"content":"Essays on probability, decision theory, and market behaviour.\n","externalUrl":null,"permalink":"/topics/probability-and-markets/","section":"Topics","summary":"Essays on probability, decision theory, and market behaviour.\n","title":"Probability \u0026 Markets","type":"topics"},{"content":"Essays where scientific practice meets social and technical constraint.\n","externalUrl":null,"permalink":"/topics/science-and-society/","section":"Topics","summary":"Essays where scientific practice meets social and technical constraint.\n","title":"Science \u0026 Society","type":"topics"},{"content":"","externalUrl":null,"permalink":"/series/","section":"Series","summary":"","title":"Series","type":"series"},{"content":"Essays on sound, media, and cultural form. New work will appear here as it is published.\n","externalUrl":null,"permalink":"/topics/sound-and-culture/","section":"Topics","summary":"Essays on sound, media, and cultural form. New work will appear here as it is published.\n","title":"Sound \u0026 Culture","type":"topics"},{"content":"Investigations are grouped by topic. Topics are editorial taxonomies, not peer-reviewed subject classifications.\nTopic Focus Mind \u0026amp; Machine AI, cognition, novelty, multi-agent systems Probability \u0026amp; Markets Uncertainty, decision, markets Code \u0026amp; Systems Programming, infrastructure, protection Science \u0026amp; Society Biology, automation, institutions Sound \u0026amp; Culture Music, media, cultural form Use the topic links in the site header, on the homepage, or on individual essays to explore related work.\n","externalUrl":null,"permalink":"/topics/","section":"Topics","summary":"Investigations are grouped by topic. Topics are editorial taxonomies, not peer-reviewed subject classifications.\nTopic Focus Mind \u0026 Machine AI, cognition, novelty, multi-agent systems Probability \u0026 Markets Uncertainty, decision, markets Code \u0026 Systems Programming, infrastructure, protection Science \u0026 Society Biology, automation, institutions Sound \u0026 Culture Music, media, cultural form Use the topic links in the site header, on the homepage, or on individual essays to explore related work.\n","title":"Topics","type":"topics"}]