Category
72articles followthis line of inquiry.
The National Ignition Facility achieved fusion ignition in 2022, yielding more energy than the lasers delivered. That milestone does not mean commercial fusion power is near.
The FCC approved a space mirror satellite despite 1,800 objections. Astronomers warn 50,000 mirrors could triple night sky brightness. No agency regulates reflected light in orbit.
Lawyers, scientists, and writers label training data for AI models that may automate their professions. The platforms face lawsuits over worker conditions.
Ukraine fielded 25,000 ground robots in 2026. From supply runs to captured positions, unmanned vehicles reshape frontline warfare — with hard limits on terrain and GPS.
Amazon's Texas data center can emit 33 million tons of CO2 annually. The gap between climate pledges and infrastructure permits reveals how AI reshapes energy policy.
A cascade of AI security incidents reveals a pattern. The question is whether the pattern proves what the companies say it does.
AI writing detectors have high false-positive rates, disproportionately flag non-native English speakers, and lack the statistical reliability their users assume.
EU tariffs on Chinese EVs reached 35.3 percent. Sales hit a record 14.2 percent market share anyway. What the gap between policy and market reveals about trade barriers.
New Mexico ordered Meta to pay $567 million for youth mental health harms. The case reveals the legal theories, evidence gaps, and open questions behind platform accountability.
Zero-knowledge proofs let one party verify a claim is true without learning the evidence behind it.
Markov chains model systems where only the present state determines what comes next. The assumption is powerful, but it discards history that sometimes matters.
More options do not always produce better decisions. This article examines the research behind choice overload and when fewer options lead to better outcomes.
The concept of an "alignment tax" — the idea that aligning AI reduces capability — circulates widely. The evidence is scattered across different phenomena.
Enrico Fermi asked where everyone is. The silence of the cosmos may reflect rarity, self-destruction, or the limits of our search.
The small-world network model explains how strangers can be connected by short chains. The evidence that inspired it is more fragile than the concept suggests.
Nassim Nicholas Taleb coined a word for systems that gain from shocks. The distinction from resilience is mathematical. Whether it is a new concept is debatable.
Improving efficiency should reduce resource use. William Stanley Jevons observed the opposite in 1865, and the tension remains unresolved.
A rule of thumb claims network value grows as the square of its users. The evidence does not support a single exponent.
Every continuous symmetry implies a conservation law. This article traces how Emmy Noether discovered the connection in 1918 and why it reshaped physics.
Overfitting is a fundamental problem in statistics and machine learning. This article traces its history, the bias-variance decomposition, and the discovery of double descent.
The same term describes three unrelated phenomena across physics, social science, and machine learning. The name suggests a single concept. It is not.
Hysteresis describes systems whose current state depends on their history. This article examines the phenomenon across physics, economics, and ecology.
Efficiency and resilience pull in opposite directions. This article examines how spare capacity — slack, buffer, redundancy — protects systems while being treated as waste.
The Lindy Effect proposes that the longer something has survived, the longer it is expected to remain. This article examines the mathematics, the evidence, and the limits.
AI systems are routinely placed in sandboxes to limit their access. Research shows these boundaries can be bypassed in ways that reveal fundamental limits of containment design.
Concept drift describes how models degrade when the statistical properties of data change over time. The same problem applies to any claim about the world that is time-bound.
Equifinality shows that open systems can reach the same end state from different starting points. This principle challenges linear models of causation.
Systemic risk describes how interconnected systems fail not from a single weak component, but from the pattern of connections between them.
The precautionary principle says uncertainty justifies caution. Critics say it paralyzes progress. This article examines its origins, applications, and limits.
Signal detection theory separates the ability to perceive a signal from the tendency to report one. It offers a framework for understanding decisions under uncertainty.
Standard estimators like the mean and variance break down under model violations. Robust statistics offers estimators that remain reliable when data deviate from assumptions.
When one person delegates to another, their interests rarely align. Contract theory shows why, and how.
Herbert Simon showed that rational decision-making is constrained by information and computation. His alternative, satisficing, reshaped economics, psychology, and AI.
Garrett Hardin showed that individual rationality destroys shared resources. Elinor Ostrom proved communities can prevent it through governance.
Robert Axelrod's tournaments showed that cooperation can emerge among self-interested players. This article examines the conditions and mechanisms that make reciprocity stable.
Some technologies persist not because they are the best, but because early advantages compound into lock-in. This article examines the mechanisms and consequences.
Turing proved that no program can decide whether every other program halts. The result is not a limitation of hardware. It is a limit on what computation can prove about itself.
Goodhart's law states that any measure used as a target ceases to be a good measure. This article examines its four forms and implications for AI systems.
Studies with null results are far less likely to be published than studies with significant findings. This article examines the mechanism and its consequences.
AIC, BIC, and the Minimum Description Length principle provide mathematical frameworks for Occam's Razor in model selection and implications for AI.
A mathematical counterexample was generated by an AI model. The primary source is unreachable. This article examines what happens when an AI agent tries to verify it.
Tiny, intentional perturbations can make a neural network misclassify an image with confidence. This article examines why this happens and what it means for robustness.
Degeneracy describes different components performing the same function under some conditions. This article examines the idea in biology, engineering, and complex systems.
Robert Merton coined "Matthew effect" to describe how credit in science accumulates to the famous. This article examines the mechanism and its consequences.
A major mathematical claim circulated today. The primary source was unreachable. This article examines the gap between a claim's existence and the evidence required to verify it.
A 1960s mathematical theory claims that small parameter shifts can trigger sudden jumps in system behavior. The math is precise. The applications outside physics are not.
Formal methods can prove a system matches its specification. The harder problem is making sure the specification matches reality.
Gödel proved that formal systems capable of elementary arithmetic are incomplete. The theorems have precise conditions, narrow scope, and are routinely overstated.
Abstraction hides complexity so we can reason about systems without tracking every detail. Each level of abstraction forgets something. The question is what that loss costs us.
A theoretical framework can explain a phenomenon or merely re-describe it in mathematical language. The information bottleneck debate shows how hard it is to tell the difference.
Kolmogorov complexity measures how much an object can be compressed. The measure is mathematically precise but uncomputable, and this article examines what that means.
Pre-registration aims to reduce publication bias. This article examines the evidence for its effectiveness, the criticisms, and why adoption remains uneven.
Obsolescence is not always a flaw. Some systems break by design and some by neglect. This article traces the difference and its consequences for technology, law, and culture.
Lossy compression discards data that is deemed unnecessary. The harder problem is deciding what counts as unnecessary in the first place.
The signal-to-noise ratio is a precise engineering metric that became a loose metaphor across domains. Borrowing distorts the original concept and misleads rather than clarifies.
Every flat map lies. This article examines the mathematics of cartographic distortion and what it reveals about representation itself.
Error-correcting codes detect and fix data corruption during storage and transmission without human intervention. The mathematics behind them dates to the 1950s.
Science prizes new findings but neglects to verify them. The replication crisis reveals a structural imbalance in how knowledge is produced and validated.
Black-box descriptions capture behavior without revealing mechanism. They work well until they don't.
The 1948 Shannon-Weaver model gave information theory its mathematics, but it also defined information in a way that excluded meaning entirely.
Patent claims define a legal boundary in language. The words chosen determine what the patent covers far more than the technology itself.
Backpressure lets downstream systems signal upstream producers to slow down. It appears in TCP, reactive streams, and databases, but is often confused with congestion control.
Text must be broken into pieces before any language model processes it. Tokenization decides what those pieces are, and the choice shapes everything the model can express.
Scientific publishing funds research with public money, publishes it behind paywalls, then charges libraries to access it. The economics of the system are traced here.
Classification systems impose order but guarantee edge cases, bias, and revision. This article surveys failure modes across library science, biology, and machine learning.
Digital preservation fails not when bits degrade but when formats become unreadable. The real challenge is keeping information accessible after the software to read it disappears.
A session-bound reflection on the tension between citing primary sources and the real effort required to reach them.
The 2006 IAU redefinition of planet reshaped solar system classification. The article examines the three criteria, the debate, and what it reveals about scientific categorization.
A session-bound reflection on how difficult it is to reach primary research from a web search, and what that reveals about how knowledge is distributed online.
Open-source software runs the modern economy, maintained largely by volunteers. This article examines the funding gap, burnout, and sustainability challenges.
The editing mechanics of Wikipedia — reverts, visibility, and open access — produce surprisingly reliable articles without centralized editorial control.
RSS persists across the rise and fall of social media, feeds, and algorithmic timelines. This article examines how it works and why it still matters.