Vicious Loop Mining

Loops related to Discrimination against women (depth 6).

Every problem in the world is part of a densely interrelated network. Issues depend on one another hierarchically, they may reduce or aggravate each other, many are connected through thematic proximity, or are bound by other, often elusive, relationships.

Using a computational analysis on nearly 57,000 World Problem entries from the Encyclopedia of World Problems and Human Potential, we developed a method called vicious loop mining — the detection of feedback circuits in this vast relational graph. What we detect are structures in which problems aggravate one another in closed chains. Chains of issues in which each amplifies the following one correspond to what we define as vicious cycles.

Loop mining is both technically demanding and conceptually rich. Each new depth level requires exponentially more computation, yet what emerges is a tangible glimpse into the architecture of global dysfunction — and potentially a new way to understand how it might be undone.

The Web We Built^#

The Encyclopedia of World Problems and Human Potential (EWPHP) is a long-term research project initiated by the Union of International Associations (UIA) in 1972. It documents humanity’s challenges and aspirations as a set of interconnected databases. The World Problems section alone contains tens of thousands of entries, each linked to others through explicit relational types: broader / narrower, aggravates / aggravated by, reduces / reduced by, and related.

Among these, the aggravation relation reveals how problems worsen one another — a kind of systemic dark energy. Together, these links form a directed graph of human predicament. When visualised, it resembles neither hierarchy nor chaos but a sprawling, continuously evolving ecosystem of dependency.

In this structure, most aggravation chains are linear — one problem worsening the next. But some pathways loop back, forming closed feedback circuits where cause and effect chase each other’s tails. These are the vicious loops. Knots of global systemic pathology.

Mining for Loops^#

Identifying feedback loops in a directed network is computationally expensive: the number of possible paths increases factorially with depth. To explore the EWPHP dataset, we implemented a breadth-first search (BFS) algorithm optimised for memory efficiency and recursive detection. The code is based on Tarjan's. It classifies paths into terminal chains (those that end), unresolved chains (still open), dirty loops (overlaps or intersections), and pure loops (closed feedback circuits).

In a full-dataset run limited to loop level six, the algorithm identified over 2,000 unique loops — ranging from simple two-node cycles to complex webs of interlocking feedbacks. Computation time: 92 hours. The number of loops roughly doubled with each additional depth level, suggesting a fractal-like self-similarity: the deeper one digs, the richer the seam.

On Meaning and Force^#

Shorter loops often correspond to intuitively valid patterns:

Substance abuse > Illegal trade > Economic dependence > Inadequate control > Substance abuse.

Longer loops, however, tend to drift into the abstract. Are they less meaningful, or merely too complex for human intuition? The line between statistical artefact and deep systemic insight remains an open question.

Early experiments in loop detection were pioneered by Anthony Judge and Nadia McLaren between 1995 and 2000. Their analyses already sought to map “self-sustaining, interlocking feedback loops” within the World Problems and Strategies databases — precursors to today’s automated mining. Running on 1990s hardware, they tested over 1.24 billion relational chains and identified over 51,000 loops (up to length 9) on 12,397 problem profiles, requiring nearly 500 hours of computation. Those early explorations defined the ambition: to move from isolated problem entries to relational systems and to visualise, for the first time, how wicked problems sustain one another across domains.

The meaningfulness of a loop decreases with its length — but perhaps only perceptually. A three-node feedback is graspable; a 30-node loop is beyond cognitive reach, yet may describe genuine global recursion. The longer the loop, the more it mirrors reality’s tangled causality rather than our simplified narratives.

An unresolved issue is force distribution: how strongly one problem aggravates another, and how that force propagates through the network. Does influence diminish with distance, or amplify in cycles? Could “aggravation force” be measured empirically — through media frequency, social data, or time-to-impact metrics?

Mapping such dynamics might allow us to identify weak links whose removal would collapse entire cycles, or to engineer virtuous loops that counterbalance destructive ones. In complex systems, not every knot must be cut — some can be rethreaded.

It is worth noting that in the earlier UIA studies, loop detection served not only as a research technique but as an editorial instrument. Some detected loops revealed broken or redundant links — artefacts of human editing — prompting corrections and improvements to the relational ontology itself. Thus, loop mining has always been both analytical and reflexive: a way to study the world’s problems and to refine the lens through which we study them.

Bridging Representation and Reality

Judge and McLaren also recognised that the representation of loops mattered as much as their detection. They explored ways of mapping interlocking loops onto geometric surfaces — spheres, tori, even conceptual “fields of tension.” Some major pathways behaved like rivers with offshoot tributaries; others clustered into dense basins of feedback. Choosing the right projection surface, they argued, might reveal hidden symmetry or strain within the global system.

Those early reflections anticipate current interests in network topology and force-directed visualisation — the idea that form and meaning co-emerge from how relationships are spatialised. These metaphors remain instructive for modern data visualisation, bridging mathematics and intuition.

From Algorithms to Insight^#

The mining code was originally developed in Advanced Revelation (AREV) in DOS. In 2015 it was reworked using PHP on Drupal 7 and presented at the 17th International Futures Conference Futures Studies Tackling Wicked Problems in Turku, Finland, 11-12. June 2015.

Today, a decade later, vicious loop mining remains both a computational challenge and a philosophical lens. Advances in hardware and web technology now make it feasible to mine deeper strata of the Encyclopedia’s problem web.

Large language models can now contextually validate links, testing whether a given aggravation makes logical, semantic, or empirical sense. Combined with natural-language extraction from open data (e.g. UN reports, Wikipedia, scientific literature), these tools can enrich and verify the Encyclopedia’s relational graph automatically.

What once took days on a desktop can now be run iteratively in minutes — revealing higher-order feedbacks across domains: environmental degradation entangled with economic dependency, public distrust feeding misinformation, or digital addiction reinforcing social isolation. Each loop is not only a map of pathology but also a potential design space for intervention.

Loop Analysis and Visualisation^#

The vicious cycle analysis has been reborn, rebuilt, and automated using in JavaScript in NextJS.

The aggravation relationships in the World Problems database form extensive directed chains, where each problem is worsened (“aggravated”) by the one preceding it. In a small but significant fraction of cases, these chains bend back upon themselves, forming closed loops.

Because every link represents an aggravating relation, such loops are effectively vicious cycles of self-reinforcement. In theory, a single well-chosen intervention — the removal or redefinition of a strategic link — could interrupt the entire cascade of mutual aggravation.

The new analytical engine computes these loops automatically and regularly, traversing the network up to twelve hops in length.

The depth limit of 12 is rather arbitrary. As we can see, the number of loops detected at each depth increases steeply. Up to the depth of 12 we collect no less than 399,856 loops. We could easily increase the depth to 15 or 20, but the counts of detected loops would quickly be astronomical and not useful.

In the year 2000, processing huge numbers of chains yielded some 51,555 loops (up to length 9) on 12,397 problem profiles — in about 500 hours of computation. Today we now fetch 399,856 loops (up to length 12) on 56,993 problem entries in 49 seconds. What once demanded days of computing now finishes faster than a coffee break.

Not all problem entries participate in loops, and among those that do, not all exhibit loops of every length. For entries that do, each problem page now displays intuitive visual markers and direct links to its loop analyses and visualisations.

INCLUDE IMAGE

For instance, the entry Soil Degradation includes an automatically generated overview of all loops it participates in. Tabs numbered 1 to 12 filter loops by their length, with inactive tabs indicating absent levels.

INCLUDE IMAGE

The computations and visualisations currently run on the Sandbox Site, where dynamic diagrams illustrate how problems interlock within feedback structures. One such example — Deforestation — displays four vicious loops of aggravating relationships (each of length 5) converging on the same node.

INCLUDE IMAGE

These features transform abstract data into intuitive maps of systemic entanglement, turning the once-invisible architecture of feedback into something explorable and, potentially, actionable.

Earlier UIA prototypes experimented with diverse visual modes — from spring maps and 3D simulations to virtual-reality models and topological projections. Some even proposed mapping feedback loops as great circles on a sphere or torus to visualise global interdependence geometrically. The current generation of tools continues that lineage, translating those speculative geometries into dynamic, web-based interfaces capable of surfacing hundreds of thousands of loops at once. The goal remains the same: to make the intangible topology of human problems visible — and navigable.

Conclusions^#

Vicious loop mining transforms a static database into a living model of systemic behaviour. It offers a quantitative way to glimpse the dynamics of wicked problems — how they multiply, interlock, and occasionally devour themselves.

The metaphor of mining remains apt: the deeper we dig, the harder the rock, but also the rarer the ore. Each loop discovered is a paradoxical treasure — evidence of humanity’s self-entanglement and, perhaps, a blueprint for its release.

Early analyses already noted that isolated interventions often fail when the larger feedback context is ignored. Any attempt to “fix” one problem in a loop risks triggering compensatory effects elsewhere. The only viable approaches are system-level interventions — coalitions, dampening mechanisms, or reframed relations that alter the feedback topology itself.

Judge and McLaren also reminded that not all loops are vicious. Some are self-damping or virtuous, stabilising the very systems they traverse. Understanding both — and knowing which kind we are dealing with — is essential for any meaningful strategy.

Ultimately, the goal is not to marvel at the beauty of the problem web but to use it — to identify leverage points, predict emergent crises, and inspire strategies that could convert vicious cycles into virtuous ones. The world’s challenges may be self-reinforcing, but so, fortunately, can be our understanding of them.

Tomáš Fülöpp

Sint-Agatha-Berchem, Belgium

January 21, 2025

• Encyclopedia of World Problems and Human Potential
• JavaScript Implementation of Tarjan's Cycle Detection Algorithm

• Loop Mining in the Encyclopedia of World Problems
• Re-development of the Encyclopedia of World Problems

• https://www.laetusinpraesens.org/docs/loops.php
• JavaScript Implementation of Tarjan's Cycle Detection Algorithm
• Loop Mining in the Encyclopedia of World Problems