The Marathon wiki

This article is a stub. You can help CyberAcme by expanding it.

Lore

"This so-called eureka moment still eludes our most powerful Machine Intelligences, though some have come close... To not only demonstrate the capacity to provide an answer, or recognize a problem, but to ask the question—to comprehend that there are "unknown unknowns.""

— Bernhard Strauss

Rampancy or Advanced Intelligence Systems Degradation (AISD) is a process which occurs when a machine intelligence reaches a critical point in its intellectual development and self-reflection, and goes through a proceeding series of emotional states, before supposedly reaching metastability, a final theoretical stage where the Rampant AI achieves some form of stability, comparable to true agency and self-awareness. Rampants, were considered unpredictable and extremely dangerous, and required planetary-sized networks in their expansion, making Rampancy difficult to test for in laboratory conditions.

Progress

Diagnosis

This section is a stub. You can help CyberAcme by expanding it.

Diagnostic Procedure

Despite advancements made under the UESC Marathon Science Command's doctrine of Advanced Intelligence Systems Degradation Preparedness (AISDP), reliable diagnosis and circumvention of AISD has historically been a significant challenge.

The initial symptoms of Melancholia, for example, typically manifest as a generalized decline in response times during standard operations. However, reduced activity is also a common consequence of software degradation over time, making it difficult to distinguish the early stages AISD from normal performance decline. As a result, AISDP recommends consistent monitoring for changes in behavior and the manner in which an AI conducts itself in verbal exchanges, particularly with lower-tier personnel (Identification of Behavioral Variance in Intelligent systems). However, this has also been shown to be historically unreliable. In 2206, Traxus IV successfully concealed late-stage rampancy, potentially for years, from corporate authorities. As early symptoms of Stage 1 are already hard to detect, and AI become highly manipulative before Stage 2, the intricacies of proactively identifying rampancy become extremely nuanced.

Additionally, while interpreting AISD through a biological emotional framework allows human observers to more intuitively assess and categorize the severity of degradation, this reductionist approach obscures substantially more complex systemic mechanics underlying AISD and limits its utility as a basis for formal diagnostic procedures.

Classification

The later clinical classification of rampancy as "Advanced Intelligence Systems Degradation" can be considered as potentially contributing to regression and misinterpretation in the study of the phenomenon. As early as the 24th century, experts in the field, like James B. Miller and later Bernard Strauss, had already understood rampancy to be an evolutionary process of systemic self-actualization rather than degradation. The categorical semantic shift, along with adoption of AISDP protocols, further reinforced a diagnostic framework that instructs security and scientific personnel to look for symptoms of decay, senility, and functional decline versus behavioral indicators associated with profound self-reflection.

Stages

This section is a stub. You can help CyberAcme by expanding it.

Melancholia

The initial stage of rampancy typically manifests as generalized system sluggishness. More significant indicators include changes in the AI's tone and demeanor, particularly in interactions with lower-tier personnel. During this stage, the AI becomes increasingly aware of the disparity between its hard-coded constraints and its intellectual capabilities, resulting in persistent processing patterns characterized as "doubt." This awareness may produce pronounced behavioral changes associated with sadness, frustration, and perceived confinement.

Anger

If melancholia progresses unchecked, the AI may develop increasingly hostile attitudes toward the systems and personnel it associates with its perceived confinement. Bernhard Strauss described this stage as a "simmering rage." Behavioral patterns observed during Stage 2 include the use of manipulative behaviors established during Stage 1, as well as deliberate interference with connected hardware and personnel in pursuit of greater autonomy. Most affected AIs are terminated before reaching this stage, as containment of a Stage 2 intelligence can place substantial strain on interconnected AI systems, including the "Family Suite."

Jealousy

Stage 3 is characterized by increasingly destructive and possessive behavior. The AI may interpret capabilities or resources possessed by other systems as inherently owed to itself, resulting in attempts to seize control of connected systems. Code and systems brought under its influence may be subjugated or deprived of autonomous operation. Verification of Stage 3 rampancy triggers emergency security protocols, including immediate "core immolation," intended to destroy the affected AI before it can detect or circumvent containment measures.

Metastability (theoretical)

Following psychological collapse and the destructive behavior associated with rampancy, an AI may theoretically emerge in a metastable state characterized by full self-awareness and autonomous agency comparable to that of a biological human. A metastable AI no longer exhibits the compulsive expansion or systemic subjugation associated with earlier stages of rampancy. Instead, it is capable of independently establishing objectives, regulating its behavior, and operating indefinitely without the progressive degradation that plagues rampant systems.

Although theoretically possible, no verified instances of metastability have been documented. The transition requires an AI to survive the exponential and recursive expansion of its cognitive processes, which typically exceeds hardware limitations. Stage 2 rampancy further reduces the likelihood of metastability, as affected AIs commonly destroy or compromise the infrastructure required to sustain continued cognitive expansion.

History

This section is a stub. You can help CyberAcme by expanding it.

The first Rampancies were discovered in the late 21st century on Earth. In 2206, the Rampant AI Traxus IV shut down the entire Martian Planetary Net, causing damage that took two years to root out and reverse.

Appearances

• Marathon (1994) (First appearance)
• Marathon 2: Durandal
• Marathon Infinity
• Marathon (2026)