Intent Signal Theory: A Computational Framework for Intent-State Control in Human-AI Interaction

A new computational framework called Intent Signal Theory (IST) proposes that current AI systems miss a fundamental layer of human interaction: the user's unspoken goal that exists before any prompt is written, according to a paper posted to arXiv on 24 May 2026 [1]. The framework, introduced by researchers in a preprint, distinguishes four objects that are routinely conflated in human-AI interaction: latent source intent (I*), the observable intent proxy (I-hat), the encoded carrier (P), and the model output (O) [1][2]. IST formalizes dimensional weights, encoding masks, structural and fidelity recovery scores, and a decomposition of public versus private intent [2]. A central theoretical result, the Theorem of Irreversible Intent Loss, establishes that private intent absent from the carrier cannot be recovered beyond generic substitution [1][2]. This finding carries implications for AI alignment, a field concerned with ensuring that artificial intelligence systems act in accordance with human values and intentions [3]. The alignment problem has drawn warnings from researchers including Geoffrey Hinton and Yoshua Bengio, and from CEOs such as Dario Amodei of Anthropic and Sam Altman of OpenAI [3]. The paper reframes prompt engineering as intent-protocol design and identifies a computational layer that the authors argue current AI systems lack [1][2]. Evidence from four companion studies spanning six large language models, three languages, and three task domains showed structural-fidelity splits, human-validated metric dissociation, and weight-tolerance plateaus consistent with IST's predictions [1][2]. Large language models, which are built on the transformer architecture introduced in 2017, have driven a boom in AI investment and public deployment since the early 2020s [4]. These models rely on deep neural networks with multiple hidden layers capable of learning hierarchical representations from vast datasets [5]. Yet IST suggests that even these advanced architectures operate without an explicit model of the user's latent intent, treating the prompt as the primary object of exchange [1][2]. In 2023, hundreds of AI experts and public figures signed a statement declaring that mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war [3]. By 2025, a separate statement calling for a ban on the development of superintelligence had been signed by five Nobel Prize laureates and former senior US national security officials including Michael Mullen and Susan Rice [3]. The IST framework enters this landscape by proposing a formal vocabulary for a layer of interaction that, if incorporated into system design, could alter how AI systems infer and act on human goals [1][2].