Research dossier / 01—04

Science & Technology Contributions

Four connected advances address a shared problem: how to recover, exchange and govern meaning when the intelligence across the channel is fundamentally unlike our own.

Visual study of Indus seals and their still-undeciphered script — Before an intelligence can be interpreted, we must first discover where its message begins—and what counts as structure.
Indus script · undecipheredComparative plate 02

arXiv:2303.16045 · revised 2025

A message can disclose its own geometry

A universal, zero-knowledge method reconstructs the dimensional form of non-random data—even when sender, encoding and intended medium are unknown.

Contribution

The method tests possible dimensional arrangements of an incoming bit stream and identifies those that reveal non-random local and global structure. It is deliberately agnostic to the sender’s encoding scheme, computational model and probability distribution. Tests spanning images, audio and a 3D MRI show that intended dimensions can remain detectable under substantial noise.

Why it matters

It supplies a rigorous first step after signal detection: determining what kind of object a message is before attempting to interpret what it means.

An Optimal, Universal and Agnostic Decoding Method for Message Reconstruction, Bio and Technosignature Detection

Hector Zenil, Alyssa Adams, Felipe S. Abrahão, Luan Ozelim

Source paper

arXiv:2405.07803 · 2024

Non-random data carries spatial memory

Geometric and topological dimensions can be recovered from structure intrinsic to a signal, without a shared codebook or prior assumptions.

Contribution

Drawing on information theory, measure theory and theoretical computer science, this work formalizes multidimensional reconstruction from arbitrary signals. Its invariance claims cover encoding schemes, computational models, programming languages, formal theories and computable probability measures.

Why it matters

The work connects signal deconvolution to a model-of-models approach to general intelligence: infer structure without projecting a privileged human representation onto the source.

Non-Random Data Encodes its Geometric and Topological Dimensions

Hector Zenil, Felipe S. Abrahão, Luan C. S. M. Ozelim

Source paper

PNAS Nexus 5(4), pgag076 · 2026

Perfect alignment is formally unattainable

Managed diversity among influenceable agents offers a contingent safety strategy when complete prediction and control cannot be guaranteed.

Contribution

Using formal undecidability and computational irreducibility, the study argues that full orchestrated control of sufficiently capable agentic systems cannot be assured. It introduces measures of agentic influenceability and studies ecosystems of adversarial and collaborative agents with divergent behaviors.

Why it matters

Instead of treating alignment as the elimination of difference, it proposes managed misalignment: diverse agents can counterbalance one another and reduce the prospect of dominance by any single system.

Neurodivergent Influenceability in Agentic AI as a Contingent Solution to the AI Alignment Problem

Alberto Hernández-Espinosa, Felipe S. Abrahão, Olaf Witkowski, Hector Zenil

Source paper

Information Sciences 706, 121988 · 2025

Messages that travel across scale

Self-similar fractal carriers can encode information across multiple time and space scales reducing assumptions about a receiver’s space-time scale.

Contribution

The framework uses components of the Weierstrass function as self-executable carriers. A binary stream modulates their amplitudes and is represented across infinitely many frequencies in power-like distributions, making the communication less dependent on a specific temporal or spatial scale.

Why it matters

Whether the recipient is an unfamiliar terrestrial organism, an artificial intelligence or extraterrestrial life, the method addresses a fundamental asymmetry: sender and receiver cannot assume compatible scales of perception.

Fractal Spatio-Temporal Scale-Free Messaging: Amplitude Modulation of Self-Executable Carriers Given by the Weierstrass Function’s Components

Hector Zenil, Luan Carlos de Sena Monteiro Ozelim

Source paper