New Research · Generative Closure Theory

What makes capabilities emerge in language models?

My new paper models language models as reflexively autocatalytic constraint systems. It proposes measurable, falsifiable tests for when internal constraints become mutually supporting—helping explain robust reasoning, transfer, and recovery from error.

Read the New Paper Explore GCT →
Armando Vieira
150+ Publications
2000 Early Deep Neural Networks Paper
25+ Years Experience
PhD Physics
ISO AI Committee

Featured In

Fortune Towards Data Science The CIO Magazine Finextra TEDx
Early AI Work

Pioneering Deep Neural Networks Before the Boom

Long before the 2012 ImageNet moment made deep learning famous, my research explored large neural networks for real scientific measurement problems. In a 2000 paper on Rutherford backscattering data, we used networks with up to 10 layers and showed that they could deliver fast, accurate analysis for a difficult inverse problem.

The point was practical, not fashionable: after training, the network could recognize spectra almost instantly, opening a path toward automated online analysis and optimized experimental conditions.

View the 2000 Paper
2000

Large Neural Networks for Scientific Inference

Rutherford backscattering neural network analysis figure
RBS neural-network analysis from the early deep-learning work.
Architecture Large neural networks, up to 10 layers
Domain Rutherford backscattering spectra
Result Fast recognition after training
Why it matters Practical deep architectures before the mainstream wave
Recent Work

Generative Closure Theory (GCT)

I am currently developing Generative Closure Theory, integrating the Experiential Coherence Framework (ECF) with Reflexively Autocatalytic and Food-generated (RAF) network theory. This programme studies how adaptive systems construct, preserve, and transform the constraints governing their own behaviour.

The goal is to move beyond treating intelligence as only prediction or computation, and toward models of systems that maintain coherence under pressure, reorganize under contradiction, and sometimes cross thresholds into new forms of order.

Open GCT Page View Papers →
GCT ECF RAF Markets Society Mind

Applications

  • Revolutions and institutional phase transitions
  • Market structure and clinical ecosystems
  • Reinforcement learning and coherence
  • Consciousness and active inference
New · Flagship Paper Generative Closure Theory · LLMs

Language Models as Reflexively Autocatalytic Constraint Systems

Why can models with similar validation loss display very different capabilities? This paper introduces behavioural and mechanistic closure measures, a training objective for internal constraint states, and a reproducible experimental programme for testing robust capability emergence.

Central empirical claim

Closure-sensitive measures should predict robust capability onset beyond scale, compute, loss, and unperturbed accuracy.

Read the paper About Generative Closure Theory →
ECF

Explaining Revolutions Through ECF

Macro-level change as coherence breakdown, constraint reconfiguration, and phase transition.

Read paper →
Application

ECF Dynamics in Financial Markets

Applying coherence-field thinking to market structure.

Read paper →
RL

Coherence and Reinforcement Learning

How consistency and coherence constraints reshape the interpretation of learning systems.

Read paper →
AI Alignment

On AI Alignment

A dedicated site on why AI alignment is not only a technical problem, but a deeper question about goals, power, institutions, and human responsibility.

Visit the AI Alignment Site Read the Book
alignment.armandoscience.com

Alignment as a societal, not technical, endeavour

  • Technical limits of objective alignment
  • Institutions, incentives, and governance
  • We can design Good AI
Books

Books

Research

Publications

Most recent articles →
Book

Introduction to Deep Learning Business Applications

A practical guide for leveraging deep learning in enterprise settings, covering implementation strategies and real-world case studies.

View on Amazon →
Research Paper

Deep Neural Networks Before the Boom

In 2000, we used large neural networks, including architectures up to 10 layers, for Rutherford backscattering data analysis, showing that deep architectures could solve real scientific inference problems years before the deep learning wave.

View 2000 Paper →
Research

Research Publications

Peer-reviewed research outputs focused on artificial intelligence methods, applications, and standards.

View Publications on Google Scholar →
Opinion

Opinion Articles

Long-form essays on AI strategy, governance, and societal impact.

Strategy

A Manifesto for Future Data Analysts

What I teach my students about the future role of data analysts and scientists in the age of AI.

Read article →
Governance

AI Is Not Just a Tool

An argument for treating AI as infrastructure that redistributes power, incentives, and accountability.

Read article →
Estratégia Feb 18, 2026

AI e a Ilusão da Estratégia Automatizada

Uma reflexão sobre responsabilidade, cultura e pensamento estratégico na era da automação.

Ler artigo →
Generative AI Jan 15, 2025

What Dr House teaches us about AI

Lessons from Dr House about diagnostic reasoning, uncertainty, and AI.

Read more →
AI Strategy Jan 8, 2025

LLMs for Business Analytics

How to use LLMs effectively in business analytics workflows.

Read more →
Ethics Dec 28, 2024

Responsible AI: impossible to align

Why alignment debates can become a category mistake in responsible AI.

Read more →
View All Articles
Research

Research Topics

Research Topic

Artificial Neural Networks

Architectures, learning dynamics, and optimization methods for robust, scalable neural systems.

Research Topic

Impact of AI in Society

How AI reshapes labor, institutions, trust, and human decision-making across social systems.

Research Topic

Applications of Artificial Intelligence in Science

Using AI to accelerate discovery in physics, biology, medicine, and complex scientific modeling.

Research Topic

Conscious Machines

AI models are incredible but they lack consciousness. I have a solution for that.

Read article →
Armando Vieira
About

Dr. Armando Vieira

With a PhD in Physics and over 25 years of experience in artificial intelligence, I have dedicated my career to bridging the gap between cutting-edge research and practical business applications.

My journey began in academic research, where I published pioneering work on deep neural networks. Today, I help organizations navigate the AI revolution through strategic advisory, education, and implementation support.

As a member of the ISO AI Committee, I contribute to shaping international standards for artificial intelligence. I am also the co-founder of Medgical.ai, bringing AI innovation to healthcare.

Download Full CV Watch TEDx Talk →
Entrepreneur

Co-founder, Medgical.ai

Medgical is a Series A AI-powered clinical documentation platform that automatically generates medical notes and reports from consultation audio, reducing administrative burden for physicians, with more than 100k consultations.

Visit Medgical.ai
Contact

Contact

Reach out directly via email or social channels.

✉️ armando@armandoscience.com ✍️ Medium 💼 LinkedIn