Reflexive Artificial Intelligence

What is missing in a language model that answers brilliantly but cannot say why it answered as it did?

This book transfers Johannes Heinrichs’ theory of reflection to the architecture of artificial intelligence. Its starting observation: today’s language models process signs and their statistical relations, but do not observe their own processing. If you ask such a system why it chose an answer, you get descriptions, not explanations. From this the book develops a proposal: to build AI systems with explicitly separated layers of reflection – one layer that works (L0), one that observes this working (L1), one that regulates it (L2), and one that aligns it with reasons and values (L3).

The book is addressed to developers, researchers and architects of AI systems. It says openly where the limit lies: Heinrichs himself wrote practically nothing on AI. The philosophical foundations – the four elements of meaning it, I, you and medium, the distinction between implicit and explicit reflection, the fourfold articulation of society – come from him; their transfer to AI architectures is the author’s work, and the text marks what belongs to whom in each case.

Structure

• Part I: The problem of current AI – pattern recognition without reflection, the problem of self-regulation, why language models do not “understand”.
• Part II: Foundations in the logic of reflection – the four elements of meaning as an architectural principle, implicit and explicit reflection in systems, the question of consciousness (with Searle’s “Chinese Room”), dialectical subsumption.
• Part III: Practical applications – reflexive agents, theory of mind, multi-agent systems with a you-level, metacognitive layers.
• Part IV: Concrete implementation – a Reflexive System Architecture (RSA) with the layers L0 to L3, conflict resolution between the levels modelled on a four-chamber model, reflexive semantic networks (RSN) and an evaluation and development platform (IEDP).
• Part V: Outlook – ethics case studies, the dilemma that the more reflexive an AI becomes, the less it is a mere tool, a research agenda and a proposal for collaboration.

An appendix offers a three-week self-study course, a workshop concept for development teams, exercises and an annotated bibliography.

What is special

The concepts do not remain abstract: there are numerous code sketches in Python for the architectures. The book understands itself as a research programme and not as a finished system; it names what is missing itself – empirical testing and collaboration between philosophy, computer science and AI practice.

Relation to the other books

Those who want to get to know the philosophical foundations begin with Introduction to Reflexive Systems Theory or, in more detail, with The Discovery of Johannes Heinrichs, whose chapter 13 shows where Heinrichs locates the difference between sign and enactment. Applications of the same logic of reflection to cooperation and conflicts are shown in Reflexive Systems Theory. The sources on which the book relies are Heinrichs’ Integral Philosophy and Dialectics as Logic of Reflection.

Bibliography

Year

2026

Edition

Johannes Heinrichs Edition · Introductions and applications by other authors

Licence

Zero Marginal Cost License v1.3

Downloads

This work is free to read and share under the Zero Marginal Cost License.

More by this author →