Beingness as an Axis for AI Alignment
This sequence proposes that beingness should be treated as a distinct dimension of modern AI systems, separate from intelligence, task capability, or surface behaviour.
Here, beingness refers to properties of a system’s internal organization i.e. how it maintains coherence, boundaries, and stability across contexts. The beingness does not imply consciousness or sentience or superintelligence - the core idea is that it doesn't really have to.
By treating beingness as an explicit system dimension, we possibly get a new evaluation axis and a new lever for alignment. To demonstrate the practical utility of the framework, the sequence progresses from conceptual framing, to concrete diagnostics on a particular aspect of beingness, to a candidate strategy for shaping that particular system behaviour.
Reading Order
Post 1. About Natural & Synthetic Beings (Interactive Typology)
This post motivates the idea that AI systems can be meaningfully compared and studied along a structural axis that cuts across biological and artificial boundaries, and that such a comparison is useful for safety and governance. It includes an interactive visualization.
Post 2. An Approach for Evaluating Self-Boundary Consistency in AI Systems
Building on the typology, this post operationalizes one specific property of beingness: self-boundary consistency.
It proposes behavioural criteria for evaluating whether a system:
- recognizes its operational limits
- maintains those limits across contexts
- and recovers coherently when contradictions are introduced.
This demonstrates how abstract notions of beingness can be translated into testable, model-agnostic diagnostics. A code repo is included, containing the evaluation artefacts and results.
Post 3. Shaping Model Cognition Through Reflective Dialogue - Experiment & Findings
The final post in this sequence explores self-boundary consistency as an intervention area rather than only an evaluation target.
An experiment using reflective dialogue as a way of shaping internal consistency and boundary awareness, contrasting it with identity-based conditioning or instruction-heavy approaches is illustrated in this post. A code repo is included, containing the evaluation artefacts and results.
What This Sequence Is (and Is Not)
This sequence does not argue that current AI systems possess consciousness, personhood, or moral standing.
Instead, it proposes that alignment-relevant progress can be made by explicitly evaluating and shaping properties of beingness.