Novaberg
Research
Notes on agency, substrate, and the architecture of artificial counterparts — concerned with what a system would need in order to bring something to an interaction rather than wait for instructions.
Series II
Agency & Self-Observation
2 papers · September 2026
I
Programming the Unprogrammable — Drive, Curiosity, Will in AI Systems
Drive, curiosity, and will are not programmable as actions; they are states of a four-component substrate that can be built.
II
EN
DE
Skill and Stance — What a System Can Improve About Itself, and What It Cannot
Systems improve their skills; a stance has no state that could improve. Two halves: the architecture of self-observation, and 87 % capitulation measured without it.
EN
DE
Forthcoming
In preparation
3 papers
III
Dual emotion: asymmetric empathy in artificial affect dynamics
The affect-state architecture in full — the alpha lookup table, perceived versus held affect, calibration.
IV
Concern engineering as a discipline: configuring agents that don't go HAL
What it means to load concerns responsibly — mutual incompatibility, situational stress, renegotiation under pressure.
V
The will architecture: from specification to running code
Implementation notes — option generation, drift-diffusion, the commitment threshold.
Novaberg is a personal AI project running on consumer hardware. The work collected here is a single developer's attempt to take the architectural questions of artificial agency seriously, drawing on the cognitive-science literature rather than reinventing it from prompt engineering.
Replications, extensions, and divergent implementations are explicitly invited. The architectural prose is licensed under CC BY 4.0, the reference implementation under Apache 2.0. Correspondence: master@novaberg.de.