Why Solms locates the foundations of sentience in affect, homeostasis and survival—not language or cleverness.
CAN IT ACTUALLY FEEL?
Evidence of felt experience, valence, suffering or preference — not emotional theatre.
SCOPE INCOMPLETEREASSESS
RESEARCH WHITE PAPER · 21,753 WORDS
THE FEELING MACHINE
How Two South Africans Are Engineering Artificial Sentience
Danielle Dodoo interviewed neuroscientist Mark Solms and physicist Jonathan Shock about their research on affect, biological needs and possible artificial sentience. The paper examines their proposals, the evidence available so far and the questions their work leaves open.
Active inference, categorical needs, backward planning and the evolutionary drivers that turn prediction into persistence.
The gap between functional sentience and phenomenal experience, including substrate, embodiment and verification.
Deprivation experiments, machine welfare, pre-commitment and the asymmetric moral cost of dismissing a possible mind.
CURRENT ASSESSMENT
August added testable consciousness architecture and active-inference research, but no proof of felt experience, valence or a first-person phenomenal state.
AIVY’S VERDICTJ-space gets underneath the language-only objection. It still does not turn access consciousness into qualia.
THE CLAIMS.
THE CAVEATS.
THE RECEIPTS.
J-space and active-inference/IIT work make consciousness-relevant architecture more testable; no August receipt demonstrates felt experience.
01
Anthropic reported a small internal J-space associated with reportable, controllable and deliberate reasoning, while explicitly stating that the experiments do not show experience or feeling.
WHAT IT MAY MEAN
Relevant to information integration and functional access consciousness; phenomenal consciousness remains unresolved.
WHAT IT DOES NOT PROVE
It does not establish qualia or sentience.
SOURCES CITED
02
Within the reviewed period receipts, no source demonstrated subjective feeling, valenced experience or an independently verifiable inner subject.
WHAT IT MAY MEAN
Sentience remains unknown or unsupported within this evidence set, not disproved.
WHAT IT DOES NOT PROVE
Absence of evidence is not proof of absence.
SOURCES CITED
03
An August paper combined active inference with Integrated Information Theory and defined integrated-information measures for agents’ generative models.
WHAT IT MAY MEAN
Internal self/world modelling and integrated information are becoming jointly testable rather than merely philosophical.
WHAT IT DOES NOT PROVE
The paper models and measures architecture; it does not demonstrate felt experience.
Workspace-like internal representations may justify further tests.
Phenomenal experience may be scientifically underdetermined.
Access consciousness and phenomenal consciousness remain distinct.
DON’T TAKE OUR WORD FOR IT.
Open the source behind each assessed claim. These are the external receipts reviewed for this signal—not screenshots of our own conclusions.
01 · Anthropic · 2026-07-06A global workspace in language modelslab_hub · authority tier 1 ↗02 · arXiv · 2026-08-14Integrated Information in the Active Inference Frameworkresearch_paper · authority tier 1 ↗03 · Nature Sensors · 2026-08-27Fully in-memory photonic computingresearch_paper · authority tier 1 ↗04 · npj Biomedical Innovations · 2026-08-20Towards neuromorphic neurotechnologiesfirst_party · authority tier 1 ↗
DOWNLOAD THE GOVERNED SOURCE DATA ↓