Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Report on Candidate Computational Indicators for Conscious Valenced Experience

Andres Campero

TL;DR

This work surveys 13 computable indicators proposed as constituents of conscious valenced experience, drawing on animal, clinical, philosophical, evolutionary, neuroscientific, and AI literatures. It casts these indicators in computational terms and organizes them around decision-making, learning, and internal processing, while clarifying cross-relations and limitations among indicators. By articulating concrete mechanisms—from forward modeling and RL to predictive processing and hierarchical self-representation—the paper outlines candidate architectures for valenced consciousness and raises implications for AI safety, ethics, and scientific understanding. The study also emphasizes that indicators are neither universally necessary nor sufficient and calls for further work to map them onto phenomenology and moral status across species and artificial systems.

Abstract

This report enlists 13 functional conditions cashed out in computational terms that have been argued to be constituent of conscious valenced experience. These are extracted from existing empirical and theoretical literature on, among others, animal sentience, medical disorders, anaesthetics, philosophy, evolution, neuroscience, and artificial intelligence.

Report on Candidate Computational Indicators for Conscious Valenced Experience

TL;DR

This work surveys 13 computable indicators proposed as constituents of conscious valenced experience, drawing on animal, clinical, philosophical, evolutionary, neuroscientific, and AI literatures. It casts these indicators in computational terms and organizes them around decision-making, learning, and internal processing, while clarifying cross-relations and limitations among indicators. By articulating concrete mechanisms—from forward modeling and RL to predictive processing and hierarchical self-representation—the paper outlines candidate architectures for valenced consciousness and raises implications for AI safety, ethics, and scientific understanding. The study also emphasizes that indicators are neither universally necessary nor sufficient and calls for further work to map them onto phenomenology and moral status across species and artificial systems.

Abstract

This report enlists 13 functional conditions cashed out in computational terms that have been argued to be constituent of conscious valenced experience. These are extracted from existing empirical and theoretical literature on, among others, animal sentience, medical disorders, anaesthetics, philosophy, evolution, neuroscience, and artificial intelligence.
Paper Structure (18 sections, 6 figures)

This paper contains 18 sections, 6 figures.

Table of Contents

  1. Consciousness Research
  2. Other computational properties:
  3. Phenomenological Qualities:
  4. Relationships to other indicators:
  5. Functional Mechanism:
  6. Functional Mechanism:
  7. Functional Mechanism:
  8. Other computational properties:
  9. Functional Mechanism:
  10. Other computational properties:
  11. Phenomenological Qualities:
  12. Functional Mechanism:
  13. Other computational properties:
  14. Functional Mechanism:
  15. Phenomenological Qualities:
  16. ...and 3 more sections

Figures (6)

  • Figure 1: Anticipatory Behavioral Autonomy
  • Figure 2: Seymour's RL Model
  • Figure 3: Hierarchical Forward Model. Two different depictions of the same model with three tiers. The first level is the sensory processing. A second monitoring circuit, forms second order awareness ($IM_{1}$ on the right), and receives a copy of the input to predict the output of the sensory module, the prediction is compared to the actual output to update the model. A second internal model $IM_{2}$ is responsible for conscious awareness and uses the input and the Global Input from other processors to predict the second-order prediction. The prediction of $IM_{2}$ is broadcasted globally and the error is used to update the second model.
  • Figure 4: Conscious Turing Machine
  • Figure 5: Imperativism about pain. Signalling game from Martinez and Klein 2016 martinez2016pain. The game is played in the presence of inflammation. The environment can be in states S1, S2, S3 of a nail receiving severe, mild, or null mechanical stimulation, respectively. The sender (a nail nociceptor perhaps) has four available messages; the receiver (the motor cortex) can protect the nail with high, medium, low, or null priority (A1,A2,A3,A4 respectively). (a) Reflects the shared payoffs of the game where in the absence of inflammation, the best action to a mild mechanical stimulation is low priority protection and to a severe stimulation it is high priority protection. The Nash equilibrium is given by receiver and sender responses (b) and (c). If all states are equiprobable, the mutual information between action and message is higher than that between state and message ($\mathbf{I(A;M)>I(S;M)}$), hence the messages are predominantly imperative. These messages can be reasonably interpreted as the imperatives: "protect with very high, high, low, or null priority" (M1, M2, M3, M4 respectively).
  • ...and 1 more figures