Artificial Sentience

Acknowledgements: Based on ideas from Karl Friston's Active Free Energy principle and Predictive Coding formulations.

Introduction:

In a highly simplified nutshell, Active Inference seeks to relate changes in the cognition of our inner world model, to changes in our sensory inputs, using several matrices of probabilities. We plan to borrow from those concepts, and from Deep Learning techniques, to explain ideas like awareness and free-will. For instance, if our cognition and choices are, indeed, governed by probabilities, which are, in their essence, determined by prior conditions, how do we explain the sense of free-will? And, is free-will "real"? My personal take on that is yes.

The journey starts by modeling some core Generative Models in Active Inference, and then trying to transport them to a Deep Learning architecture. Longer term objectives include helping to explain disorders such as Autism, and designing LLM architectures with an emphasis on sentience. Why? Because text is just a sequence of symbols, that can be used in an easily customizable environment, to further develop our understanding of the biological roots of sentience and consciousness.

Throughout our work, we will develop several simulations and list them here:

1) Infinite Regress.

2) Goal-directed behavior and surprise.

3) A Prototype Model of Self-Awareness?

4) Appendix to: A Prototype Model of Self-Awareness?

5) Curiosity, Motivation and Discomfort.

6) Tonic Dopamine as the Engine of Motivation: A Computational Exploration

7) Phasic Dopamine, Cue Sensitization, and the Emergence of Addiction: A Computational Exploration

8) Fear as the Counterpart of Motivation: A Computational Exploration.

9) From Simulation to Embodiment: Toward Artificial Motivation, Curiosity, and Fear.

10) Building Proto-Affective Agents with Active Inference