Artificial Emotional Intelligence
This diagram illustrates a decoupled, two-stage neural architecture designed to bridge raw sensory perception and ethical decision-making. Rather than training a single end-to-end "black box" network that maps camera pixels directly to robot motor commands, the system splits the problem into Affective Appraisal (Stage 1) and Affective-Moral Decision Making (Stage 2).
By splitting the AI brain of the care agent into an Emotional Interpreter (Stage 1) and a Moral Decision-Maker (Stage 2), the system moves beyond mechanical task execution. It perceives human vulnerability beneath surface frustration, allowing it to deliver care that is both physically safe and deeply respectful of human dignity.
This plugs into SALAI 3 as part of it's moral sub-system
Synthetic EmotionAI can receive a negative reward and learn to avoid an outcome, but does that mean the AI actually experiences anything? This diagram explores the distinction between simply processing reward and penalty signals and having persistent, system-wide internal states resembling pain, distress or joy.
The interesting question isn't whether machines can simulate emotion. It's whether architectures involving internal sensing, vulnerability, global state changes and recurrent self-perception could ever turn simulation into something more. We often talk about AI being "rewarded" or "penalised", but those are essentially learning signals, not evidence that an AI feels pleasure or pain.
This diagram explores a fascinating possibility: perhaps something resembling artificial emotion would require much more, including internal sensing, something meaningful to lose, system-wide state changes and continuous feedback about the AI's own condition. Whether that could ever produce genuine subjective experience remains an open philosophical question, but it's an intriguing direction to think about.
- Nociception = non-detection of harmful stimulation (opposite of perception).
- Constructed pain = the brain integrates the signal with attention, context and other brain processes to classify the situation as painfall.
- Dopamine / reward prediction error (RPE) = broadly, differences between expected and actual outcomes contribute to learning. The graphic emphasises that this learning signal is not itself pleasure. The delta between the baseline dopamine level determines the perception
- Joy = portrayed as a broader whole-system state rather than a single electrical reward signal.
- Brain areas including the prefrontal cortex, insula, anterior cingulate and ventral tegmental area (VTA) are shown as parts of this larger system.
(IMHO)
