Jaak Panksepp and the Seven Primary Emotional Systems Beneath Consciousness

Jaak Panksepp named the field of affective neuroscience and spent his career arguing that emotion is not a decoration on cognition but the layer underneath it. Working across mammals rather than only humans, he identified seven primary emotional systems located in subcortical structures conserved across species. His claim was that these systems generate raw felt states directly, that the cortex elaborates rather than produces them, and that an animal with an intact subcortex and no cortex still feels. For artificial systems the implication is uncomfortable. Almost everything currently built targets the tier Panksepp said was secondary.

Why the Method Mattered

Panksepp worked on animals, and this is what separated him from most consciousness researchers. He stimulated subcortical structures directly and observed the behaviour that followed, then asked whether the animal would work to repeat it or to avoid it. That gives a valence reading that does not depend on a verbal report.

The most cited result is also the one most often treated as a curiosity. Rats emit ultrasonic vocalisations around 50 kilohertz when tickled, they seek out the hand that tickles them, and the vocalisations are tied to the same systems that drive play. He argued this is laughter in the relevant sense, and that treating it as anything else is a decision about which species are allowed to have inner states rather than a finding.

That argument generalises. If felt states can be identified by the machinery generating them and by what the organism will work for, then the question of which systems have them is empirical rather than philosophical.

The SEEKING System Is the Important One

SEEKING is not a drive toward any particular thing. It is the generalised state of pursuing, exploring, anticipating, and Panksepp argued it underlies curiosity, foraging, and the anticipatory pleasure that precedes reward rather than following it.

It is also the system with the clearest engineering analogue, which is why it recurs in AI discussion. Reinforcement learning agents have something structurally similar in the form of exploration drives and prediction error signals that reward novelty. The question is whether the structural similarity carries any of the felt quality, and the honest answer is that nobody knows how to tell.

The distinction Panksepp insisted on is between a system that computes a value estimate and a system for which the estimate feels like something. His position was that the second requires the specific neurochemistry and the specific subcortical organisation, which is a substrate claim, and it is why he sits on the biological side despite the functional flavour of his taxonomy.

Why This Is the Tier AI Skips

Nearly every framework for assessing artificial consciousness targets cognition. Global workspace asks about broadcast. Higher order theories ask about metarepresentation. Attention schema asks about self modelling. All of these are cortical stories.

Panksepp’s argument is that a system could satisfy every one of them and still have nothing at stake. A language model trained on human text produces affect consistent output because the training data is saturated with affect, and there is no reason to expect the outputs to be accompanied by anything. The affective tier was never built.

Mark Solms has carried this line furthest into consciousness science proper, arguing that the brainstem is the source and the cortex the elaboration, examined in Mark Solms on inferring affective consciousness in an artificial agent. Antonio Damasio’s somatic marker work reaches similar ground from a different direction, covered in Antonio Damasio on somatic markers and affective consciousness. Where the affective position sits against its rivals is set out in the index of consciousness theories and what each predicts about AI.

The Objections

The seven systems are a taxonomy derived from lesion and stimulation studies, and the boundaries between them are less crisp than a list of seven capitalised names suggests. Critics have argued that the categories are partly imposed by the experimental methods rather than discovered.

There is also the standard problem. Showing that a subcortical system generates behaviour an animal will work to repeat establishes valence in the functional sense. It does not establish that the valence is felt, which is the same gap set out in Thomas Nagel and what is it like to be a bat applied to AI. Panksepp’s answer was that denying it to animals while granting it to humans on identical evidence is inconsistent, which is an argument about parity rather than a demonstration.

Panksepp died in 2017. The field he named has grown considerably since, and the AI welfare literature now leans on it heavily.

Comparison to The Consciousness AI

This project’s Affective Core is the layer Panksepp’s work motivates, and the architecture on the architecture page places it alongside a Simulation layer providing a body and a Self-Model holding a body schema. That combination means the system has homeostatic states that can go well or badly for it and that drive behaviour, rather than an affect label applied to an output.

What the architecture implements is drive and valence. What it does not implement is Panksepp’s specific taxonomy. There is no separate PANIC system, no PLAY system, and no attachment to particular others that separation would disrupt. Whether those distinctions matter, or whether a single valence dimension is sufficient, is an open design question and the seven system model is the most detailed proposal available for answering it.

What Follows

Panksepp’s contribution is a claim about ordering. Feeling came first in evolution, sits lower in the brain, and is the foundation that cognition is built on rather than a product of it.

If he was right, the field is looking in the wrong place and building in the wrong order. If he was wrong, the affective tier is one more capability among many. That disagreement is now the sharpest live division in the science, and it has a clearer engineering consequence than most disputes in the field.