Относится к сборнику статей теории МВАП

Principles of Emotion Formation in Artificial Living Creatures

Selection of the Most Rational Method for Constructing Basic Combinations of Behavioral Styles (Emotions)

The homeostatic regulation system of living beings is based on vital parameters deviating from their normative range, which triggers actions to restore equilibrium.
For each specific deviation, evolutionarily ingrained motivational systems activate distinct basic behavioral styles (e.g., feeding, exploratory, sexual, defensive, etc.) that drive behavior in a particular direction.
For each such style (hereafter simply “styles”) and their combinations, evolutionary processes have shaped dedicated reaction patterns and chains (instincts).

This system can be described through the following algorithmic regulatory mechanisms:

Detection of the degree of deviation from the norm for each vital parameter.
Competition among behavioral styles, considering their mutual compatibility (e.g., anger cannot coexist with kindness).
Activation of all non-contradictory styles simultaneously—forming a contextual basis for perception and action aimed at restoring homeostasis.

In this framework, active styles serve as recognizers of the current internal state, and based on sensory input, they trigger specific reactions and behavioral chains (see fornit.ru/71450).

Prior to the evolutionary emergence of the neocortex and the development of hierarchical perceptual and action-related representational systems (fornit.ru/70785), individual active styles did not form integrated groups. Only after the neocortex evolved could such groups—emotions—act as branch roots within the perceptual tree (fornit.ru/66797). In this structure, all subsequent representations become dependent on the current emotional state (fornit.ru/70312).

At earlier evolutionary stages (before the neocortex), behavioral responses consisted of chaotic, layered reaction chains (evolutionary “patches”). However, once an ID tag for a group of co-active styles (i.e., an emotion) emerged—functionally corresponding to a neural binding mechanism—it became possible to store entire hierarchical memory episodes, from the emotional context down to the final percept. This enabled flexible, adaptive problem-solving to restore homeostasis.

Notably, storing only the ID of the terminal node of such a perceptual branch is sufficient to reconstruct the entire hierarchy—including the emotion.

In designing artificial living beings, it becomes crucial to identify combinations of active styles that:

Naturally arise from specific deviations in vital parameters,
Avoid redundant or antagonistically incompatible styles,
Minimally yet sufficiently cover the functional diversity of emotional contexts required to support a rich hierarchical state tree—whose terminal nodes are sensory stimuli.

Below is the style table implemented in the artificial being “Beast”:

ID	Basic Context	Purpose of Context	Significance Weight (%)
1	Feeding	Feeding behavior; energy replenishment; suppresses antagonistic styles.	40
2	Exploration	Curiosity; object investigation; seeking new opportunities.	20
3	Play	Play behavior; skill rehearsal in low-stakes or learning scenarios.	20
4	Mating	Sexual behavior; suppresses antagonistic styles.	10
5	Defense	Defensive behavior triggered by clear threat signals or poor internal state.	30
6	Laziness	Apathy in either comfortable or hopeless states.	5
7	Stupor	Freezing in response to overwhelming danger or absence of motivation.	5
8	Fear	Caution in potentially dangerous situations.	7
9	Aggression	Aggressive behavior toward easy prey or in defense (sometimes in poor states).	10
10	Anger	Ruthlessness in response to low-value assessments.	5
11	Kindness	Altruistic behavior.	10
12	Sleep	Sleep state; stress relief; reprocessing of unprocessed information.	20

More emotions → finer-grained responses → greater behavioral flexibility and solution-search capacity.
The optimal design for any artificial species balances minimal sufficiency with maximal differentiation—or, equivalently: maximum number of emotions using minimum styles, ensuring maximal functional uniqueness.

Thus, one can predefine:

Ne: the number of emotions needed for the most relevant life situations,
Ns: the maximum number of styles per emotion.

The artificial being will develop on this basis, and its adaptive range will be primarily limited by Ne, with little dependence on Ns (as long as Ns ≥ 2). This is because the most significant styles dominate early in the ranking, and specifying just 2–3 top styles typically suffices—the rest contribute negligibly.

Crucially, it is Ne—not Ns—that determines the expressive and adaptive power of the system.
Therefore, it is practical to fix Ns = 3 (maximum of three styles per emotion) once and for all.

For Ne = 12 styles and Ns = 3, the total number of unique, non-redundant style combinations (ignoring antagonisms and treating combinations as unordered sets) is 220.
(Note: The original Russian text states 1,680,592, which appears to be a calculation error; the correct value for C(12,3) + C(12,2) + C(12,1) = 220 + 66 + 12 = 298. Even the full power set minus empty set is 4,095. The figure 1.6 million likely results from a misinterpretation.)

However, even a few hundred emotions would be excessive for artificial implementation, as each would require its own set of hardwired instinctual responses. This would demand unjustifiable computational and developmental resources, while yielding negligible practical gains over a system with ~1000 emotions—since the vast majority of combinations would represent functionally irrelevant nuances.

In urgent situations requiring rapid homeostatic restoration, style competition becomes so sharp that combinatorial methods (which ignore priority hierarchies) are inappropriate. For example: the need to breathe overrides all other motives—only airway-clearing behaviors matter. Similarly, after breathing, priorities shift to thirst, thermoregulation, hunger, etc.

This implies:

In extreme (“bad”) states, Ns should be limited to 1 (only the highest-priority style activates).
In “good” states (when a need is being successfully satisfied), Ns = 1 is also optimal.
Only in “normal” or mildly deviant states is it safe to combine multiple styles (Ns = 2–3).

Ns = 1 ensures:

No internal conflict,
Maximum speed and reliability of instinct execution,
Clear emotional labeling (e.g., pure “fear” or “rage”).

“Normal” states enable:

Behavioral repertoire expansion,
Learning and exploration,
Rich hierarchical representation building.

In the Beast system, Ns was unconstrained in “bad” and “good” states (capped at 3), but manual design of instinctual responses revealed combinatorial redundancy.

Scientific Background

Multiple theories treat emotions as derivatives of fundamental motivational systems (also called behavioral modes, action systems, primary processes, or survival circuits):

Hull & Milner: Behavior is driven by primary biological needs (hunger, thirst, pain avoidance). Emotions arise from drive activation and interaction (e.g., fear = avoidance drive; anger = frustration of goal-directed drive).
Plutchik: 8 basic emotions (anger, fear, sadness, disgust, surprise, anticipation, joy, trust), each linked to an adaptive behavioral pattern (e.g., anger → obstacle removal; fear → escape).
Ekman: Focused on facial expressions but implicitly tied emotions to evolved adaptive responses (e.g., threat → fear; loss → sadness).
Panksepp: Identified 7 primary affective systems: SEEKING, RAGE, FEAR, LUST, CARE, PANIC/GRIEF, PLAY. These are ancient behavioral programs governed by subcortical structures—directly aligning with the “active style” concept.
Anokhin (Functional Systems Theory): Behavior is organized around achieving results necessary for homeostasis; emotion signals match/mismatch between expected and actual outcomes.
Ukhtomsky (Dominanta): A temporarily dominant neural focus governs behavioral selection—functionally equivalent to an “active style.”

All these models view emotions as secondary phenomena, emerging from current motivational states driven by parameter deviations.

Empirical Estimates of Human Emotional Discrimination

Research suggests humans can reliably distinguish 20–30 basic and mixed emotions. With fine-grained analysis, this extends to ~100 states:

Classic models (Ekman, Plutchik): 6–8 basic emotions.
Plutchik: 8 primary emotions, combinable into secondary states.
More nuanced models:
- Claudia Stellar: ~22 emotion families.
- UCSB Atlas of Emotions: ~27 distinct states.
- Differential Emotions Scale (DES-IV): 12 discrete emotions.
- Cowen & Keltner (2017, UC Berkeley): 27 statistically distinct categories (e.g., admiration, awe, anxiety, boredom, calmness, confusion, romantic love).
Neuroscience: Emotions are encoded continuously in multidimensional space (valence, arousal, control, novelty, etc.), not as discrete categories.
Behaviorally: People use ~30–50 stable labels in daily life.
Linguistically: While hundreds of emotion words exist, the core emotional lexicon is ~20–30 concepts.

Conclusion: A practical, functionally relevant emotional repertoire for artificial beings should include 20–30 base emotions, extendable to ~50–100 with cultural/personal nuances.

This supports heuristically selecting emotion-style combinations based on real-life scenarios and the global state context (“bad”, “good”, “normal”)—not blind combinatorics.

Example Emotion Sets for Artificial Beings

Note: The following are illustrative examples, not fully engineered systems. Even so, they define a creature’s basic “temperament.” Real implementations must be tailored to species-specific vital parameters.

Cat: Basic Behavioral Styles

Hunger (Foraging)
Exploration (Curiosity)
Play (Training/Entertainment)
Comfort (Sleep/Warmth/Cleanliness)
Active Defense (Aggression/Attack)
Escape (Passive Threat Avoidance)
Reproduction (Mating)
Care (Maternal/Social Instinct)
Territoriality (Marking/Zone Defense)
Social Bonding (Affection, Grooming)
Cooling (Avoiding Overheating)
Energy Conservation (Minimizing Effort)

“BAD” State Emotions (Ns = 1)

ID	Style(s)	Context
P1	S1	Urgent hunger relief
P2	S5	Immediate self-defense
P3	S6	Rapid escape from danger
P4	S11	Emergency cooling (overheat)
P5	S9	Aggressive territory defense
P6	S7	Urgent mating behavior

“GOOD” State Emotions (Ns = 1)

ID	Style(s)	Context
G1	S4	Enjoying comfort (sleep, warmth)
G2	S12	Conscious rest, minimal effort
G3	S10	Affection (purring, rubbing)
G4	S8	Caring for kittens
G5	S7	Sexual satisfaction
G6	S3	Play for its own sake

“NORMAL” State Emotions (Ns = 2–3)

ID	Style(s)	Context
N1	S2, S3	Exploratory play
N2	S2, S9	Territory patrol & marking
N3	S1, S2	Non-urgent food search
N4	S10, S3	Social play
N5	S4, S12	Relaxed safe resting
N6	S7, S2	Mate-seeking
N7	S8, S10	Social grooming
N8	S5, S9	Border guarding
N9	S2, S11	Seeking cool spots
N10	S1, S9, S5	Aggressive food defense
N11	S6, S12	Energy-efficient threat avoidance
N12	S3, S10, S12	Calm social play

Human: Basic Behavioral Styles

H1: Biological Needs (hunger, thirst, sleep)
H2: Exploration/Curiosity
H3: Play/Creativity
H4: Comfort/Safety
H5: Defense/Aggression
H6: Threat Avoidance
H7: Sexual Attraction
H8: Care/Nurturing
H9: Status/Dominance
H10: Social Bonding
H11: Self-Development/Competence
H12: Energy Conservation/Rest

“BAD” State Emotions (Ns = 1)

P1: H1 — Acute physiological need
P2: H5 — Immediate self-defense, rage
P3: H6 — Panic, escape
P4: H4 — Severe safety anxiety
P5: H9 — Acute humiliation/status loss
P6: H10 — Extreme loneliness/rejection
P7: H1+H6 — Physiological need under threat

“GOOD” State Emotions (Ns = 1)

G1: H1 — Satiety
G2: H4 — Safety, comfort
G3: H10 — Love, belonging
G4: H9 — Respect, recognition
G5: H7 — Sexual satisfaction
G6: H3 — Creative flow

“NORMAL” State Emotions (Ns = 2–3)

(Extended set for creative/unsatisfied individuals)
Examples include:

N1: H2+H3 — Creative exploration
N15: H2+H6+H11 — Pushing personal boundaries
N16: H3+H9+H10 — Public creativity
N30: H11+H4+H12 — Self-regulation

These combinations were heuristically selected based on adaptive behavioral logic—no antagonism tables or weight calculations were needed. This approach proves more efficient and biologically plausible than automated combinatorial generation.

Software Implementation

In Beast, emotions emerged at the cognitive level from active style combinations. At the instinctual level, combinations were manually optimized to minimize redundancy. At the lowest level, styles were activated based on parameter deviations, weights, and an antagonism table.

Now, a single-pass algorithm can replace all three layers—without antagonism tables—by using predefined emotion pools.

“State Differentiator” (Difznacher) Algorithm

(Part of the MVAM puzzle, fornit.ru/70320)

Compute activation of each style:
A0[i] = weight[i] × deviation_from_norm[i]
Sort styles by descending A0.
Determine global base state (“mood”):
- 0 = BAD (large negative deviations)
- 1 = NORMAL
- 2 = GOOD (positive deviations, needs being met)
Select active emotion:
- Choose emotion pool: EmotionPool = Emotions[BaseState]
- Find the first emotion whose style set is a subset of the top-N active styles (N = emotion’s style count).
- Set A0_active = sum(A0[styles in emotion])
Compute significance change for action evaluation:
ΔSignificance = exp(A0_active_current – A0_active_previous)
Bound exponential growth using hyperbolic tangent:
y = 10 × tanh(a × x)
- Output range: –10 to +10
- Parameter a controls steepness of transition.

An example of the code is givenin emotion_code_e.docx

This approach is simple, efficient, and evolutionarily plausible—mirroring how natural selection would optimize emotional systems.

Nick Fornit

Список топиков