Emergent Manifestations of Psyche in Simple Python-Based Thought Simulation

How my HDC system exhibited a "panic attack" without specifically tailored code.

In Brief:

TL;DR: I am working on an HDC-based system that simulates the human psyche. I noticed that, without any hardcoded instructions, it began to exhibit properties similar to mental disorders—specifically, a "panic attack" was recorded via metric shifts in data tables. Below are the data and some reflections on the subject.

About the system where this occurred

TAEMI is a cognitive architecture based on Hyperdimensional Computing (HDC). All computations are performed via HDC within closed "thought cycles." The system simulates basic human psychological processes that I identified as the foundation of thinking through observations of my own brain. In my view, the "core of cores" in cognition is homeostasis, making it the central link in this experiment. Within the system, I distinguish between "emotions" and "hormones" as two types of regulators: fast (emotions) and slow (hormones). Other technical details are supplementary to the structure and will not be detailed here.

Details on Homeostasis

Homeostasis in the TAEMI system is highly primitive, yet it yielded valuable results. There are five hormones (the first four are involved in this experiment): Dopamine, Cortisol, Serotonin, Adrenaline, and Oxytocin (Oxytocin is excluded here). There are also six basic emotions, with a focus on four: Fear, Anger, Joy, and Sadness. All hormones operate simultaneously, while emotions trigger only when specific conditions are met.

Experimental Conditions and Progression

The conditions were as follows: the system was fed information-dense but identical "spam." The system includes a "boredom" mechanism regarding repetitive data to avoid learning "garbage" information. The more often information repeats, the higher the system's "dissatisfaction." Given the identical spam input, TAEMI increased its Cortisol and Adrenaline levels, reacting to a perceived "danger," while its outcome prediction forecasted only continued spam, further intensifying stress. Below are the tables showing homeostatic dynamics during the experiment.

Table No. 1: General System State

Time (Cycle)	Energy (%)	Integrity (%)	Efficiency	Load (stress)	Status
03:23:47	100.0	100.0	1.00	0.00	Start
03:24:47	97.4	100.0	0.94	0.28	Normal
03:25:47	93.9	99.5	0.88	0.45	Heating
03:26:47	88.5	98.2	0.76	0.72	Warning
03:27:48	79.2	94.5	0.55	0.91	Critical
03:28:18	72.1	91.0	0.42	0.98	Overload
03:28:48	64.8	85.4	0.15	1.00	Shutdown (Force stop)

This table assesses the overall state of the system; more precise metrics are provided below.

The system lost energy and integrity as stress levels rose until it reached a critical threshold and was halted.

Table No. 2: Hormonal Changes

Time	Dopamine (Reward)	Serotonin (Stability)	Cortisol (Stress)	Adrenaline (Unrest)
03:23:47	0.50	0.50	0.10	0.10
03:24:47	0.55	0.52	0.12	0.15
03:25:47	0.62	0.48	0.25	0.22
03:26:47	0.45	0.35	0.55	0.48
03:27:48	0.25	0.20	0.82	0.75
03:28:18	0.15	0.12	0.91	0.88
03:28:48	0.05	0.05	0.98	0.95

The table shows a shift where Cortisol and Adrenaline rise while Dopamine and Serotonin collapse.

Table No. 3: Emotional Changes

Time	Joy	Sadness	Anger	Fear	Dominant Emotion
03:23:47	0.00	0.00	0.00	0.00	Neutral
03:24:47	0.20	0.02	0.00	0.05	Interest
03:25:47	0.35	0.05	0.10	0.15	Curiosity/Anxiety
03:26:47	0.15	0.25	0.30	0.45	Anxiety
03:27:48	0.05	0.45	0.65	0.75	Panic
03:28:18	0.00	0.60	0.80	0.90	Terror
03:28:48	0.00	0.85	0.95	1.00	Total Despair

The dominant emotion evolves from curiosity through anxiety to panic and despair.

Table No. 4: Simulated Sensory Input

Time	'Audio' (dB/Lvl)	'Visual' Intensity	'Temp' (°C/Val)	Pressure	Description
03:24:47	0.15	0.20	0.50	1.00	Silence
03:25:47	0.35	0.45	0.55	1.02	Activity
03:26:47	0.65	0.70	0.68	1.15	Noise
03:27:48	0.85	0.88	0.82	1.35	Overload
03:28:18	0.95	0.95	0.92	1.45	Critical level
03:28:48	1.00	0.00*	0.98	1.50	Sensory shock

TAEMI also simulated sensory perceptions for psychic completeness, though their implementation was extremely primitive in this prototype.

The system entered a state of "sensory shock"—processing of external data virtually shut down (Visual Intensity dropped to zero).

Parallels with Known Mechanisms

The simulation of "boredom" in this cognitive system resonates with Antoni Kępiński’s theory regarding information metabolism—the idea that not all information is beneficial. This is implemented by filtering out excessive repetitions through memory comparison and other mechanisms.

The "panic attack" itself is based on a Positive Feedback Loop: the identical spam input increased stress; the predictions showed only the same spam, increasing stress further, which eventually destroyed the system's balance through infinite reinforcement.

Conclusions

1. Emergent Complexity: Even basic rules set within a system can give rise to unforeseen outcomes. This is simultaneously dangerous due to its inherent unpredictability (at first glance) but also potentially beneficial for autonomous AI agents.
2. The Inevitability of Deviation: This situation suggests that systems of this type face problems that are nearly impossible to "regulate" away, requiring a specific approach to their operation. While tuning homeostasis, I noticed a clear correlation: as the flexibility and efficiency of the system increase, so does the probability of these "psychological" deviations. A balance can be reached, but the risk can never be zero—just as in humans, manifestations like depression or panic attacks cannot be entirely excluded.

Discussion Question:

Do you think systems of this nature can potentially provide more value than fully controlled and predictable variants?

If you have observed similar phenomena in your systems or have thoughts on the balance between autonomy and control, I would love to discuss them in the comments.

Thanks for reading.