Why Emotion Matters

On Affective Neuroscience and AI

What are you feeling right now? Fear? Joy? Maybe you feel hungry. Or maybe you are feeling courageous—having overcome your fears to achieve a difficult goal. Plato referred to these emotions, feelings, desires, will, and passions as hormic or impulsive traits. Many believe the brain is a kind of computer and that emotions are vestigial remnants—like the appendix—made superfluous by big brains and logical reasoning. Shouldn’t reason be the master of unruly and untrustworthy passions? Artificial Intelligence has no need for emotion. Or does it? This post addresses that question.

First Things First

Before we get started, I need to be precise regarding a few terms:

• Emotion is a set of physiological states and with specific locations in the brain. If you stimulate certain emotion centers, you will get an appropriate affect and feeling for that emotion center.

• Affect is a behavioral response to an emotional state. Facial expressions, body language, vocal intonations, heart rate, galvanic skin response, and other non-volitional behaviors are examples of affect. Polygraph examiners, poker players, fortune tellers, and actors make it their business to detect or imitate affect.

• Feeling is the subjective experience of an emotion. Being subjective, scientists have no direct way of measuring someone’s feelings. Animals cannot report on their feelings and humans often have difficulty putting their emotions into words. All feelings have three attributes:

  • A subjective experience that is unique and consistent with emotion and affect. Fear, anger, and familial love are each unique subjective experiences.

  • Valance ranges from negative values (or bad-for-survival) to positive values (or good-for-survival). Sweet berries are good (positive valance), wasp stings are bad (negative).

  • Feelings include a level of intensity or arousal. Blueberries are sweet (good), but honey is sweeter (better). A skinned knee feels bad, but giving birth or passing kidney stones feels worse. Arousal ranges from weak to strong. It measures the meaningfulness of an experience or concept and its likelihood of entering long-term memory. Being told that the Battle of Hastings was in 1066 will not interest most people (weak arousal) and most will forget that fact. But military and English historians may find this more interesting (stronger arousal) and so are more likely to commit the fact to their long-term memory.

• Concepts are activations of multiple, interconnected neurons. Just as there are no solitary neurons, there are no atomic concepts. Imagine grabbing a thread in a spider’s web and pulling it—you end up pulling the threads closest to that thread. That is what the activation of a concept is like. We address the mechanics of semiotics and non-representational meaning in a future post.

• Homeostasis is self-regulation of internal bodily processes. When the core body temperature becomes too high or too low, we sweat or shiver. If blood oxygen is chronically low, the body manufactures more red blood cells. There are many more examples of how our bodies maintain internal equilibrium.

What are emotions good for?

1. Emotions motivate homeostasis. Not all homeostasis is automatic. Sometimes, we have to intervene in a way that restores equilibrium. We get feelings of hunger, thirst, pressure to defecate or urinate, tiredness, extreme temperature, loss of breath, and more. Homeostatic emotions increase feelings of discomfort (increasing arousal of negative valance) until we act to return our internal systems to normal levels. So we inevitably eat, drink, pee, poop, sleep, put on a sweater, move out of the sun, catch our breath, and so on.

2. Emotions manage memory and learning. The brain is not a video recorder with an infinitely large storage capacity¹. Experiences or concepts become part of long-term memory only if they are meaningful to our survival and well-being. Experiences with weak arousal do not get committed to long-term memory.
   
   Picking up a cold iron skillet doesn’t form a persistent memory because it is low arousal and inconsequential to my survival. The memory of the act fades quickly from my episodic memory. However, if I pick up a hot iron skillet with my bare hand, I feel a burning pain. You could say the experience burns itself into my memory because of very strong values of arousal (high importance). Next time I’ll grab a hot pad! Any experience—good or bad—is meaningful and worthy of retaining if it contributes to my future survival. Recall that the primary role of intelligence is to enhance survival through adaptation — see my post What is Natural Intelligence?

3. Emotions prioritize actions. A third function of emotion is to prioritize competing cognitive action paths. Our brain responds to situations with a single behavioral response. But before acting, it first weighs multiple potential responses or action paths—consciously or unconsciously—in a winner-take-all competition. These parallel action paths include processes described as (ordered fast to slow) polysynaptic reflexive, instinctual, habitual, associative learning, predictive learning, reasoned, or some combination². Sometimes time-to-act is more important than action optimization. Facing an armed terrorist or an approaching tornado are situations where time-to-act may be more important than response optimization. In those cases, FEAR may motivate rapid, instinctive behaviors over slower, more reasoned responses.

Evidence of Emotional Memories

A fascinating story from the earliest days of neuroscience suggests two things. First, that emotions exist in memories physically apart from declarative (factual) or episodic (experiential) memory. And second, that emotional memories influence behavior unconsciously.

In 1911, a French physician named Edouard Claparede encountered a female patient with damage to both sides of her hippocampus. This kind of damage makes the consolidation of short-term memory into long-term memory impossible. His patient was incapable of creating any lasting episodic or declarative memories. Every visit Dr. Claparede made to his patient was like his first: he had to introduce himself to her as if it were the first time. Since her procedural memory was intact, she engaged in social rituals and shook hands with Dr. Claparede. One day, Dr. Claparede hid a sharp pin in the palm of his hand. When they shook hands, the pain of the prick startled her, but it was superficial and healed quickly. Because of her condition, she forgot about the incident. Despite this, when Dr. Claparede visited this patient again, she refused to shake hands with him and could not explain her own behavior. She had never been hesitant to shake his hand before. The act of shaking hands previously solicited a response with neutral or positive valence. Yet, now the doctor’s outstretched hand invoked a strong, negative feeling.

There is more recent evidence that emotional memory exists separately from declarative memory, episodic memory, and language.

• By removing the neocortex from rat pups, Panksepp³ showed the neocortex was unnecessary for the expression of primary emotions.

• Merker⁴ showed that human children born without a cerebral cortex could still display a full range of primary emotions.

• Stroke victims who have lost their ability to speak or even think in words maintain their emotional memory. A stroke in the left neocortex can cause this disability.

Neither cognitive ability nor the ability to think in words is necessary for affective behavior or feelings. This is significant because it refutes a popular theory that language and declarative memories are necessary for all emotions. Your dog really loves you... even if Fido cannot express himself in words.

What is Affective Neuroscience?

Affective Neuroscience, a term coined by Jaak Panksepp⁵, applies neuroscience to the study of emotion. Similar to cognitive neuroscience's study of functional brain centers, affective neuroscience identifies emotional brain centers, nerve pathways, and neurotransmitters.

Jaak Panksepp (1943-2017). Photo by Henry Moore Jr., CVM/BCU, Washington State University.

Panksepp identified four forms of emotion: homeostatic, sensory, basic, and constructed.

I. Homeostatic emotions

Homeostatic emotions motivate us to act in order to maintain normal operational levels.

II. Sensory emotions

Sensory emotions originate from sensing our environment. We recognize healthy food by its sweet and savory flavors and smells. We avoid bitter or rotten foods. Pain, loud noises, and rapid motion shock us and alert us to potential threats.

III. Basic or learned emotions

Basic emotions in humans include at least seven primary emotional systems in different locations in the midbrain. Unlike homeostatic and sensory affect, which are entirely innate, basic emotions comprise innate primary emotional systems activated by learned memories.

Panksepp identified seven primary emotional systems⁶ in the rat midbrain. He localized each system. Stimulating a basic emotion system produces behaviors characteristic of that emotion. The first four invoke positive valance. The remaining three invoke negative valance. Panksepp capitalizes the names of these systems to remind us that each has an autonomic, anatomical, behavioral, semantic, and subjective (feeling) aspect.

1. SEEKING: Enthusiastic expectancy, Anticipation, goal-driven motivation

2. CARE: Nurturing & loving

3. PLAY: Joy in social interaction

4. LUST: Sexual excitement, sexual desire

5. FEAR: Anxiety

6. RAGE: Anger

7. GRIEF: Loneliness and sadness

There are likely more emotional centers to be discovered. CURIOSITY may be one⁷.

Most readers will readily identify the feelings associated with these basic emotions…except perhaps SEEKING. The SEEKING emotion drives “wanting” rather than “liking”. It motivates us to climb mountains, do our taxes, or simply to get out of bed in the morning. It is what drives us to attain goals. Panksepp describes SEEKING as a positive emotion, though a goal denied can trigger frustration (RAGE).

Mammals learn most basic emotions, but sometimes they are innate. For example, rats show FEAR when they encounter the smell of cat or ferret fur. Their first response is to freeze still—as if to avoid detection from movement. We know it is innate because dog fur does not solicit the same response. Immature mammals (including human infants) show GRIEF when separated from their parents.

How do experiences get bound to specific basic emotions if they are not hard-wired? How does an experience trigger the correct basic emotion? Imprinting might be one mechanism. Scientists have studied imprinting in birds, but similar processes exist in fish, insects, and mammals. A newborn fixes its attention on the first object it experiences through sight, sound, and touch and becomes emotionally bound to that object. The emotion triggered in this case is CARE. Separation triggers GRIEF⁸.

Whereas homeostatic and sensory emotions are genetically determined, both genes and the environment shape basic emotions. Panksepp believed (as do I) that the neocortex arrives in human infants as a blank slate. The seven (or more) emotional centers listed above serve as building blocks for constructing more semantically complex emotions described next.

IV. Constructed emotions

Panksepp referred to a fourth type of emotion but did not elaborate on it as he did not consider the science mature enough to even speculate. I hold no such reservations. The following is my best guess of this fourth type of emotion.

Direct experience is not the only trigger of emotions. Concepts, activated by reading, speech, or even daydreaming, can trigger emotional responses. Complex concepts⁹ can trigger constructed emotions that activate one or more basic emotional centers. Examples of feelings from constructed emotions include pride, shame, piety, resentment, revenge, schadenfreude, justice, rivalry, and courage.

Transcendental concepts, concepts which do not originate from the senses, can trigger complex, constructed emotions¹⁰. My nation's flag is a visible symbol that represents a transcendental concept (United States of America and everything it implies). When I view the flag and think about what it means, I feel pride. PRIDE (the constructed emotion) combines the basic emotions of CARE (devotion, compassion) with PLAY (social joy). But, if you are a flag burner, then the Stars and Stripes trigger a very different set of emotions. We now turn our attention to another constructed emotion: COURAGE.

What is Courage?

Courage is perseverance toward goal attainment in the presence of danger, difficulty, pain, grief, or self-doubt. Synonyms of courage include bravery, determination, fortitude, resolve, and valor. It is not the same as fearlessness. Courage means we acknowledge fear, but we do not allow it to prevent us from exploring new opportunities, solving hard problems, or doing what is right.

COURAGE is a constructed emotion in which SEEKING triumphs over FEAR. You probably already have a grasp of what FEAR is. SEEKING is the motivation to reach a goal. Each is a basic emotion. When FEAR triumphs over SEEKING, we have HOPELESSNESS. SEEKING and FEAR combined form a constructed emotion that ranges from COURAGE (SEEKING > FEAR) to HOPELESSNESS (SEEKING < FEAR).

If FEAR dominates your behavior, I can tell you to be brave. Or you can read self-help books. But words don't have the power that direct experience and active involvement does. The best thing you can do to overcome FEAR is to leave the environment that reinforces HOPELESSNESS, and go somewhere to learn COURAGE. There is a school for that. That is the topic of the post that arrives after this one.

The Evolution of Emotion

Cognition evolved from simple reflexive processes to innate processes (simple and fixed action patterns), then to associative learning, habits, predictive learning, and reasoning. We carry these processes within us today. Similarly, the four types of emotion — homeostatic, sensory, basic, and constructed — emerged over time in order and remain within us. Child development and the evolving umwelt both recapitulate this evolutionary emotional progression.

Recall that the Animate Domain of the umwelt originated when animals became mobile and could sense predators or prey at a distance (see my post, The Story of Intelligence). FEAR originated in response to danger from predators. Social collaboration between peers benefited from PLAY. Animals born dependent on caring parents for food and protection benefited from CARE. These emotions all originated in the Animate Domain.

Finally, the emergence of transcendental knowledge resulted in emotions that are more complex, more abstract… and occasionally even fanciful or maladaptive. For example, we no longer fear being eaten by a lion. Instead, we fear not getting enough likes on our Facebook page. In a strange, recursive maladaptation, we have become fearful of fear itself. We also get worked up from conspiracy theories.

The evolution of intelligent organisms, human development, and the umwelt each recapitulate the four levels of emotion.

Competing Emotions

If emotions exist to save me from dangerous food and activity, why do I have a passion for hot peppers, smelly cheese, intermittent fasting, horror movies, whitewater kayaking, and roller coasters? Shouldn't I want to avoid those things?

You can find the answer in the question itself: you have a passion. You constructed a new emotion that overrides earlier, innate emotions that guided you as a child. Not everyone likes hot peppers or horror movies, but nearly every infant dislikes hot or bitter flavors, hunger, cold baths, loud noises, surprise, and visceral disorientation. We can learn to like things that repulsed us in our youth because constructed emotions can overrule simpler emotions.

Some of us enjoy hot peppers, scary movies, and roller coaster rides for the same reasons addicts enjoy heroin. Panksepp claims that moviegoers and amusement park visitors know these venues are safe despite causing the release of powerful FEAR hormones. When this happens, we get a shot of hormones into our blood stream that resembles opioid drugs.

I eat hot peppers because I enjoy the rush of endogenous opioids even more than the discomfort of a hot pepper. I also enjoy theater, music, art, and comedy if they trigger emotional responses within me. Elevator music and cheap hotel room artwork fail as art because they don’t trigger an emotional response. Hotel room art is decorative and designed to trigger the weakest possible arousal. Its aim is to be innocuous and unobtrusive.

I practice intermittent fasting because I have a goal of losing weight. Emotion influences the action selection taking place in my brain. It is a continuous tug of war between my desire to eat and my desire to lose weight. Often, desire to eat wins. Will or will-power is the measure of discipline needed for a constructed emotion to prevail over simpler ones.

Constructed emotions can overrule homeostatic, sensory, and basic emotions. In this way, emotions promote behaviors we would have avoided when we were younger. This process of a learned emotion overruling an innate emotion has a parallel in cognitive neuroscience: actions based on learning and reasoning often overrule reflexive and innate responses. But if stress, drugs, or alcohol are present, or if time is of the essence, then the will can diminish and innate responses will take over.

Emotion in Robots

Someday soon, the number of baby boomers in their 80s and 90s will surpass the number of available in-home caregivers. Robots could ease this problem. Would you rather have a robot caregiver programmed at the factory with AI or one that learns and understands your needs, wishes, and feelings? A robot programmed with AI is equivalent to a mechanical sociopath (it does not care). A robot with the CARE emotion cares for you but also learns and adapts to the job.

What if my emotional robot has a mental breakdown and goes on a killing spree? Would you ask the same question about a human caregiver whose history you know much less about? We like to dwell on the negative aspects of emotional dysfunction and ignore the benefits of healthy emotion: learning and wise judgment. AI does not learn (see my post, AI Does Not Learn) nor does AI show judgment.

Besides, consider that these are robots. Because robots have replaceable parts and archivable memories, they have no need to fear death. Robots do not need sexual reproduction, so LUST and gender becomes unnecessary. That alone prevents a lot of courtship display and dominant male behaviors. I can see no reason an in-home care giving robot needs RAGE—so rule out a killing spree. A robot’s umwelt is going to differ from a human umwelt. It makes no sense to duplicate the same emotion centers into robots as are found in humans. SEEKING, PLAY, CARE, and GRIEF should be sufficient for a care giving robot.

I am not suggesting we program emotional intelligence into robots. I did that successfully in a commercial chatbot 25 years ago. My post, The Illusion of Intelligence, describes my emotionally intelligent chatbot. So called emotional intelligence simulates emotion...it is not authentic. Instead, I propose we recreate the same natural intelligence found in animals into robots…it will include genuine emotion. Doing so requires profound changes in AI and the cognitive sciences.

What is Artificial General Intelligence Missing?

Sam Altman of OpenAI and others believe Artificial Intelligence (AI) will soon reach human-level performance. Human-level AI is called Artificial General Intelligence or AGI. Altman and others believe that AI only needs scaling to achieve AGI: larger neural networks and larger training sets.

AGI-seekers are misguided. We can never reach a human-level of cognition in computers until we understand human-level cognition in humans. And we cannot hope to understand human cognition without also understanding emotion. This includes explaining what might be called hormic events: arts, music, theater, comedy, sporting events, amusement parks, social media, travel, hunting, and church. These venues exist to stimulate our emotions.

AI is a wonderful technology, but it doesn’t work like animal cognition. Today’s AI lacks essential elements to achieve human-level performance. Autonomous learning and emotion are two of those elements. I have already written how "machine learning" and "supervised learning" in computers are not learning at all (see my post AI Does Not Learn). The recognition of emotion as a fundamental component of cognition is long overdue.

Some readers might think that reinforcement learning (RL) is the future of AGI and that RL’s reward function resembles a simple emotion. By now, they should realize that ‘simple’ is overly generous. Unlike RL’s fixed reward function, emotions are dynamic. They can redefine themselves in service to survival.

Summary

• Emotion is a critical component of homeostasis, learning, memory management, and action selection. Without some form of emotion, computers will remain dependent on human programmers for knowledge and will never attain the bootstrapped learning of humans.

• Jaak Panksepp's ground-breaking work in Affective Neuroscience provides the only model I know that makes sense in neuroscience, animal behavior, semiotics, and evolutionary biology.

• Courage is an emotion constructed from basic emotion centers FEAR and SEEKING.

• Willpower is the ability of a constructed emotion to overrule a simpler emotion. In this way, emotions invent themselves and replace less adaptive emotions.

• Feelings are subjective experiences that defy direct scientific measurement. This makes them fodder for speculation on consciousness, spiritualism, and mind-body dualism. For my perspective on consciousness, see my post The End of Consciousnessism.

Jaak Panksepp influenced the model of emotion described in this post. His model is not the only one, though I believe it to be the best. For a recent comparison of the leading three models of emotion, see the 2022 book, A Short-Cut to Understanding Affective Neuroscience by Lucy Biven. An excellent source for affective neuroscience is Jaak Panksepp’s 1998 text, Affective neuroscience: The foundations of human and animal emotions.

---

Please subscribe to Intelligence Evolved (it’s free!) to receive notifications about the next post, Courage. I will describe teaching courage from my perspective as a former instructor of outdoor experiential education programs.

1. There are persons with savant syndrome that have exceptional memory recall, but they also have significant social or intellectual impairment. See Wikipedia for Savant syndrome.

2. My forthcoming book describes each of these modes of cognition in greater detail.

3. Panksepp J., Normansell L., Cox J. F. Siviy S. M. (1994). Effects of neonatal decortication on the social play of juvenile rats. Physiology & Behavior, 56, 429–443. 10.1016/0031-9384(94)90285-2.

4. Merker B. (2007). Consciousness without a cerebral cortex: A challenge for neuroscience and medicine. The Behavioral and Brain Sciences, 30, 63–81. 10.1017/S0140525X07000891.

5. Panksepp J. (1998). Affective neuroscience: The foundations of human and animal emotions. New York, NY: Oxford University Press.

For a summary of Jaak Panksepp's scientific career, see Davis KL, Montag C. Selected Principles of Pankseppian Affective Neuroscience. Front Neurosci. 2019 Jan 17;12:1025. doi: 10.3389/fnins.2018.01025. PMID: 30705615; PMCID: PMC6344464.

6. Ibid.

7. Ahmadlou, Mehran, Janou H. W. Houba, Jacqueline F. M. van Vierbergen, Maria Giannouli, Geoffrey Alexander Gimenez, Christiaan van Weeghel, Maryam Darbanfouladi, et al. “A Cell Type–Specific Cortico Subcortical Brain Circuit for Investigatory and Novelty-Seeking Behavior.” Science 372, no. 6543 (May 14, 2021). https://doi.org/10.1126/science.abe9681.

8. To learn about fascinating and disturbing experiments on imprinting in rhesus monkeys, see the Wikipedia entry on Harry Harlow.

Harry Harlow Monkey Experiments: Cloth Mother vs Wire Mother

Harlow, H. F., Dodsworth, R. O., & Harlow, M. K. (1965). Total social isolation in monkeys. Proceedings of the National Academy of Sciences of the United States of America, 54 (1), 90.

Harlow, H. F. & Zimmermann, R. R. (1958). The development of affective responsiveness in infant monkeys. Proceedings of the American Philosophical Society, 102,501 -509.

9. Panksepp refers to these as "projected emotions". I use "constructed" to avoid confusion with another psychological term for a type of defense mechanism.

10. For background on transcendental concepts, see my post The Story of Intelligence - Part One.