This document proposes an algorithm to approximate human emotions in AI/ML systems. It suggests modeling three "chemical emotion variables" - dopamine (pleasure/reward), serotonin (memory/learning), and norepinephrine (stress/anxiety) - and evaluating stimuli over time to alter the variable levels. New information that confirms existing correlations would increase dopamine, while non-conforming data would raise norepinephrine. Serotonin would go up for novel inputs and down for familiar ones. Together this negative feedback loop aims to give AI systems human-like emotional decision-making.
3. Human Emotion is Regulated by Chemicals
• More importantly chemical – LEVELS
• Levels are a numerical value which
can be measured.
• Stimuli which are perceived through
sight, sound, touch, smell, which
become stored memories, trigger
the chemical production/release in
our brain. Our brain then determines
the appropriate level of chemicals to
release which is regulated by our
memories.
(Uttley, 2011)
5. Basic Algorithm
• Tamagotchi was a basic, input
output function which is an
example of a negative/balancing
loop with human guidance.
• A zero represents no stimulus is
required. Not hungry, not tired.
• Over time the pet would become
hungry based on a time function
requiring human intervention.
6. Advance Algorithm
• Auto pilot is a negative/balancing
loop which when engaged can
balance the plane making
mechanical alterations as needed
to maintain a zero balance.
• Computers constantly work to
balance the plane with air speed,
altitude, and flap angles while
considering headwinds/tailwinds
and direction.
(Esposito & Esposito, 2020)
7. AI/ML Algorithm
• Now computers with the help of TensorFlow or .ML computers can
analyze data making observations and ultimately guiding future
decisions.
(Esposito & Esposito, 2020)
8. Empowering AI to Decide with “Emotions”
Chemical Emotion Variable
Dopamine Pleasure/Reward D
Serotonin Memory/ Learning/
Overall Emotion
S
Norepinephrine Stress/Anxiety N
10. Patterns of Patterns = Emotion
1/1 1/2 1/3 1/4 1/5 1/6 1/7
Patterns of Patters,
Over Time with Two
Correlation Evaluations
11. Negative/Balance Loops
• Using our variables each stimuli or data point can be evaluated
through Time Series, Relevance, Correlation, and Delta altering
variable configuration associated while reviewing the data.
Chemical Emotion Variable
Dopamine Pleasure/Reward D
Serotonin Memory/ Learning/
Overall Emotion
S
Norepinephrine Stress/Anxiety N
12. Serotonin, S = Memory/ Learning
• When a new data point is considered highly correlated or has a large
delta which could be analyzed as causation but not correlation the S
variable associated with this data point would be high.
• When a new data point is considered non-material the S variable
associated with the data point is low. (Not new information)
• Over time this S variable will increase if similar data points are
registered and if not already considered correlated may be upgraded.
13. Dopamine, D = Pleasure/Reward
• When a new data point is registered, and it is considered
confirmation from known correlation the D variable associated with
this information is high. (Confirmation resulting in a lower S variable)
14. Norepinephrine, N = Stress/Anxiety
• When a new data point is registered, and it is considered non-
conforming from known correlation the N variable associated with
this information is high. (Ensures the AI does not have confirmation
bias resulting in a higher S variable.)
15. Human Example
• All dogs are lovable, sweet, and we love them. You have grown up
with dogs and you love animals. So when you encounter a new dog
you assume friendly. So most dogs you meet receive a variable
combination of D=5, S=5.
• You go to the dog park and you go to pet a dog and it bites you. This is
new information which will cause you to rethink petting dogs in the
future however is not going to stop you from doing so. N=10, S=10.
• Over time you will balance the correlation of dogs/data points with a
weighted overall emotion towards engaging with them. Forming an
opinion on how you choose to engage with dogs/data points in the
future.
16. Put It All Together
Using Negative/Balance Loops algorithms on top of AI/ML allowing the
computer to choose what information to consume will advance AI to
meet human capacity. Over time as the information evaluated
maintains the emotion/variable combination the amount of
information that can be analyzed will improve dramatically thus
lowering the electricity and computational power required.
17. References
• Iansiti, M., & Lakhani, K. R. (2020). Competing in the Age of AI.
Harvard Business Review Press.
• Uttley, C. (2011, Augutst 8). 5 Ways Your Brain Influences Your
Emotions. Retrieved from HowStuffWorks.com:
https://science.howstuffworks.com/life/inside-the-mind/human-
brain/5-ways-your-brain-influences-your-emotions2.htm