[
next
] [
prev
] [
prev-tail
] [
tail
] [
up
]
Part I
Suffering as error signalling
The first part will explore the very definition of suffering, existing proposals on how suffering comes about, and how these can be understood by the theories of AI and evolution
2
Defining suffering
2.1
Medical definitions of pain
2.2
Medical and psychological definitions suffering
2.3
Ancient philosophical approaches to suffering
2.4
Two main kinds of suffering
2.5
Using the pain system for broadcasting errors
3
Frustration due to failed plan
3.1
Agents, states, and goals
3.2
Planning action sequences, and its great difficulty
3.3
Frustration as not reaching planned goal
3.4
Defining desire as a goal-suggesting mechanism
3.5
Intention as commitment to a goal
3.6
Heuristics can help in planning
4
Machine learning as minimization of errors
4.1
Neurons and neural networks
4.2
Finding the right function by learning
4.3
Learning as incremental minimization of errors
4.4
Gradient optimization vs. evolution
4.5
Learning associations by Hebbian rule
4.6
Logic and symbols as an alternative approach
4.7
Emergence of unexpected behavior
5
Frustration due to reward prediction error
5.1
Maximizing rewards instead of reaching goals
5.2
Learning to plan using state-values and action-values
5.3
Frustration as reward loss and prediction error
5.4
Expectations or predictions are crucial for frustration
5.5
Unexpected implications of state-value computation
5.6
Evolutionary rewards as obsessions
5.7
Reward maximization is insatiable
6
Suffering due to self-needs
6.1
Self as long-term performance evaluation
6.2
Self as self-preservation and survival
6.3
Self-related suffering as intrinsic frustration
7
Threat as anticipation of possible frustration
7.1
Decision-making under uncertainty
7.2
Risk aversion and economic gambles
7.3
Fear, threat, and predictions
7.4
Threat as prediction of possible large frustration
7.5
Interplay of threat and frustration
7.6
Threats and the level of intelligence
8
Fast and slow intelligence and their problems
8.1
Fast and automated vs. slow and deliberative
8.2
Neural network learning is slow, data-hungry, and inflexible
8.3
Using planning and habits together
8.4
Advantages of categories and symbols
8.5
Categorization is fuzzy, uncertain, and arbitrary
9
Summarizing the mechanisms of suffering
9.1
Frustration on different time scales
9.2
Frustration based on desires, expectations, and general errors
9.3
Self, threat, and frustration
9.4
Why there is frustration: Outline of the rest of this book
[
next
] [
prev
] [
prev-tail
] [
front
] [
up
]