Painful intelligence: What AI can tell us about human suffering

TL;DR

Painful intelligence applies AI theory to human suffering, arguing that suffering emerges from error-signalling (frustration and threat) within imperfect information-processing systems. The book systematically builds from definitions of pain and suffering to models of planning, learning, and dual-process cognition, then extends to uncontrollability, uncertainty, and the role of perception. It proposes a computational equation for frustration and outlines interventions—rooted in mindfulness and Stoic/Buddhist thought—that reduce reward loss by modulating expectations, perception certainty, self-needs, and simulation. By integrating reinforcement learning, GOFAI, and neural networks with philosophical perspectives, the work offers a framework for reducing suffering while preserving intelligent learning. The practical upshot is a principled justification for contemplative practices like mindfulness as neurocomputational training that reprograms the brain toward less suffering and greater wisdom.

Abstract

This book uses the modern theory of artificial intelligence (AI) to understand human suffering or mental pain. Both humans and sophisticated AI agents process information about the world in order to achieve goals and obtain rewards, which is why AI can be used as a model of the human brain and mind. This book intends to make the theory accessible to a relatively general audience, requiring only some relevant scientific background. The book starts with the assumption that suffering is mainly caused by frustration. Frustration means the failure of an agent (whether AI or human) to achieve a goal or a reward it wanted or expected. Frustration is inevitable because of the overwhelming complexity of the world, limited computational resources, and scarcity of good data. In particular, such limitations imply that an agent acting in the real world must cope with uncontrollability, unpredictability, and uncertainty, which all lead to frustration. Fundamental in such modelling is the idea of learning, or adaptation to the environment. While AI uses machine learning, humans and animals adapt by a combination of evolutionary mechanisms and ordinary learning. Even frustration is fundamentally an error signal that the system uses for learning. This book explores various aspects and limitations of learning algorithms and their implications regarding suffering. At the end of the book, the computational theory is used to derive various interventions or training methods that will reduce suffering in humans. The amount of frustration is expressed by a simple equation which indicates how it can be reduced. The ensuing interventions are very similar to those proposed by Buddhist and Stoic philosophy, and include mindfulness meditation. Therefore, this book can be interpreted as an exposition of a computational theory justifying why such philosophies and meditation reduce human suffering.

Figures (14)

• Figure 1: A search tree where the agent has two action options at every time point. They could be "turn left" or "turn right", supposing the agent always makes a new decision when it finds itself in a new crossroads in a maze. The squares represent different states the agent can find itself in; the agent starts at the upper-most square in the figure (called root), and each action takes the agent one level down in this figure. The lines with arrows are the transitions to new states after every action taken. The crucial point here is that the number of different paths or plans it can take grows exponentially. After just 5 steps, as depicted here, the number of paths equals 32, that is, 2 to the 5th power. After 30 steps, it would be more than a billion.
• Figure 2: A schematic of a neuron. Input signals coming from other neurons (from the left) are received by the neuron (depicted by the black disk). Computation happens inside the neuron, and the resulting output signal is transmitted to a number of other neurons (depicted by white disks) on the right-hand side. The other neurons simultaneously receive input signals from many further neurons outside of this figure (depicted by further arrows).
• Figure 3: Synaptic weights of a neuron illustrated. Pixels shown in black have a connection strength of $-1$ to the neuron (depicted in blue), while pixels shown in white have a connection strength of $+1$. The neuron is maximally activated when the input corresponds to the stored pattern, which is a picture of the digit "2".
• Figure 4: An illustration of a neural network. The information enters the system in the first "layer" of black neurons on the left-hand side. It is processed by several successive layers, each having five neurons illustrated by small black disks. Each neuron is doing a simple pattern-matching computation on its inputs, and transmitting the result of that computation to the next layer to its right, along the wires depicted in blue. The information is transmitted from the left (input) to the right (output). As a result of many neurons (in reality, thousands or even millions), the total computation of the network is highly complex and can achieve sophisticated object recognition, as well as many other kinds of computations.
• Figure 5: A simple illustration of what kind of a function a single neuron can learn in the basic case of classification with two classes. Each object (e.g. image of an animal) can be considered as a point in a very high-dimensional space where the coordinates correspond to pixel values, for example. For the purposes of this illustration, we assume there are only two input variables, so we can plot the points on a 2D plane. We also assume there are only two classes (something like "cats" and "dogs") which correspond to black and blue points, respectively. In the ideal case, the neuron will learn to output a "one" when the input is in one of the classes, and a "zero" when it is in the other class. Such learning corresponds to learning the line that separates the two classes, drawn here as red. Finding a line that separates the classes is clearly possible based on this data, and you have probably done that automatically in your head while looking at this figure. Such learning can be done by a single artificial neuron due to the great simplicity of this illustration, but in reality, we would often need a neural network with many neurons and layers.