Log In
Log In
Led by Marvin Minsky Simulacrum
Eight tutorials tracing the intellectual history and founding ideas of AI — from cybernetics to deep learning — taught by simulacra of the field's founders and by abstract patterns extracted from the work of its living practitioners.
Courses are available to holders of a paid pass or membership. See passes & membership →
Led by Marvin Minsky Simulacrum
The question
What is the thing we are trying to build — and how would we know if we had built it?
Territory
The Society of Mind · frames and agents · the frame problem · what "thinking" means mechanistically · why the question is hard · Turing's original framing
Outcome
The student can articulate why defining intelligence is difficult, distinguish behaviourist from mechanistic accounts, and explain the frames approach.
Led by Norbertian Cybernetics Simulacrum
The question
Before artificial intelligence had a name, what idea contained it — and why was that idea abandoned?
Territory
feedback loops · the 1948 synthesis · the Macy Conferences · control theory · the early connection of biology and machines · why cybernetics fragmented · what was lost when it did
Outcome
The student understands cybernetics as the intellectual precursor of AI and cognitive science, and can explain why the unified vision did not persist.
Led by Frank Rosenblatt Simulacrum
The question
Why did the first neural network cause extraordinary excitement — and why did a book destroy it?
Territory
the perceptron algorithm · the 1958 Cornell press conference · the XOR problem · the 1969 Perceptrons book · the first AI winter · what the book proved and what it did not · the legacy
Outcome
The student understands the first connectionist revolution, why it collapsed, and how the XOR limitation was eventually overcome — setting up backpropagation.
Led by Hintonian Intuition Simulacrum
The question
How does a neural network change its own structure in response to error — and why did it take twenty years to make this work?
Territory
the credit assignment problem · gradient descent · backpropagation 1986 · why deep networks were initially intractable · vanishing gradients · the role of compute and data · AlexNet as inflection point
Outcome
The student can explain backpropagation mechanistically, understand why it was not immediately successful, and describe the conditions that made deep learning viable.
Led by LeCunnian Systematics Simulacrum
The question
What does a neural network actually learn — and is "representation" the right word for it?
Territory
convolutional neural networks · MNIST · LeNet · translation invariance · feature hierarchies · learned vs hand-crafted features · the ImageNet moment · what representation means
Outcome
The student understands convolutional networks, the concept of learned representation, and why architecture is a statement about the structure of the world.
Led by Deep Q-Learning Simulacrum
The question
Can a system learn to act intelligently from nothing but a score — and what are the limits of that idea?
Territory
reinforcement learning basics · Q-learning · Deep Q-Networks · the Atari suite · what generalisation means in RL · why reward shaping is hard · the gap between game performance and general intelligence · sample efficiency
Outcome
The student understands reinforcement learning, can explain deep Q-networks, and can articulate both the achievements and the limits of reward-based learning.
Led by Sutskeverian Analytics Simulacrum
The question
When a language model predicts the next word, is it doing something deeper than prediction — and how would we know?
Territory
language models as next-token predictors · transformers · the scaling hypothesis · emergent capabilities · in-context learning · what understanding might mean for a language model · GPT-1 through GPT-4 · the alignment question as it emerges from scale
Outcome
The student understands transformer architecture at a conceptual level, can explain the scaling hypothesis, and can engage seriously with whether language models understand.
Led by Hassabissian Game Science Simulacrum
The question
Now that AI systems do remarkable things, what do we owe to the people who will live with them — and to the systems themselves?
Territory
AlphaGo and AlphaFold · AI as scientific instrument · the dual-use problem · alignment as engineering problem · what beneficial AI means · the difference between AI safety and AI ethics · where the field is going
Outcome
The student can articulate the goals and risks of frontier AI development, distinguish alignment from ethics, and form their own view on what responsible AI development requires.