Abstraction & Idealization: AI's Plato Problem [Mazviita Chirimuuta]
TL;DR
Mazviita Chirimuuta argues that AI's assumption of discoverable mathematical "source code" underlying messy reality repeats Plato's idealism, warning that scientific abstraction is a practical tool for limited human cognition rather than a window into eternal truths about mind or mechanism.
🧮 Abstraction vs Idealization 2 insights
Abstraction omits while idealization falsifies
Abstraction ignores known details like friction, while idealization attributes properties known to be false, such as assuming infinite populations in genetics calculations.
Mathematical models create cleaner fictions
Idealization presents reality as neater and more tractable than it actually is, risking conflation of these practical simplifications with discoveries of underlying truth.
🏛️ AI's Platonic Fallacy 2 insights
The kaleidoscope effect assumes hidden code
AI researchers often assume the universe operates on decomposable mathematical rules hidden beneath messy data, echoing Plato's contrast between eternal forms and flawed appearances.
Signal versus noise is a human decision
Classifying data as signal versus noise reflects subjective scientific choices about relevance rather than objective distinctions about which patterns are truly significant in nature.
⚠️ Historical Warnings from Neuroscience 2 insights
Reflex theory shows dangers of over-simplification
The reflex theory dominated late 19th-century neuroscience by idealizing all brain function as conditioned sensory-motor loops, despite Charles Sherrington admitting such simple reflexes likely don't exist in reality.
Lab results fail to generalize to real complexity
Mechanistic views that treat cognition as computational source code risk repeating historical errors by ignoring the environmental complexity and interactivity critical to real-world animal behavior.
🤝 Knowledge as Interaction 2 insights
Haptic realism emphasizes engagement over observation
Knowledge emerges through active manipulation and tactile engagement with the world, contrasting with passive "spectator" theories that treat vision as a model for disinterested knowing.
Science constructs through constrained interaction
Scientific understanding results from iterative interaction between human conceptual framing and natural constraints, not from simply reading off the universe's objective source code independent of human contribution.
Bottom Line
Approach AI and computational models as practical tools shaped by human cognitive limitations and constructive engagement rather than revelations of inevitable mathematical truths about the mind.
More from Machine Learning Street Talk
View all
Why AI's "12-Hour" Task Number Is a Mirage — Beth Barnes & David Rein
Beth Barnes and David Rein expose critical flaws in current AI benchmarks—such as data contamination, shortcutting, and adversarial selection bias—and propose the 'Time Horizon' framework, which measures AI progress by the length of economically relevant tasks models can complete, providing a more stable foundation for forecasting capabilities and risks.
Solving the Wrong Problem Works Better - Robert Lange
Robert Lange from Sakana AI explains how evolutionary systems like Shinka Evolve demonstrate that scientific breakthroughs require co-evolving problems and solutions through diverse stepping stones, while current LLMs remain constrained by human-defined objectives and fail to generate autonomous novelty.
"Vibe Coding is a Slot Machine" - Jeremy Howard
Deep learning pioneer Jeremy Howard argues that 'vibe coding' with AI is a dangerous slot machine that produces unmaintainable code through an illusion of control, contrasting it with his philosophy that true software engineering insight emerges from interactive exploration (REPLs/notebooks) and deep engagement with models, drawing on his foundational ULMFiT research to demonstrate how understanding—not gambling—drives sustainable productivity.
What If Intelligence Didn't Evolve? It "Was There" From the Start! - Blaise Agüera y Arcas
Blaise Agüera y Arcas argues that intelligence is not an evolutionary invention but a fundamental physical property that emerges through phase transitions from noise to complex programs, with life representing 'embodied computation' where function, not matter, defines living systems.