claim

active

claim:if-the-universality-hypothesis-is-broadly-true-it-raises-the-exciting-possibility-that-artificial-neural-networks-could-predict-features-previously-unknown-in-biological-neural-networks

If the universality hypothesis is broadly true, it raises the exciting possibility that artificial neural networks could predict features previously unknown in biological neural networks.

Speculative extension of universality to neuroscience, with high-low frequency detectors as a candidate prediction

Source paper

extracted_from

Zoom In: An Introduction to Circuits

(2020) · Chris Olah · Nick Cammarata · Ludwig Schubert · Gabriel Goh +2

Neighborhood — ranked by edge-count

Hypotheses (1)

hypothesis

We hypothesize that high-low frequency detectors, if predicted by artificial neural network universality, might be found in biological neural networks.
supports
Specific cross-domain prediction mentioned by neuroscientists in conversation with the authors

Claims (1)

claim

Analogous features and circuits form across models and tasks.
extends
Third of three speculative claims asserting that learned features are not model-specific but represent universal solutions to learning problems

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

If the Universality Hypothesis is correct, an artificial system trained to perform the same tasks that lead to consciousness in a human infant might exhibit consciousness as well.claim0.849
Paper's extension of Olah's Universality Hypothesis to the domain of consciousness
Superposition hypothesis: neural networks represent more features than dimensions using almost-orthogonal directions.hypothesis0.827
Explanation for why dictionary learning can recover many more features than dimensions.
If we translate the Universality Hypothesis to the problem of consciousness, would it follow that an artificial system trained to perform the same tasks leading to consciousness formation in a human infant would exhibit consciousness as well?question0.821
Key open question linking mechanistic interpretability universality to machine consciousness
We tentatively hypothesize that if an artificial system were trained to perform the same tasks leading to consciousness formation in a human infant, the system would exhibit consciousness as well, by analogy with the Universality Hypothesis.hypothesis0.812
Paper's uncertain extension of mechanistic interpretability universality to consciousness
Neural networks show substantial alignment with biological representations in the brain, driven by shared task and data constraintsclaim0.808
Extends convergence argument to brain-machine alignment
Artificially evolved neural networks and robots often lack modularity unless it is directly selected for, and exhibit inefficiencies from evolutionarily duplicated sub-structures.finding0.798
Evidence that evolved machines share biological property of non-optimal modularity, blurring the distinction
Neural networks, trained with different objectives on different data and modalities, are converging to a shared statistical model of reality in their representation spaces.quote0.797
The paper's central thesis statement, presented prominently after the abstract
Linear representation hypothesis: neural networks represent meaningful concepts as directions in their activation spaces.hypothesis0.789
Foundation for interpreting features as linear directions.