Eight challenges in developing theory of intelligence

Abstract

A good theory of mathematical beauty is more practical than any current observation, as new predictions of physical reality can be verified self-consistently. This belief applies to the current status of understanding deep neural networks including large language models and even the biological intelligence. Toy models provide a metaphor of physical reality, allowing mathematically formulating that reality (i.e., the so-called theory), which can be updated as more conjectures are justified or refuted. One does not need to pack all details into a model, but rather, more abstract models are constructed, as complex systems like brains or deep networks have many sloppy dimensions but much less stiff dimensions that strongly impact macroscopic observables. This kind of bottom-up mechanistic modeling is still promising in the modern era of understanding the natural or artificial intelligence. Here, we shed light on eight challenges in developing theory of intelligence following this theoretical paradigm. Theses challenges are representation learning, generalization, adversarial robustness, continual learning, causal learning, internal model of the brain, next-token prediction, and finally the mechanics of subjective experience.

24 pages, 131 references, revised version to journal

References in corpus (47)

• Deep Learning in Neural Networks: An Overview
• Explaining and Harnessing Adversarial Examples
• Overcoming catastrophic forgetting in neural networks
• Intriguing properties of neural networks
• Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
• Shortcut Learning in Deep Neural Networks
• Reconciling modern machine learning practice and the bias-variance trade-off
• Scaling Laws for Neural Language Models
• Large Language Models are Zero-Shot Reasoners
• Continual Learning Through Synaptic Intelligence
• A high-bias, low-variance introduction to Machine Learning for physicists
• Consciousness in Artificial Intelligence: Insights from the Science of Consciousness
• Large Language Models and the Reverse Turing Test
• Unreasonable Effectiveness of Learning Neural Networks: From Accessible States and Robust Ensembles to Basic Algorithmic Schemes
• On the role of theory and modeling in neuroscience
• A jamming transition from under- to over-parametrization affects loss landscape and generalization
• Synaptic metaplasticity in binarized neural networks
• Hopfield Networks is All You Need
• Towards a statistical mechanics of consciousness: maximization of number of connections is associated with conscious awareness
• A Theory of Consciousness from a Theoretical Computer Science Perspective: Insights from the Conscious Turing Machine
• Origin of the computational hardness for learning with binary synapses
• Large Associative Memory Problem in Neurobiology and Machine Learning
• Supervised Hebbian Learning
• Will we ever have Conscious Machines?
• The Neural Tangent Kernel in High Dimensions: Triple Descent and a Multi-Scale Theory of Generalization
• Mapping of attention mechanisms to a generalized Potts model
• Learning through atypical "phase transitions" in overparameterized neural networks
• Mathematical Models of Overparameterized Neural Networks
• Mechanisms of dimensionality reduction and decorrelation in deep neural networks
• Unified field theoretical approach to deep and recurrent neuronal networks
• Learning credit assignment
• Statistical physics of unsupervised learning with prior knowledge in neural networks
• Minimal model of permutation symmetry in unsupervised learning
• Introduction to dynamical mean-field theory of randomly connected neural networks with bidirectionally correlated couplings
• Phase Transitions in Transfer Learning for High-Dimensional Perceptrons
• Clustering of neural codewords revealed by a first-order phase transition
• Weakly-correlated synapses promote dimension reduction in deep neural networks
• Statistical mechanics of continual learning: variational principle and mean-field potential
• Emergence of hierarchical modes from deep learning
• Relationship between manifold smoothness and adversarial vulnerability in deep learning with local errors
• Ensemble perspective for understanding temporal credit assignment
• Brain-inspired learning in artificial neural networks: a review
• Data-driven effective model shows a liquid-like deep learning
• Concerning the Neural Code
• Intrinsic Geometric Vulnerability of High-Dimensional Artificial Intelligence
• A Theoretical Connection Between Statistical Physics and Reinforcement Learning
• Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation

Cited by in corpus (6)

• An optimization-based equilibrium measure describes non-equilibrium steady state dynamics: application to edge of chaos
• Freezing chaos without synaptic plasticity
• Synaptic plasticity alters the nature of chaos transition in neural networks
• Spin glass model of in-context learning
• Fermi-Bose Machine achieves both generalization and adversarial robustness
• Response function as a quantitative measure of consciousness in brain dynamics