catastrophic forgetting
The tendency of neural networks to lose previously learned information when trained on new tasks, once seen as a fundamental limitation.
As defined in Carl Shulman (Pt 1) — Intelligence explosion, primate evolution, robot doublings, & alignment
Other definitions
The problem that occurs when learning new information destroys memory of previously learned patterns; happens in simple learning systems like competitive learning that lack protective mechanisms for stable memories
How Does Each Brain Make A Mind? Adaptive Resonance Theory (ART) & Consciousness | Stephen GrossbergThe problem in AI systems where learning new information causes the system to lose previously learned knowledge. A key challenge in continual learning where adding new training destabilizes or overwrites old learning.
Demis Hassabis: Why AGI is Bigger than the Industrial Revolution & Where Are The Bottlenecks in AI
Where it comes up · 8
- How Does Each Brain Make A Mind? Adaptive Resonance Theory (ART) & Consciousness | Stephen Grossberg3 claims
- Rich Sutton, The OaK Architecture: A Vision of SuperIntelligence from Experience - RLC 20252 claims
- Carl Shulman (Pt 1) — Intelligence explosion, primate evolution, robot doublings, & alignment1 claim
- Dwarkesh Patel Podcast - The next big breakthrough will be AIs learning on the job1 claim
- AI: Grappling with a New Kind of Intelligence | World Science Festival1 claim
- 153 - Possible Minds1 claim
- Ai Sota #20261 claim
- Charan Ranganath1 claim
Most important claims · 10 of 11Search all claims for “catastrophic forgetting” →
- Most of my work emerged from analyzing self-organization in both development and learning, particularly solving the stability-plasticity dilemma—how we learn quickly without catastrophic forgetting while also maintaining stable memories—which I addressed through Adaptive Resonance Theory (ART) introduced in 1976.How Does Each Brain Make A Mind? Adaptive Resonance Theory (ART) & Consciousness | Stephen Grossberg
- Human children are the best example we have of a system that truly works as an intelligence, generalizing from comparatively small amounts of data using innate structure, whereas current AI relies on statistical inference from huge datasets, is fooled by adversarial and superficial features, and suffers catastrophic forgetting when rules change slightly, so nothing we have is even in the ballpark of a four-year-old.153 - Possible Minds
- Mid-training uses the same algorithm as pre-training but is more specialized—e.g. focusing on long-context documents—and is positioned so the quality content is the last thing the model sees, because neural networks suffer catastrophic forgetting where learning something causes other things to be forgotten.Ai Sota #2026
- GPT-4 cannot plan in the mathematical sense, but it cannot because it is frozen in time after training; in principle, continuous learning could address this limitation, though we currently do not know how to implement continuous online learning without catastrophic forgetting.AI: Grappling with a New Kind of Intelligence | World Science Festival
- Every animal or organism that can autonomously solve the stability-plasticity dilemma has a form of consciousness—consciousness is not unique to humans but is a fundamental property of any adaptive system that can learn and predict without catastrophic forgetting.How Does Each Brain Make A Mind? Adaptive Resonance Theory (ART) & Consciousness | Stephen Grossberg
- RL concentrates gradient updates only on what is relevant to getting the outcome right, making its updates sparse, which matters for continual learning because on-the-job learning must not overwrite everything else the base model knows.Dwarkesh Patel Podcast - The next big breakthrough will be AIs learning on the job
- Competitive learning shows that humans and various mammals learn recognition categories (cell populations that fire selectively to spatial patterns of activation across feature-selective neurons), but experiences catastrophic forgetting when input patterns are numerous, statistically drifting, or overlapping—solving this requires the addition of top-down learned expectations to prevent catastrophic forgetting.How Does Each Brain Make A Mind? Adaptive Resonance Theory (ART) & Consciousness | Stephen Grossberg
- Current deep learning methods do not support reliable continual learning—despite decades of work on catastrophic forgetting and recent work on catastrophic loss of plasticity—and this is a major limitation of applying runtime learning to complex real-world problemsRich Sutton, The OaK Architecture: A Vision of SuperIntelligence from Experience - RLC 2025
- Human intelligence is powered by having both a 'dumb' episodic memory that records that something happened once at a point in time, and accumulated semantic knowledge; when a surprising situation violates expectations, an episodic memory is formed that can later supplement knowledge to reason about novel cases, letting us stop on a dime and change without erasing everything learned — a powerful solution to Grossberg's Stability-Plasticity Dilemma that lets us learn from far less data.Charan Ranganath
- The OAK architecture requires solving reliable continual learning for deep neural networks, which currently fails due to catastrophic forgetting and loss of plasticity, with some solution approaches including continual backprop and meta-learning of new featuresRich Sutton, The OaK Architecture: A Vision of SuperIntelligence from Experience - RLC 2025