To learn a new thing can be to lose an old one. A system trained on a fresh task may overwrite the very weights that held its former skill. Memory and plasticity war within the same parameters.
The claim
Catastrophic forgetting is the abrupt loss of prior competence when a learner acquires something new. The knowledge is not gently revised but erased, as new demands reshape shared parameters. What was known simply vanishes.
Beneath the surface
The mechanism is interference. When the same parameters must serve old and new tasks, optimizing for the latter can destroy the settings the former relied on. There is no separate shelf on which to keep the past.
A reframing
The tension is between stability and plasticity. A system rigid enough to remember struggles to learn; a system flexible enough to learn struggles to remember. The two virtues pull against each other.
The trade-off
Remedies are partial. One can protect important weights, rehearse old data, or grow new capacity, but each buys memory at some cost. No method fully reconciles the demands.
Where it breaks
The implication is that continual learning is unnatural for these systems. Minds that learn across a lifetime solve a problem our artifacts find hard. Sequential experience is a challenge, not a given.
The larger point
Catastrophic forgetting is the erasure of old skill by new learning in shared parameters. It exposes the deep tension between stability and plasticity. Learning continuously, which minds do easily, remains hard to engineer. The principle rewards the patience to state it precisely and the humility to mark its limits. Precision reveals what it truly claims; humility reveals where it quietly fails. Between these two disciplines lies genuine understanding, which is never the possession of a conclusion but the grasp of why the conclusion holds and exactly how far it reaches.
