Most systems are built to remember everything. Storage is cheap. Logs are infinite. Nothing needs to be thrown away.

Humans don’t work like that. Forgetting isn’t a failure of memory. It’s how memory stays usable. If everything were equally accessible, nothing would matter.

We forget details so patterns can survive. We forget instances so categories can form. Memory compresses itself by discarding specificity. What remains is not accuracy, but direction.

AI systems treat forgetting as an error state. When something is lost, it’s framed as degradation. So we overcorrect. We archive. We snapshot. We retrieve context long past its relevance.

The result is not intelligence. It’s clutter.

Long-term memory isn’t about recall. It’s about selection. About deciding what should continue to influence future decisions and what should quietly decay.

A system that never forgets becomes brittle. It overfits to its own history. It mistakes noise for signal.

Humans revise memory constantly. We overwrite. We simplify. We lie to ourselves a little. Not because we’re irrational, but because we’re adaptive.

Memory is shaped by usefulness, not truth.

Intelligence requires decay. Without it, systems accumulate weight they can’t carry.

The question isn’t how much a system can remember.

It’s what it chooses to forget.