Our summary
A cellular automaton is a grid whose cells update using nearby cells and a fixed rule. Justin Hedge studies a small moving pattern in such a grid. After one cell is removed, a change to one local rule makes the next whole grid match the undamaged version.
The same three-by-three view needs one answer on the healthy path and the opposite answer after damage. A rule that remembers nothing beyond that view cannot represent both histories. The study also replays a saved Jev answer table, without making fresh calls while the grid runs.
The saved table matches 231 of 512 grid cases; separate tests supplied with cell counts match eighteen of eighteen. Those are distinct tests of a recording, not current model quality. A repair mode given the target is an upper bound, not learned memory or evidence of biological regeneration.
Key takeaways
- Check whether identical local inputs need different outputs before expecting a memoryless rule to solve the task.
- Keep saved model answers and new model tests separate when reporting results.
- A method handed the desired target shows what extra information can buy, not that the system discovered memory.