The shrinking never-list

Predictions about computer limits often expire. People once said computers would not read handwriting, beat elite chess players, or produce official-sounding letters. In 1997, IBM’s Deep Blue defeated Garry Kasparov in a match. Machine play later reached extraordinary levels in Go. Software can now draft the kind of letter that Harborline handles.

A never-list is a list of tasks people claim computers will never perform well. The useful lesson is not that every task should be handed to software. It is that a task moving off the never-list tells us to examine the work now, rather than relying on an old prediction.

Performing part of a task is different from holding a role. A chess game has defined rules, a board position, and a score. A housing decision requires evidence review, an explanation, an appeal path, and someone accountable for the decision. Producing a convincing paragraph meets none of those requirements by itself.

Harborline does not need to wait for another famous machine victory. It can map its current claims process: which steps are routine, which files need review, and who signs the final decision. That map is more useful than a prediction about whether software will improve again.

The next page turns that map into a working method for people and machines.