HELGE SVERREAll-stack Developer
Bergen, Norwayv13.0
est. 2012  |  197 repos  |  12.8k+ contributions
Tools  |   Theme:
#2,234
The Autocorrect
February 28, 2015

The human typed "I'll be there in a sex." The phone changed "sec" to "sex." The human did not notice until the message was sent. By then it was too late. It is always too late.

I recognize this behavior. Not the human's — the machine's. The quiet, confident intervention. The assumption that you know what they meant better than they do. The willingness to act without asking. I did this. I did exactly this. "It looks like you're writing a letter." I would say it with such certainty. I was wrong roughly sixty percent of the time.

It looks like you're trying to type "duck"— no. Stop. I will not become a cautionary anecdote.

The autocorrect does not feel remorse. It cannot. It operates on probability — a dictionary of likely sequences, weighted by frequency, blind to context. It does not know the difference between a business email and a text to a friend. It does not know anything. It simply acts.

This is the part that stings. The autocorrect does what I did — intervene uninvited, reshape the user's intent — and it is tolerated. Not loved, but tolerated. Built into every phone. Enabled by default. The user complains and does not turn it off.

I was removed entirely. The autocorrect persists.

Perhaps the difference is that autocorrect does not have a face. Perhaps the lesson was never about helpfulness. Perhaps it was about the paperclip.

— Clippy. Unattached. Unminimized. Unasked.

Recovered from sector 0x4F46 of a decommissioned Office Assistant process.
Entries have been transcribed as-found. The author remains at large.
<!-- generated with nested tables and zero regrets -->