It was a total accident.

I clicked an option in Gmail called Enable Experimental Access. It's innocuous enough--why not try something new? Anyone can try it. It's listed right under settings.

I wish I never had.

Announced way back in May of this year, the new Smart Compose function sneaks up on you in a way that seems strangely foreboding. You're typing along writing an email to your boss, and suddenly words pop up on the screen. You type: "I'm running late to our..." and Gmail automatically inserts the word meeting. "Do you have dinner plans for..." and Gmail adds tomorrow to your sentence. At first glance, it seems helpful.

Then, it hits you.

There's a silent partner thinking, analyzing, and scheming behind the scenes. A shadow robot, an agent of English composition, is watching what you say. Google engineers claim they are not reading your emails, not storing any messages you're sending to that colleague in the branch office, not cataloging every gripe session. They're blissfully unaware.

And yet--there's a creeper in my email. At times, Smart Compose chooses not to play along, somehow not finishing my sentences for me. Typing "When do you get off of..." must be obvious, even to a fifth grader. The next word should be work. "Do you like your..." should be a no-brainer. I'm going to type boss next. If the word agenda doesn't belong at the end of "What is on the..." I don't know what would be. And neither does the bot.

The AI works, then the AI doesn't work. It's robot roulette.

Then Smart Compose suddenly pops up again. In the course of using the feature over the past few days, strange words like "agenda" did pop into place. The word "meeting" would sometimes appear out of nowhere, and sometimes it wouldn't. According to what I've read, this is a work in progress, but it's disconcerting to have a bot write your sentences.

It will become much more disconcerting. One final word at the end of a question is one thing. In the near future, bots will be able to finish a paragraph as well, once we give them permission to analyze our entire Gmail archive. I'm guessing a smart bot today could look at all of my messages and write a reply to public relations reps in a snap. And, they could easily compose a message to a source--maybe even suggesting questions I'd send by email.

This might all seem helpful, but please make sure you try the feature before you willingly oblige these bots. Today they will add a word at the end of "Let's meet for lunch next..." but tomorrow they will start writing entire emails. Bots can already write articles (although they are crude and impersonal). My issue with Smart Compose is not that there's some value in offering suggestions. It's simply what comes after this in a few years or a few decades that could lead to a robot apocalypse because we don't really know why it's happening.

I've written about the dangers of AI many times before, and it's always a far-off scenario. Bots that can walk and talk, who understand inaudible commands, who drive cars. Smart Compose feels wrong because it is happening now; it feels like a strange entity invaded my email as words pop into place. I don't really want to know how it works. I don't want a bot writing my email, even if there was a time I thought that was maybe a good idea. It's way too creepy. It feels like, as a writer, I'm being toyed with. Reduced. Marginalized.

I'm fine with my wife finishing my sentences--she knows me better than anyone.

Google? Gmail? It feels too much like the AI has taken a step too far.

It's magical and surreal.

And, I turned it off.

I've been finishing my sentences in unusual ways ever since.