Mechanical Turks All the Way Down

I jabbed the accept button on one of the many touch surfaces that suddenly popped into reality in my shared office, home, and daycare. Can’t be too eager on your first day of the job.

There was a low din of my five-year-old arguing again about what xe would or would not eat for breakfast. Thank god for the AR distraction-canceling face mask. It was cranked up to 98% reality canceling. With it on I could safely ignore the mounting skirmish between little Tovio and my partner.

Mechanical Turks All the Way Down

Illustration by Justice.

This was my second job today. A basic anniversary gig like I’d seen in the training. There were a lot of notes and I quickly scanned them for any red flags, but accepted immediately. Taking jobs fast shows good teamwork which means my teamwork score goes up which, algo willing, will lead to more lucrative jobs.

If I’m not in the algo’s top tier, we can’t afford the next level of healthcare we need. The raw number of medical claims sitting in my personal email could be felt through the unread count floating above it. Inhalers, lung rehabilitation treatments, microextracorporeal membrane oxygenation machines. The list doesn’t end.

Next week is a special anniversary of working for my principal. He is middle-aged (36) and works at a mid-sized company. He has most recently purchased a new pair of shoes, some weekly groceries, and a measles-mumps-rubella-ebola-coronavirus home booster immunization kit. I want it to be really special. He tends to enjoy things like cold food, travelling alone — not sure if by design or low desirability — and collecting little AR-tagged stickers from his travels. More than anything he is striving to be more efficient in his life and get more out of it.

It went on and on like that.

For a second I felt that I was lost daydreaming. It wasn’t that I was daydreaming and zoned out — even though it said a lot, the message didn’t say much.

The final comment was unexpected:

How do I know that you are a person or a machine?

That’s exactly what I was thinking.

Was this a corp test? How did I know this requester was a real person? I’ve been warned by a friend who also works this gig that they will slip these in every now and then to make sure I’m not using personal algos to do my work.

While I was trying to orient myself, three windows filled with top-10 lists for presents for managers of 3 to 5 years. One window included a sock collection that matched shoes this group had bought, another popped up a guided meditation app focused on dealing with the uncertainty of life, and the third suggested a wine-flavored powdered alcohol collection of the best wines ever made.

The sock collection had already been suggested to the client by the algo.

How uninspired. Just like I expected. I also felt a teeny bit thankful. These uninspired suggestions were why I had a job in the first place.

The client apparently felt the same, adding to the channel:

Your suggestion is highly predictable. Our models output the same ranking.

We are looking to expand past our current bottleneck. The requirement of novelty is high for this principal. Churn is highly likely without a high-importance gift.

I dictated quickly based on the training I had just received:

I know it feels unnatural. The important part is that you are asking these questions. We show care by doing these actions even if they aren’t 100% by ourselves.

The display flashed an icon to be more casual. I continued:

Just think about all the times we’re saving people from horrible arguments. That reduces everyone’s suffering. The authenticity is that you care.

An immediate reply:

Acknowledged. What is your recommendation? On standby.

This reply made me pause in thought again. Am I helping someone’s machine buy themselves a gift? Is there solidarity between people and machines working the same gigs? Is this like training my replacement? Or being a slave to a machine? At least more of a slave than usual.

The system dinged at me that I was taking too long. I had half a mind to sabotage this job in some way. I remembered the company had a one-strike law for “conspiring against the machines.” We jokingly referred to it as the Terminator Law.

The requestor pinged at the same time:

The optimal time to get this to them will be within the next hour and thirty-five minutes. Acknowledge?

I said as fast as I could:

I recommend the Chupa Chups reproduction taste machine. It gets at something that your ‘principal’ posted a long time ago in the archives.

It was a long-shot item on the list that the system had spit out with low confidence and high uncertainty.

I got a quick reply:

Your follow up suggestion was not the highest confidence score for a gift. If I boost that original memory, it will change the ranking. It scores very high on nostalgia for my principal. 🙌👉👈👾💞

It left a bunch of gibberish in the chat, but then who knew:

eNo1zEEOgzAMRNGrzFXYsOISJJliSxAjx2ng9kWV2P//JrglCxy9BSzxRghhXugNm35ZoYF0Q/qxViRq3Rp4ZZ6BIXSi9SzvMazvBdnqZ9f8BBryB2f1x1/W8QMJXSke

Machines are weird.

The channel was closed, the Chupa Chups machine was ordered, a card was generated with some pithy platitude appropriate for the situation, and a star was added to my budding virtual jar.

Another job flashed on the surface immediately. I accepted it eagerly to push off the questions that arose in people helping machines help people.

While starting to scan the details, a part of my mind was stuck on the last job. I wondered when the bots could better calculate human behavior and recommend the same tasteful gifts as I did. Maybe the novelty of a minimum viable human picking gifts would wear off.

Then I had to turn up the distraction cancellation as I could hear the breakfast ordeal escalating.