AI training

How a computer learns from lots of examples

The short answer

AI training means showing a computer a huge pile of examples until it can spot the pattern on its own. For cleaning, the examples are hours of video of real jobs. The more kinds of jobs it sees, the better it gets at new ones.

Four first-person views looking down at the work, both hands in each: a school restroom sink, a dental office counter, a warehouse locker room floor and an office kitchen counter. — Four buildings, four jobs. Variety teaches more than the same job over and over.

One frame, with the kinds of labels a training team adds. Drawing, not a real recording.

Like breaking in a new hire

You don't hand a new cleaner a rule book for every sink. They shadow someone for a week, and by Friday they know the mop goes in the bucket after the trash is pulled because they watched it happen.

AI training works in a similar way. A computer program watches thousands of hours of video, and each time it guesses what comes next and gets it wrong, it adjusts a little. Over millions of small fixes, the guesses get better.

What the computer sees in one clip

Take a 30-second clip of a cleaner emptying a trash can. To you it is nothing, but to the computer it is a lot of small lessons.

Where the hands are in each frame, and how the fingers close on the liner
Which tool is in use: a can, a liner, a cart
What step comes next, like tying the bag before lifting it
What the room looks like: tile floor, stall doors, a paper towel holder

Why variety beats more of the same

If the computer only sees one office break room, it learns that one room. Show it a school restroom, a dental office and a warehouse locker room, and it starts to learn the job itself.

That is one reason these teams want real work from many different buildings. A countertop wiped 40 times for the camera teaches very little.

Why this matters to your crew

Every account your crew covers is a different building, with its own sinks, floors and trash setup. A computer that is learning gets a lot out of that mix. Our review also looks for real work at a normal pace, so the hours from a regular shift are the ones that count.

What other owners ask

Does the computer know it is watching my company?

It learns from what is in the frame: hands, tools, surfaces and steps. That is why we ask your crew to keep private areas and personal information out of the shot.

Why does slowing down or repeating a task hurt?

The computer copies what it sees, so if it sees fake, slow motions, it learns fake, slow motions. That is why our review turns down hours like that.