If you wrote numbers or letters as precisely as the triangles are drawn in the example, I doubt computers would have much of a problem reading them. We've got fast algorithms for digit classification that are pretty much perfect on the MNIST dataset with just 0.21% error (and I'm pretty sure there are errors in the dataset itself, which brings the real error rate down). MNIST is an easy image recognition task compared to others but it's not done with clearly written numbers.

Here's an example of some that were misclassified by an algorithm: http://www.concordia.ca/content/concordia/en/research/cenpar...

Machine-readable codes also have space-efficient error correction built in, something that plain text does not support. A text label can become unusable with the loss of a single character.

Only if the code has no error correction built in. If the code is presented to the user to write down it's perfectly possible to have redundancy in the code.

Then it's not plain text anymore (not human readable)...

How so? You can easily have error correcting codes that are human readable.

Yep, but drawing readable triangles is much easier for a lot of people

Seems like a more durable solution would be making an encoding of the alphabet that's easier for machines to read.

Like the one we've used on bank cheques* for all of time

MICR E13-B: http://www.micr-fonts.com/MICRfont/micrfont.html

(*British. <grin>)

I meant one that could be created by hand, but that works for ones made by printer.

Or on snail mail.

Morse code, alpha bravo charlie, ASCII... there's actually a long history of doing this. Maybe not one meant for easy of memorization and drawing though. Even Mark Whatney had to be taught Morse code on Mars.

(Yes the phonetic alphabet is meant for ease of hearing by people, but writing a program to understand 26 distinct sounds can't be nearly as hard as natural language transcription.)