We had a lot of fun creating today’s April Fools joke (actually we enjoy doing it every year) and we thought... if we were going to write an article about artificial intelligence, why not use it to create all the images in the post as well? Little did we know how hard it would be.

To create an image you provide the A.I. with a bunch of keywords and it “imagines” images for you. It’s absolutely remarkable to throw things at it like “Orangutan” + “campfire” and it creates something hyper surreal such as this…

Aside from a few quirks like the hovering fire, it fully understands what a little campsite could theoretically look like including a tent structure (with a cozy blanket inside), camp pots and bowls for a meal, as well as a dense woodsy night scene. We just keep feeding it words in anticipation of seeing what on earth it can come up with next.

What’s fun is watching the A.I. start to create the image as it gives you four variations. For example, “man + canoe + serene lake + oil painting” slowly generates the four different images on the left which transform into the ones on the right…

There are subtle differences with each creation, but if you look closely you can see where the A.I. quickly doesn’t understand canoeing…

Paddles look like oars
Higher chance you will get a plastic double bladed paddle or some bamboo variant vs. a wooden single blade
Paddles will have broken shafts that will start in one place and randomly protrude out of other places
It does not understand what a PFD or a Lifejacket looks like, you usually get some weird puffy vest with straps
Don’t even bother asking for portage packs, food barrels, etc. Usually that suggestion will go ignored
Canoes vary wildly in materials and number of thwarts. It has come a long way however. In the previous version any time you asked for a canoe it would produce a kayak… usually with about seven thwarts and zero seats.

The best however, is asking for a photo of someone portaging…. to which it has no clue what that means.

It always portrays someone either walking their canoe, or walking through their canoe. Even trying to guide it by telling it “person holding a canoe over their heads” still doesn’t seem to make it any clearer for it.

At least it seems to extrapolate that there is a trail involved. Each version of A.I. IS getting better however. I’m sure you’ve seen generated photos of people with 7 or eight fingers but that rarely happens anymore. Canoes now have less thwarts and PFDs look a little more functional. We’re getting there.

So in these computer fever dreams, hopefully one day it learns just what portaging means, and then maybe it truly can work on ways to help us avoid doing it.

Trail Mix is an assortment of outdoors related blog posts that can be fun, quirky or just thought provoking.