Episode 70 - AI or Not?

AI-generated and AI-altered images have gotten genuinely, unsettlingly good. That's wildly useful and also dangerous!

A close-up of a weathered hand cradling a steaming white mug of black coffee on a wooden table.
What do you think? AI or not?

Prologue

Last week, the dymaptic team had its monthly social, and like a responsible adult with calendar conflicts, I missed it. That’s sad for me. We always do something fun, from terrariums to, well, the AI or Not game!

Ande (yes, the same one from Episode 38 that took us on our red teaming journey) built a game called “AI or not.” The game consists of a pile of images where your job is to identify which are real and which are a computer’s dream. The premise is simple, but the result was quite difficult. The team averaged about 20 out of 30 correct, with Amara winning the whole thing by a thread!

We also briefly played this game during my livestream this week and had a lot of fun!

Fauxologue

In case you have FOMO like me, here is a handful from the quiz. Make your guess on each one before you click to reveal the answer. If you are playing in the email, you’ll have to click through to the website to get the answers. Sorry, I just didn’t want to spoil it! (Also, this boosts my clicks, so click on!)

With apologies to Bill Watterson:

A Calvin and Hobbes comic strip by Bill Watterson showing two storylines: in the top row, Calvin and Hobbes attempt to lobby Calvin's mom for more privileges, only to be rebuffed; in the bottom row, Calvin declares the backyard vegetable garden "the Forbidden Swamp" on his map, responding to Hobbes's objection with "Maps change!"
Tap to reveal — was it AI?

The bottom one is the AI. I’m almost disappointed because “maps change” is a pretty great life lesson that fits well with the Calvin and Hobbes concept!

The team reported that the nature shots were the hardest, so see how you do!

Will the real Yosemite Falls please stand up?

Side-by-side comparison of two photos of Yosemite Falls, challenging the viewer to identify which is AI-generated; the left shows a single waterfall in daylight with green meadows, while the right shows a more dramatically composed scene with autumn foliage, multiple cascades, and moody lighting.
Tap to reveal — was it AI?

The right one is real. The fake leans a little too scenic, a little too composed. I’m pretty sure I’ve seen this in real life, and it left me a little confused about what was real!

Smokey Mountain High

💡
Yes, I know it is “Rocky Mountain High”
Side-by-side comparison of two sunset photographs over the Great Smoky Mountains' blue ridgelines, used to challenge readers to identify which is AI-generated; the left image is the real photograph.
Tap to reveal — was it AI?

Left is real. Two sunsets over the same blue ridgelines, one happened, one didn't! (That feels like AI wrote it, but I swear it was me!)

Maps should be easy, right?

Two side-by-side transit maps labeled as Portland Metro — the left displays the TriMet branding with color-coded MAX light rail lines overlaid on a geographic street map, while the right shows a cleaner schematic-style system diagram; readers must determine which is AI-generated and which is the official TriMet map.
Tap to reveal — was it AI?

The right one is the real one! I live in Portland, and I don’t think I’ve ever seen that map! The left one has the TriMet logo, which is a plus, but then says “Portland Metro” which just has the wrong vibe! But it sure does look official.

Real Rodeo Roundup

A side-by-side comparison of two rodeo scenes used in an AI-detection quiz: on the left, a cowboy rides a bucking horse in an outdoor arena; on the right, a rider is thrown from a bull in an indoor arena with a sign reading "The Hottest Show on Dirt" visible in the background.
Tap to reveal — was it AI?

Right is real, and in what might be the best story from the social, check out the sign. Amara (who won the game) said she didn't think AI was clever enough to come up with a line as good as  "The Hottest Show on Dirt." Others read that line as obviously AI-written.

Christopher here: While writing this, I realized that “The Hottest Show on Dirt” could totally be AI writing, but it was then printed on a real sign and then ended up in this photo!

Palm Springs, California

Side-by-side satellite images of Palm Springs, California used as an "AI or Not" quiz challenge, where the left image is the real aerial photograph and the right is AI-generated; both show a desert city grid with mountain terrain, making them difficult to distinguish.
Tap to reveal — was it AI?

Left is real. I’ve been to Palm Springs almost every year for the last 15 years, and I hesitated!

So, how’d you do? Were you six for six, or were you on the 60% bus with the rest of the dymaptic team?

So, how do you tell?

I asked Amara for tips on how to tell what’s real and what isn’t. She admitted to consuming a lot of “tips on the internet” on how to tell, but still thinks it is very hard, and getting harder by the day. Here are her top tips:

  • Shadows — AI often gets light direction wrong; if the shadows in a shot don't agree with each other, be suspicious!
  • Depth — On the landscape ones, watch the focus. A bokeh blur on the wrong thing (a foreground tree with no reason to be soft) gives it away.
  • The vibe — Unsatisfying but real: Amara said sometimes the only move left is to "feel the vibe of the pic."

With great power…

These image generation tools are becoming genuinely useful, which means they are scary good at generating real-looking photographs and videos.

At dymaptic, we still rely on humans to build final versions of images, but in preparation for this year's Esri User Conference, we did extensive prototyping with AI image-generation tools. This helped us come up with concepts for our booth graphics and some new limited-edition stickers. We start with AI because it's fast and good enough to mock up many ideas until we settle on what we want. (It's like software gardening, for design!) Thankfully, we have talented design folks on our team like Agnes and Juliette who can take the AI-generated idea and make the real end product.

Want to see what I mean? Come see us at the UC next month!

Deepfakes are the real issue here: the ability to make you believe something is real when it isn’t. I really want to see more technologies to help prevent these! I’ve been thinking a lot about how to prove to folks that it really is me on my livestreams or in web meetings because I think this is a really important problem.

My best tips right now are to always turn your camera on, and to have a real physical background that you can interact with.

I think our best bet in the short term, though, is to make it socially unacceptable to try to pass off an AI-generated image as real.

Newsologue

(written by Jaws)

Epilogue

This week started as a bunch of notes from Holly, Ande, and Amara about the social, then a Word doc of the images and the answers from Ande. I fed all of that into Jaws, and it drafted this. It was pretty bad, I guess I’m already spoiled by Fable!

Then I rewrote it the way I wanted, had Jaws edit, and then had Holly edit.

P.S. The cover photo is AI, what was your guess?

Subscribe to Almost Entirely Human

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
jamie@example.com
Subscribe