Picture_3_1.PNG (1334×750)

Screengrab via AI Scry

See things differently.

Artificial intelligence (AI) may some day surpass the human race and lead to our extinction, but we’re all right so far. It’s technologically advanced, but it doesn’t yet understand how humans interact in the real world. 

Remember Tay, Microsoft’s teen chatbot? They released her into the wild, wild Web, innocent one moment and a racist, sexist and genocidal AI the next? Tay was an IRL Chappie, corrupted by trolls and abusers, picking up the worst parts of social media until its plug was pulled.

Of course, there’ve been other moments of AI gone terribly wrong. There was that one time we found out Google Photos tagged black people as “gorillas.” As the Daily Dot reported then, the common mistake with AI development is “a lack of awareness or understanding as to what potential harm these technologies can do, and how to prevent it in the first place.”

That’s where AI Scry comes in. The app, created by company Disk Cactus (the team behind the ingenious Emoji Keyboard), is described as a “remote viewing application powered by an alien psyche,” which is represented on an “Attention Aperture” slider that messes with the AI’s guesses of what your phone’s camera is displaying.

“The image database that our neural net is trained on, Microsoft COCO, specifically features objects that a four-year-old could recognize,” Disk Cactus co-founder Tara Shi explained. “As a result, it sees a lot of dogs, donuts, bananas, and cell phones in weird places.”

However, there’s nothing extraterrestrial about AI Scry (it’s all controlled by a team of human designers), and therein lies the lighthearted and playful aesthetic of the program. The point of it is to introduce people to the technology of artificial neural networks (also known as ANN, a model for machine learning which, in the case of AI Scry, is essentially a pattern recognition system) in a way that doesn’t evoke images of a dystopian robotic takeover.

So, in hopes of better understanding AI and ANN, I spent several days using the app throughout different parts of New York.

AI Scry

Upon downloading AI Scry in my bed the first thing I had the app try to guess was, obviously, my bed. It’s no surprise that it got half of the image correct. However, it mistook a giant coat of mine for a dog, which I found amusing. Curious to see the difference in response I would get once I changed the Attention Aperture, I moved the cursor to the far right and got...

Ai Scry

This. I don’t even know how to interpret any of it, except that it reads like the lyrics to a System of a Down B-side. It’s here that you really begin to see the flaws of AI and its objective observations, as the app becomes more sensitive to what it’s being shown. It can tell you what it sees, but not what’s going on, which speaks to the limitations of AI. Sure, I had gotten some good initial guesses from AI Scry in the comfort of my bedroom, but I needed to test it in the wild.

AI Scry

I traveled to the Brooklyn Bridge and kept AI Scry on the highest part of the slider, fully aware that the constant movement of people and change of scenery would produce some absurd results. “Young men standing while eatin’ holding from walking maze,” is somewhat accurate in a pretty poetic way. Still, moments like this confirm the developers’ note that the app lacks any “direct knowledge” of real-world objects, and, instead of recognizing things, “maps patterns received on the signal line to a sequence of word-choice probability distributions and assembles an output stream.”

Which would account for the sheer abstraction.

AI Scry

There’s something being said about a hipster here that suggests that this AI has a sense of humor.

AI Scry

I would like to think that the phrase “youth seems wieners over cake in Canada” means “dude peeing” in a more progressive alternate universe.

AI Scry

This was the first picture I took on a mid-range setting and, well, AI Scry doesn’t seem to know anything about instruments.

AI Scry

I pass this great work of art every day in Bed-Stuy and wanted to see the app’s thoughts on it. It essentially got it right, noticing one of the chains on the door, but confusing the writing on the door for a sign. Also, I can only assume that “unknown” is because it couldn’t read the graffiti. I wouldn’t expect an AI with the mental capacity of a four-year-old to know what gentrification is, anyway.

AI Scry

AI Scry can’t decide if Abraham Lincoln was a brown horse or a dude holding a dog or cat.

AI Scry

Neither I nor AI Scry understand my girlfriend.

AI Scry

AI Scry called Donald Trump a “son of a diaper.” This is easily the most accurate response the app has ever produced.
Promoted Stories Powered by Sharethrough
artificial intelligence
Here's why AI won't be replacing cybersecurity experts anytime soon
But such is not the case in the domain of cybersecurity. Although security solutions based on unsupervised machine learning do exist, relying entirely on artificial intelligence to spot cyberattacks isn’t totally practical because such systems yield a large number of false positives. You eventually need the help of human experts to find evidence of security breaches and make critical decisions.
From Our VICE Partners
Group

Pure, uncut internet. Straight to your inbox.

Thanks for subscribing to our newsletter!