Trusting the machine: AI will reveal what is art, what is beauty, and which are the hidden diamonds
- Lorenz Aschoff, EyeEm co-founder, explains what AI and machine-learning can teach us about photographs
- Machines must learn culture, aesthetics, and geo-specific trends to interpret “ beauty”
- How much longer will identifying and understanding beauty be a uniquely human trait?
Beautiful images are one of the cornerstones of our app and tech experiences. But how do you find the best images out of the millions created every day? And can AI identify beauty itself?
Lorenz is fascinated by beautiful images, and he uses computer vision technology to analyse and identify them. This non-human eye is helping him understand what makes an image truly beautiful.
Listen to his full talk here — and below we’ve extracted his key points and quotes to bring you Up To Speed…
The tough questions that get Lorenz out of bed every morning:
Can we understand beauty? And can we understand it at scale?
How do we get rid of the not-so-good photography?
How do you find out what’s worth looking at when billions of photos are taken every year?
In photography, subtle differences make all the difference:
Lorenz fed his technology famously-beautiful images to learn from and compare to similar images. It needs to be able to ask: what is the difference and why?
“When we look at an image, we need to analyse over a million different factors… it’s not as easy as saying that it’s symmetry or a certain type of composition that’s beautiful — it’s really abstract.”
Since 2010, photography has changed in fundamental ways that have been dictated by technology:
The two things that changed photography the most over the last ten years have not necessarily made photography more beautiful — but they are fundamentally different. The front-facing camera and the selfie stick has enabled selfie culture, and now a completely different style of image flooding the world.
Deciding which images to keep — and which to dispose of — is still a big problem:
Today, 80% of all photos are taken by mobile phones and the volume of photos taken has skyrocketed. Snapchat’s original ethos is one reaction to the flood of images and selfie culture: you just delete the image.
Social networks like Facebook narrow the filter of imagery because of your friendship circle. But what about filtering by beauty itself?
Sensitivity to aesthetics requires understanding of nuance, culture and human differences:
Aesthetics is trend- and location-based: what may be beautiful in Asia is not necessarily beautiful in Europe. There is not one single type of aesthetics — so being able to personalise results is key.
By providing source images, you can provide a custom algorithm to decide which images. Aesthetics are also trend-based: they change over time. This is where community monitoring is paired with algorithmic learning to understand what resonates.
The big questions: How far can we go? Will AI ever detect “art”?
“We are not that far yet. What we do well now is auto-curate — we can show the top 100 images out of a million. But there is a tiny fraction of images that are different — and only the human brain can decide this.
“Over time, we need to track changes in trends and to find outlier photographs. Computers can sometimes miss amazing photos — and finding the diamonds in the masses is the whole point.”
Get TOA.life in your inbox — and read more from TOA’s network of thought-leaders:
Sign up for the TOA.life newsletter
The Great Power Shift… to us. Welcome to our weird blockchain future where we save our jobs, own our data, and shake up governments
If your best-ever sex was with a robot, would you go back to humans?: Sex with machines — is SexTech the next big leap for human sexuality?
