homehome Home chatchat Notifications


Scientists Used Quantum Physics to Help AI Fall for Optical Illusions

Scientists have coaxed AI to see the Necker Cube illusion the way we do thanks to quantum tunneling.

Tibi Puiu
October 28, 2024 @ 6:29 pm

share Share

Optical Illusion
Credit: Fran_kie/Adobe.

Imagine staring at a drawing of a cube, a simple, elegant shape made of intersecting lines — a child’s toy, in essence. Yet, stare long enough, and suddenly the orientation of the cube appears to shift, flipping inside-out. This is the Necker Cube, an optical illusion discovered in 1832 by the Swiss crystallographer Louis Albert Necker.

Just as the Necker Cube invites us to toggle between two realities, Rubin’s Vase challenges us to see either a vase or two faces in profile, an image that continually slips from one form to another.

Both the Necker Cube and Rubin’s Vase are classic optical illusions. They confound even our best-trained brains by toggling between different interpretations. However, scientists have coaxed artificial intelligence to fall prey to the same illusions. But to get there, they had to do something rather radical — inject a bit of quantum physics.

The Challenge of Teaching AI to “See” Like Humans

Necker Cube (left) and Rubin’s Vase.

Most AI-based vision systems can recognize faces, and categorize objects, and some can even make new images from scratch. But give them a trickier challenge — like an optical illusion — and they fall flat. For a long time, computers could handle the math behind images, but they couldn’t mimic the strange, fluid way human perception shifts back and forth between interpretations.

For Ivan Maksymov, a research fellow at Charles Sturt University in Australia, this challenge represented something bigger: a chance to explore the core question of whether AI can ever truly see the way we do. “Optical illusions trick our brains into seeing things which may or may not be real,” Maksymov explains.

To help AI grapple with illusions, Maksymov turned to a principle from quantum mechanics known as “quantum tunneling.” This phenomenon lets subatomic particles seemingly bypass barriers that, by the laws of classical physics, should be impenetrable. It’s the type of physics that feels pulled from science fiction, and Maksymov wondered if it might open new doors for AI.

Building a Quantum-Tunneling Neural Network

To make AI see in this quantum-inspired way, Maksymov’s team built what they call a “quantum-tunneling deep neural network.” In regular neural networks, artificial “neurons” process and store data in layers, with each layer doing a bit more of the work to produce a final answer. But Maksymov’s design took an unconventional approach, embedding the process of quantum tunneling into the network itself.

Imagine each neuron as a tiny worker, stacking data to reach a threshold, like a person trying to jump over a fence. In normal neural networks, neurons that can’t meet that threshold remain inactive. But Maksymov’s neurons, inspired by tunneling, are occasionally allowed to “jump” through this barrier without hitting the threshold—just as an electron might pass through a wall in quantum physics. This tweak allowed the AI to react to images in ways that mirror the brain’s own flexibility.

As the AI examined illusions like the Necker Cube and Rubin’s Vase, it did something remarkable: it produced not one, but multiple interpretations. Sometimes, it saw the shaded cube face at the front; sometimes, at the back. Occasionally, it even hovered between the two, as if it couldn’t make up its mind.

“When we see an optical illusion with two possible interpretations . . . we temporarily hold both interpretations at the same time, until our brains decide which picture should be seen,” Maksymov wrote in a blog post. His AI model captured this uncanny, human-like ability to see two things at once.

A New Kind of Vision

Why push AI into this odd realm of optical illusions? Maksymov envisions applications that could transform human experiences in high-stakes fields. Imagine, he suggests, a pilot flying through thick clouds, relying on instruments to tell up from down. Misinterpretation can be deadly, and the pilot’s brain, like the AI’s, must sort through ambiguous signals to arrive at the truth. An AI capable of understanding ambiguity might one day help avoid such disorientation.

There’s also hope that quantum-tunneling AI could one day aid in medical diagnoses. When AI models are trained on ambiguous patterns, they might help detect early signs of cognitive decline, like dementia, where the ability to recognize visual patterns often diminishes. These applications could one day give AI systems that not only see images but also interpret them in a way that feels uniquely human.

It may be years, perhaps decades, before AI systems with this level of insight appear in everyday technology. For now, however, Maksymov’s quantum-tunneling neural network remains an awe-inspiring experiment, challenging the boundaries of what it means to truly “see.”

share Share

Mexico Will Give U.S. More Water to Avert More Tariffs

Droughts due to climate change are making Mexico increasingly water indebted to the USA.

Chinese Student Got Rescued from Mount Fuji—Then Went Back for His Phone and Needed Saving Again

A student was saved two times in four days after ignoring warnings to stay off Mount Fuji.

The perfect pub crawl: mathematicians solve most efficient way to visit all 81,998 bars in South Korea

This is the longest pub crawl ever solved by scientists.

This Film Shaped Like Shark Skin Makes Planes More Aerodynamic and Saves Billions in Fuel

Mimicking shark skin may help aviation shed fuel—and carbon

China Just Made the World's Fastest Transistor and It Is Not Made of Silicon

The new transistor runs 40% faster and uses less power.

Ice Age Humans in Ukraine Were Masterful Fire Benders, New Study Shows

Ice Age humans mastered fire with astonishing precision.

The "Bone Collector" Caterpillar Disguises Itself With the Bodies of Its Victims and Lives in Spider Webs

This insect doesn't play with its food. It just wears it.

University of Zurich Researchers Secretly Deployed AI Bots on Reddit in Unauthorized Study

The revelation has sparked outrage across the internet.

Giant Brain Study Took Seven Years to Test the Two Biggest Theories of Consciousness. Here's What Scientists Found

Both came up short but the search for human consciousness continues.

The Cybertruck is all tricks and no truck, a musky Tesla fail

Tesla’s baking sheet on wheels rides fast in the recall lane toward a dead end where dysfunctional men gather.