人工智慧也會被光學錯覺欺騙——這揭示了我們大腦的哪些秘密！

Artificial intelligence, or AI, is now able to 'see' optical illusions in much the same way that humans do. This surprising discovery is helping scientists learn more about how our brains work and why we sometimes see things that are not really there. Optical illusions are pictures or images that trick our eyes and brains. For example, the Moon can look bigger or smaller in the sky, even though it stays the same size. These illusions show that our brains do not always see the world exactly as it is.

Our brains are very busy, and there is too much information to process all at once. To help us, our brains focus on the most important details and ignore the rest. This helps us make sense of the world quickly, but it also means we can be fooled by illusions. Scientists have found that some AI systems, especially those using deep neural networks, can also be tricked by the same illusions that fool people. This is surprising because AI is designed to be very good at spotting details and patterns.

Eiji Watanabe, a scientist in Japan, used a deep neural network called PredNet to study how AI reacts to optical illusions. PredNet was trained by watching millions of video frames of natural scenes, similar to what people see every day. It learned the rules of how things usually look and move. When Watanabe showed PredNet the 'rotating snakes' illusion, the AI thought the picture was moving, just like people do. This supports the idea that our brains use something called 'predictive coding,' where we guess what we expect to see before we actually see it.

However, there are differences between how AI and humans see illusions. For example, when people look at the rotating snakes, they can focus on one part and make it stop moving in their mind, while the rest of the picture keeps moving. The AI could not do this because it does not have an 'attention mechanism' like humans do. This means it looks at the whole picture at once, instead of focusing on one part.

Some researchers are also using ideas from quantum physics to help AI understand illusions even better. For example, Ivan Maksymov in Australia created an AI that can switch between different ways of seeing an illusion, just like people do. This could help us understand how we make decisions and how our brains work. Scientists are also studying how astronauts see illusions differently in space, where there is no gravity. This research could help us prepare for future space travel.

Even though AI is getting better at seeing like humans, it still cannot do everything our brains can do. But by studying how AI and people see illusions, scientists hope to learn more about both computers and the human mind. This research could lead to new discoveries in science, technology, and even space exploration.