We often fall into the trap of placing ML algorithms in awkward situations that even humans would have trouble with. For instance, we feed a flat still image to

We often fall into the trap of placing ML algorithms in awkward situations that even humans would have trouble with. For instance, we feed a flat still image to an ANN and expect it to detect things that a human would need lateral head motion and stereoscopic vision to detect. We feed a large detailed colour image into a deep learning algorithm, but humans only see in colour and high resolution in a small spot in the middle of the visual field. This might sound like it would make the job harder for ML, but perhaps it's actually helpful, because there's less data to decode, and it's obvious what the most important part of the image is. Our brains are only designed to make sense of the world when it is also equipped with a body that can assist it and provide more information by turning or moving around, or just taking a closer look.