We know that a dog is a dog, regardless of whether it's sitting, running, or trying to comprehend escalators. But one of the tricky parts of designing an artificial vision system for a computer is getting it to recognize that an object is the same, regardless of perspective, position, ambient lighting, and a whole slew of other factors that our brains process easily.

If an object we already recognize moves, whether it changes position or size due to perspective, our brains perceive that it's probably the same thing we were looking at a half second before. This is scientists call "temporal contiguity," and it's believed that by learning to associate images that appear in rapid succession, we understand that they're the same thing, regardless of movement.

To test this, researchers at MIT's McGovern Institute for Brain Research, set up a visual display that broke "temporal contiguity" as we're used to it. The display would show an object decreasing or increasing in size — mimicking either approaching or receding from the screen — and then would swap it out for a different object. So a dog would change into a rhino, or something similar. They recorded the brain activity in the inferior temporal cortex (the part of the brain thought to handle object recognition) of a group of monkeys while exposing them to these images.

What they found is that the monkey's sense of recognition of the object began to change, and that some would begin to link together the two different animals as one and the same, essentially turning "dog" into "dogrhino."

This provides the clearest evidence to date that "temporal contiguity" plays a major part in how we learn to recognize objects, and could theoretically be used for training robots to recognize objects, regardless of angle and size.

Research published in Neuron.