[nextaccountic]

This is called few-shot learning (or in the limit case, with a single example, one-shot learning). One way this may be achieved by first training a very general model (maybe with huge data sets), and then fine-tuning it into a specific example (this is called transfer learning). [0]

One reason researchers suspected this must be possible is that human beings, as well as other animals, can learn stuff by watching it for just a few seconds. But we have some prior baggage, because we spent our whole lives learning.. other, vaguely related stuff, and it turns out that knowledge is often transferable.

[0] This isn't the only way, there's also meta learning