These modes are usually computational effects rather than true optical shallow depth of field. The camera estimates a depth map, then applies stronger blur to areas it decides are farther from the plane of focus.

From the answers, it can do this a few ways:

• Analyze local contrast and edge detail: areas that already look slightly out of focus have different contrast/blur characteristics, which can help separate in-focus subjects from background.
• Use autofocus/depth information: cameras with phase-detection points across the image may estimate subject distance at multiple locations.
• In some systems, combine multiple captures or slight viewpoint changes to infer depth, though not every camera does this.

If your camera seems to make two exposures, it may be combining shots or simply using a processing routine tied to that mode. But the key idea is the same: it is not changing the lens’s actual depth of field much; it is identifying the likely subject and digitally blurring other regions.

Because this is based on estimation, it can make mistakes around hair, transparent objects, low-contrast edges, or noisy backgrounds.