TL;DR:

Some advantages of a binning sensor

• Integer 2X+ zoom, variable fractional zoom
• 8K+ video capture
• Single-frame HDR, single-frame noise reduction
• Multi-output video

---

It's kind of a gimmick, but not much, because very few people capture photos at full non-binning resolution. Most manufacturers disable that mode by default, make it difficult to enable, or remove lots of features in that mode. I've only ever seen one person who always captures at full 64MP by default in their phone. Most users don't know what pixels are, and if they do a substantial amount of them also know that higher resolution doesn't necessarily mean higher quality, and lots of them don't know where to change the resolution

• Smartphone lenses are compact and cheap, I find it hard to believe they can resolve anywhere near 50MP.
• Smartphone sensors are necessarily small (much smaller than full frame) so in order to have 50MP worth of photo sites, those sites must be tiny and thus noisy, particularly in low light situations.

Probably not true 50MP but close. When Apple changed sensor the first time recently they have further changed the default resolution to 24MP to balance between full 48MP and binned 12MP photos. And that means the iPhone lenses must have resolvability at nearly 48MP. The camera app will take multiple successive 48MP frames and combine to the 24MP output

Since combining photos and processing with computational photography is standard in phone cameras nowadays, budget phones do it as well, to increase quality and reduce noise. So even when the photosites are tiny, the output would be acceptable by most people

What's more, it's possible to downscale from 48MP to 12/24MP from a single frame, or get some intermediate zoom levels and still get 12/24MP outputs which is great when no multiframe processing is available. It's oversampling as Tom Axford mentioned in another answer. Most people don't know that the 2×/3×... buttons in the camera provides the best quality and just zoom in digitally slightly to 1.3×, 1.5×... which would result in terrible output with a 12MP sensor. That means one big advantage of binning sensor is to have variable zoom between 1×-2× or 3/4×. Oh I should also mention that Apple added an option to zoom in slightly to 28 or 35mm by default, which won't be possible without a higher resolution sensor:

Default lens settings High resolution settings

Arbitrary variable zoom like this is very useful for smooth zoom while recording videos. Besides, in video mode it's almost always slightly zoomed in for image stabilization purpose, especially EIS, and a higher resolution is highly beneficial in this case because you have to rotate/skew the frame as well, instead of only scaling linearly

Furthermore, 48MP+ sensors are a must for 8K video recording. That may be the only reason Apple decided to change the sensor in newer iPhones, because Android phones already had that capability long long ago. And since they now have the new sensor, they added the other default zoom and resolution settings to utilize more of their high res output. In video recording the high noise and low chroma resolution are also not a big concern, because the frames are always moving, averaging out the chroma and luma noise

With a 108MP or 200MP sensor though, 3× or 4× would probably be terrible, but at least you'll get more useful information in many situations, for example to reduce noise. But the main usage for such sensors is still at 2× or below. That's probably one of the reasons why Samsung changed the 10× telephoto camera to 5× to compete with the 5× camera in iPhone, because they can still use 2× non-binning mode to get 10× as before, with some acceptable trade off. And they can further combine with outputs from the remaining cameras to reduce noise/increase color accuracy

• 50MP is a lot a data, which leads to bigger file sizes and unnecessarily clogging up ones phone.

Not true. As mentioned above, the default capture mode is still non-binning mode, so 12.5MP for 50MP (2×2) or 200MP (4×4), 16MP for 64MP (2×2), 12MP for 108MP (3×3)... The file size is still the same as if using a low-res non-binning sensor. In fact if you use a more modern format like JPEG-XL, HEIC, AVIF... then the output would be even smaller than JPEG files at smaller resolution. Chipsets nowadays support 320MP cameras so the raw data amount is not an issue

Another useful feature of such sensors is single-frame HDR (the classic tone-mapping HDR) or single-frame noise reduction. Traditionally HDR is done by capturing multiple photos at different exposure level, but there's a temporal downside: movements would be blurred. In single-frame HDR pixels in one diagonal will expose longer or has higher sensitivity than the pixel pair in the other diagonal. More advanced sensors can even vary the exposure levels in 4 pixels in the same color filter, making the output's dynamic range even wider

In Samsung Galaxy cameras there's also Zoom Anyplace which output multiple video streams at different crop coordinates and zoom levels. This can be useful in various cases. For example Zoom meetings already have the capability to split multiple frames for different person, but they do it in software. A hardware solution would be better