Professional YouTubers know how to attract views and clicks. This can be done by putting in creative titles that will entice viewers to find out more, or they can use interesting thumbnails that captures the attention of the viewer. However to help makes its videos more enticing overall, regardless of who uploaded it, Google is turning to neural nets for help.
For those unfamiliar with neural networks, basically this is like a model in machine learning that is inspired by biological neural networks, like those in our brain (or brains of animals). Ultimately the goal here is to train computers to be able to think like humans, as opposed to merely doing whatever they are instructed to do.
How does this apply to YouTube thumbnails? By using neural nets, Google is basically training YouTube to identify thumbnails that might create a positive image, or at least an image that is enticing to viewers. So instead of grabbing some random screencap from your video, it will try to look for portions of the video where the images are clear and “positive”.
Ultimately this should result in videos, regardless of amateur or professional uploads, to be more enticing overall. Whether or not this will work remains to be seen, however if you do plan on uploading videos, you can use custom thumbnails if it boils down to that.