The dimensionality reduction problem space rarely tries to reduce down to only two or three dimensions. The fact that in the end you are being reduced to so few dimensions allows for a lot of potential errors to not meaningfully impact the result.