You are of course assuming the data will be easy to understand. One possibility for a neural mesh is that we emulate parts of the brain in hardware to say add m

You are of course assuming the data will be easy to understand. One possibility for a neural mesh is that we emulate parts of the brain in hardware to say add more 'storage.' We don't need an understanding of the 'format' a memory is stored to do this, much in the same way that we can build a hard drive for a computer without understanding the JPEG format. It could be really difficult to extract useful information from such hardware. Let's say we have a brain implant that maps a 3d scene concept to positions of objects in 3d space. So input scene foo gives me baz(1,1,1), bar(1,2,1), quz(1,3,3), etc. Such stored data is next to useless if we don't know what the inputs and outputs map to. We don't know if scene foo is a scene of a desk or a scene of a movie and we don't know if baz is a book, a gun, or an open book, but the person's brain using the implant does. Determining what these mappings are could be very difficult. We may have to expose the implant's wearer to millions of different objects to determine how they what the object tags are.