This is pretty theoretical but really interesting! Human randomness isn't perfect by any means, but it's hard to say by how much.

If I were to think up - manually - 64 'human-random' hexadecimal characters, then how many truly random (i.e: generated from a decent random algorithm) would be equivalent for security purposes?

I'm guessing about 40-50? But I could be way off. Any research in this area?