Bill Gates was quite a fan of this concept, which Microsoft Research investigated in the early-2000’s. The POW scheme involved forcing senders to calculate a CP

Bill Gates was quite a fan of this concept, which Microsoft Research investigated in the early-2000’s. The POW scheme involved forcing senders to calculate a CPU-intensive hash such as a cryptographic checksum that could be quickly verified as valid by the recipient. The proven expenditure of CPU would indicate that the sender spent some money sending the message. It’s a neat idea in theory. But in practice, e-mail filtering works better as an exercise in content and sender categorization. The best systems use a multitude of techniques to generate features and then use machine learning to determine the high-dimensional space that contains spam messages. This space can vary for each recipient. Google’s lead antispam ops guy told me at a conference a few years ago that moving to this approach dramatically improved their spam filtering accuracy. But some amount of manual feature engineering is necessary, such as looking at the sender’s domain name and assessing things like domain age, registrar reputation, etc.