It would be interesting to deploy a network of bots that responds to those scams, posing as gullible targets. That would then significantly increase the workload of scammers and reduce their availability to conduct scams on actual people.
People are notoriously bad at spotting bots. Just a bunch of canned messages can draw people into surprisingly lengthy conversations when they are desperate to get something out of you... The biggest problem would probably be to vary the exchanges enough to prevent them from learning to pick them out after a few exchanges.
Thoughts?