Sieve of Aristothenes + induction seem most likely to me.

"For every binary sequence of length X, the Collatz iteration ends at 1. Adding more zeros clearly does not change the result, therefor adding more ones is the only path which might disprove the conjecture." Eventually you build up to a catalog of proven sequences that all other numbers must necessarily step into at some point.