I suggested something here the other day. Generally it’s good to focus on something with recognizable economic payoff to justify the significant effort that these things require at this stage.
However, I have the impression that this question is meant to be more general. “Which is good and which isn’t”, i.e. lets find a broad set of computational problems that are suitable for HTM and a complementary set where we’re sure they don’t qualify.
Well that’s not too hard, you just find yourself an evolutionary path where, starting from today’s conventional computational landscape for want of a better starting point, you can make incremental changes that gradually morph each of today’s existing, currently economically viable solutions into an HTM-based solution, all the while trying to maintain their viability as long as possible. Solutions will gradually fall by the wayside the way species become extinct, only for the fittest to remain. That’s a holistic approach for ending up with a pretty exhaustive set of winners.
The result will depend somewhat on the path chosen, but the process as a whole is much better than the current “lets throw it against a wall and see if it sticks” approach that the HTM community out there is prone to take.