The betting strategy in Loki is a simple approach to enable the use of the more sophisticated hand evaluation system. A preferable approach would be to redesign the system and attempt to make decisions based on computation of expected values (perhaps by simulations playing out the game many times). However, in our expert-strategy dependent architecture there are several more simple refinements that could be made to improve performance. Two categories of refinements are unpredictability and deception.
Unpredictability is a simple addition that makes it harder for the opponent to build a model of the program's play. Consider adding unpredictability to the betting based on an EHS' decision: we could use a linear scale so that we bet 50% of the time with a 0.50 hand, 0% of the time with a 0.40 hand and 100% of the time with 0.60 hand.
Deception includes strategies such as pure bluffing, slowplaying and check-raising. It causes the opponent to make wrong assumptions about the current state of the game. Deception can also be used to make a play that does not necessarily lead to the highest expected value for the current game but rather is intended as ``false advertising" to indirectly lead to increased profits in future hands. To be complete, all possible actions that one can witness from Loki should have a dual interpretation so no conclusions can be made with certainty. For example, in the present system (without check-raising), if Loki checks, a knowledgeable observer can infer that Loki has a hand with EHS' < 0.5.
We have included some of these deceptive strategies successfully in later versions: check-raising, pure bluffing (betting with a weak hand on the river) and balancing raises (sometimes raising instead of calling). They are currently used unpredictably (randomly). We also occasionally check-raise with a mediocre hand, so our opponents cannot infer that we always hold a strong hand when we check-raise.