Getting people to make decisions that directly impact them is a good way of figuring out which beliefs they actually use to predict the behavior of the world and which beliefs they only profess but in some way know are false.

Consider someone who claims "I have a dragon in my garage.". When someone asks to test that claim by looking, you say it's an invisible dragon. When they then say they'll listen to the dragon breathing, you say the dragon doesn't need to breathe. If they offer to throw a bag of flour over it and see the shape of the flour-covered dragon, you say the dragon is permeable to flour. Despite saying you have a dragon in your garage, all your predictions are based on not having a dragon in your garage. If you actually thought you had a dragon in your garage, you'd say "sure, let's see"; you should be perfectly willing to test that claim and accept the result. (Credit to http://lesswrong.com/lw/i4/belief_in_belief/ for the dragon example.)

The same lines of experimental questioning can generally distinguish between "I truly believe this and use it as a model to predict how the world works" and "I profess this but when it comes time to predict how the world works I do so using a model that actually works.".

On the other hand, if you build a system that rewards people for how strongly they profess their beliefs, and how much confirmation and support they get from others professing the same beliefs, you can give strong positive reinforcement to beliefs that do not model the world.