The Mantis-1 experiment
The purpose of this experiment is to train you to think probabilistically, and monitor the evolution of your performance. You will be presented with a series of questions, and asked to predict the answer. Throughout the experiment, you will be able to view various statistics indicating how well you are doing, and tips on how to improve your predictions. This experiment aims to show that very little training is required to improve the quality of one's predictions by a large margin.
Your instructions
You will be shown multiple-choice questions with exactly one correct answer. To each answer, you must assign a probability of being the correct answer. If you are unsure of the correct answer, you should assign roughly the same probability to each possible answer. On the other hand, if you are very confident that you know the correct answer, you should put a very high probability on the answer you think is correct. For each question, your score will be the decimal logarithm of the probability you assigned to the correct answer.
Your goal is two-fold. First, maximize your score, by assigning the highest possible probability to the correct answer. Secondly, give calibrated answers, that is to say among answers where you assign 20% probability, exactly 20% must be the correct answer, no more and no less. You can view your scores and calibration on your profile page, and compare with other participants. See the tutorial for some more concrete examples, or start wandering around answering questions, as you wish.
Some questions are quizzes, referencing the past and present. You will be able to see the correct answer and your score immediately after answering them. Others questions are bets / predictions, referencing the future. You can modify your answer until the closure date indicated below the question, and the question's text will indicate when you will be able to see the correct answer.