Sandia's Agent Bayes project faces a nine-month test that asks whether AI can help scientists perform Bayesian reasoning.