The strict answer to your question has to do with humans being social, motivation, and strong AI, I think. However, if we allow teachers to stay in the picture

The strict answer to your question has to do with humans being social, motivation, and strong AI, I think. However, if we allow teachers to stay in the picture, software could (and should I think) be used more to support learning. It could improve students' abilities significantly if done sensibly. Data-mining would probably be useful to tune such a program, but I wouldn't expect it to outcompete teachers on the structuring of materials and answering students' questions any time soon. The question on how to grade students is fascinating. There's an argument for one long exam at the end of your studies: you get the grade corresponding to your performance at the end of your studies. If you "get it" late in the course you're not penalised for earlier performance and, more importantly, it encourages you to take a long-term perspective on what you learn. I've experienced both teacher + frequent non-standard test assessment in school and one shot at two weeks of exams after 3 years at university. The former is problematic because grades become uncalibrated (thus losing much of their intended usefulness). My main concern with the latter is that you can end up testing things you don't intend. Calculating the grade from appropriate data captured over a long time would be great -- comparisons and profiles could be both more detailed and more stable than what a teacher could provide. (There are some things it couldn't test, like handling the pressure of final exams.) Do you think computers should have replaced teachers? Why do you think that hasn't happend?