One of the great things about games is their potential for delivering just-in-time instruction.
Most games require some sort of user assistance, whether it be instructions, in-game tutorials, or a help system. Many games integrate help and tutorial information into the gameplay as much as possible. As designers, how can we know when it is is too much? At what point does a design start to keep a player from feeling autonomous because it’s too helpful?
I think the answer to this is going to be different for entertainment games than for serious games, and there is probably variation even within serious games. Games that try to change your attitude, for example, will likely need to be much more cautious about forcing people to spend time in a tutorial than a game trying to teach math. It also depends a great deal on the complexity of the game as well as the complexity of the subject matter.
Personally, I think people should have access to tutorials and other help but there is very little they should be forced to do.
A game like Machinarium has a very short intro tutorial that gets you started. I think it is very well designed. Its “help” pages are incredibly cryptic, on purpose, in keeping with the whole spirit of the game. You even have to do a mini-challenge before you can look at the help pages.
A few years ago, we were asked to design a game to teach concepts in machine learning to middle schoolers. This one presented a huge challenge – there is nothing insightful about watching a machine while it’s learning and without an idea of how the machine is doing what it’s doing, you really can’t get any idea of what this is all about – all you see is that the machine (or program) just gets better at what it does. Also, machine learning is not part of any normal middle school curriculum.
It’s a fairly advanced topic, for good reason: it assumes a fair bit of background in probability, statistics, and CS. Our approach was to design a game where the player is the one who is asked to learn the way a machine does. The player is to create a program for a probe to be launched to Europa. It was not designed to be easy, but the intent was to give kids a feel for how machine learning works, without all the math and CS (a tall order). The publisher kept asking us to make it easier. They wanted us to add all kinds of hints and pop-ups telling players what to do next and we got into a lot of trouble for resisting. 
Making a game easier rarely makes it more fun, and providing the right kind of help in a game can be as difficult as designing the game in the first place. In the end we created a fairly substantial ‘manual’ that players could access whenever they wanted. We had offered to do far more (video tutorials, lesson plans, etc.) but the publisher didn’t want to pay for them. The game is good, but it doesn’t stand alone. The help we created is also good, but insufficient. Good games aren’t cheap.
Far too many educational games out there are nothing more than template games skinned with educational content. I wonder why people make these kinds of games. Is it because they are easy to design, or do they really not understand what it is about a game that makes it a valuable learning experience? Is it games they don’t understand, or instructional design? Both?
Jim Gee says, “When we think of games, we think of fun. When we think of learning we think of work”. (P. 43, 2007, Good Video Games + Good Learning)
Yes. And: No.
Thiagarajan is a well-respected training specialist and designer who specializes in in-class and paper-based learning and development activities. He identifies five critical characteristics of training simulation games that I’ve found quite useful (Thiagarajan & Stolovitch, 1978):
Curricular Ties: Teachers are often under tremendous pressure to make “good” use of their class time. If they can’t show that this game can meet a required curricular objective, their principal is likely to tell them they can’t use it. (This is part of the second pillar)
A number of chapters in my 
