I think it’s hard enough to teach yourself things because of several reasons:
You don’t know where to start, where is the finish line, and how to methodically move from the start to finish. Too many easy materials is boring. But jumping to harder materials without first mastering the prerequisite is equally frustrating.
Thus, the existence of a sequence of materials that will smooth out the process of climbing the learning curve is indispensable. The sequence should “horizontally” and “vertically” comprehensive. Horizontally comprehensive means the materials include all the sub-domain that makes up the totality of a certain domain of knowledge. Vertically comprehensive means that for each of those sub-domain, the sequence will be enough to take you from beginner up towards the expert level on the frontier of existing knowledge, able to contribute in producing new knowledge for the greater community of that domain.
You don’t really know, thus cannot distinguish, what is wrong and right. Thus making it difficult to do correction. Even if you know, the correction might not be as fast as it should be.
Different domain has different sensitivity on the importance of feedback loop on learning performance. In programming, you know when you are wrong. The program simply won’t run or compile. Thus the proliferation of self-taught programmers. In math, you don’t really know if you are on the right track. You need teachers, mentors, or at the very least peers that know more than you do.
Thus, you need to simulate an environment where feedback loop is readily available. The frequency and quality of feedback loop matters. An entrepreneur that consult every quarter to Silicon Valley will have a different frequency of feedback loop with an entrepreneur that lives there that has the access of daily feedback from the available tacit knowledge that is geographically concentrated there (be it in form of VC, mentor, or peer entrepreneur). Higher frequency = lower interval between each feedback loop. The difference will be 30x (maybe even more because of economies of scale, maybe less because of diminishing return).
Rewards and Incentives
You are not rewarded (or punished) for your progress (or the lack thereof) because there are no external system in which you are liable for the reward and punishment. It is also hard to persevere because you need motivation as the fuel to carry on from the start to finish. It is a very painful process to learn a new thing and it is very tempting to stop midway. After all, nobody is forcing us to keep learning it.
Thus, it is important to design rewards and incentives yourself, be it externally sourced or internally generated. External ones can be winning grants or any other recognition of your work from other people or some agency. Or it can simply having generated a product, say, a fleet of self-flying drone utilizing swarm algorithm that you can see and touch as a trophy of your progress. Internal ones could include realizing that you are smarter because you can tackle problems that frustrate you in the past with relative ease now. Or just enjoying the process itself. You can be more hardcore than that if you successfully train yourself to have the discipline of keep going, without any artificial rewards of incentives.
Regardless of those natural hurdles, self-taught person is a very filtered segment of people because only those with the comfort to endure within the lack of structure to find the structure himself, patching the built-in lack of feedback loop with other substitute, and persevere nonetheless without reward and incentives that are required by mere mortals.