I could give an extreme example and show you what's wrong here. Let's say a kid is born. And we have somehow lost all morality. Kid gets put into a room with food, drink, whatever it needs for physics experiments and left alone. Will the kid ever do, say, the double-slit experiment ? No. In fact it won't even learn how to eat. It'll starve to death. We know this, because people have done that accidentally.
Or another "experiment": show a kid only "parents" that walk on four legs. Will the kid ever learn to walk like a human ? After all, the human body is built so that walking on 2 legs is much more efficient. Again, accidental experiment was carried out, in fact many times. Disney made a completely unrealistic movie about it. Answer: no, the kid will walk on hands and feet. They do not find the optimal solution, despite exploring. Despite being forced into the ultimate form of self-directed learning.
But let's work from simple principles. How about we start by assuming that the brain is neither magical, nor can it come up with information from nothing. It must absorb information from the outside.
You can see, given those easy assumptions that the brain cannot be but almost exclusively a behavior copy machine (which is of course why you seek out help in the first place). Then you see "unschooling" if it works, DOESN'T work because of it's stated principles, but because it copies things from the parents, and/or/later teachers. Unschooling, like any other method frankly, will work incredibly if the parents are geniuses, or otherwise regular contact between the child and very smart people can be maintained over a long time, and will fail spectacularly if that doesn't happen.
Second thing to consider is the information content difference between positive information and negative information, and which random (self-directed) information will be. Positive information is showing a particular example that works, for some process. Negative information is showing that a particular way doesn't work. You can easily see that this depends on whichever is less dense in the solution space (in other words: you should show whichever is rare, not the common case). This will in practice nearly always be the positive information. In other words: the correct solution has lots of information, false tries have very little information about the world.
If you just let the child try, all they'll ever hit is the common case (in the extreme example given above: not eating the food). So you can't let that happen. You MUST show them the correct solution, or put another way you must transform the solution space for them so the rare and common case happen 50-50, and not 1-99999 like would be expected. Again rephrasing: It's very easy to fail, and very hard to succeed. So a "self-directed learning" would only ever fail. You must prevent that if you want them to learn anything at all.
AFTER doing that, yes, you can let them try and explore more by themselves, but it's easy to see that a behavior copy machine like the human brain will be able to learn incredibly faster getting shown the above things compared to trying things out itself.
This means in practice self-directed learning requires the brain to process thousands to millions (or worse: trillions) of times more information to arrive at the same level of knowledge. In the overwhelming majority of cases that'll mean it will never happen.
Third problem is that skills converge. Having one skill makes almost any other skill a tiny little bit easier to learn (even in machine learning, an imagenet trained network can learn faster to make music than a random start network, vastly disparate skills, but the ability to recognize squirrels allows you learn to make music faster). The problem with that is that 0.99^1000 is a very small number. So if you let a kid be slower, the damage that accumulates over time is incredible. If you slightly accelerate a kid, constantly, the advantage that builds over say 10 years is not only enormous, but is at that point STILL compounding. Likewise, if the kid doesn't have skills the damage will compound. Are you seriously going leave that to chance as a parent ? Wouldn't you require extraordinary evidence in favor of unschooling rather than the reverse ?
So self-directed learning is a lie. It cannot work. You cannot explore the world in a self-directed manner and expect to learn much. That doesn't make sense when you consider how the mind must work. But ! There are famous examples where it worked ! It must be able to work. Alas, no. Or at least, there's a simpler explanation. Like most "mysterious"/"natural"/"magical" processes it only works when people cheat. It works when there is another form of learning available to those kids (seeking out help, or having some other form of knowledge easily available to them, like a genius parent).
The factor that confuses people, I think, is that Montessori schools can work despite their bullshit method because of the teachers they hire. As long as that method of instruction is very rare and intriguing they will have more money and more interested, more effective teachers. And that can make a big difference, big enough that the disadvantage of their methods is compensated. But traditional instruction works reasonably effectively despite the teachers being underpaid idiots that don't understand the subjects they're teaching. Is it ideal ? No ? Fun ? Hardly. But you can easily see: traditional instruction with actually good teachers vastly outperforms Montessori schools. It also works much better in the practical situation on the ground that will exist in any large society (that practical situation is, in case you're wondering: 50% of below-average performers, and with current teacher pay, teachers will not even be below average 50% of the time, but much more). But as long as unschooling is done only by intrigued smart people, that will be masked.
The literature on it is unanimous: the thing that matters more than any other in education is how smart the parents are. Second thing is how smart the teachers are. Only third is the method of instruction used. People really hate this, because of course it amplifies class differences, makes IQ "inherited", but that doesn't make it any less true.