"Previous outcomes" is vague enough. 10, 1000 or 1mil previous spins? Both are "previous outcomes". So typically the video is correct IMO. And Reyth is probably right when saying that "it all comes down to semantics".
Kav, I think semantics are important and are not just about words, but meanings. Not being clear about meaning can lead to misunderstandings and mistakes, bad decisions etc. Besides, it seems like the video is about
semantics, because it's introduced by "Roulette has no memory".
Actually I should have been clearer in my first post. Your sentence:
Following these laws roulette takes into account previous outcomes in order to conform to probabilistic expectations in the long run.
suggests that roulette has some kind of a will because "in order" (in other words, "for the purpose of") conforming to probabilistic expectations it looks back at previous outcomes and adjusts itself accordingly. But what are "these laws" you mention? As I said, ultimately they are physical, but from a statistical POV it's the law of large numbers which determines the long run expectations, not some kind of feedback mechanism.
Obviously roulette isn't an entity but a collection of physical components which must exist in a particular configuration for outcomes to be equally likely in the long run, but if that configuration changes the outcomes will also change, although probably not perceptibly without a detailed statistical analysis. The point being that the outcomes don't change as a result of some entity called "roulette" somehow regulating itself, but because they are correlated with some change in the configuration or conditions.
I would argue that the only version of roulette which really does depend on past outcomes is a pseudo-RNG, because numbers will repeat in predictable cycles.