Blind insight refers to an unconscious learning. A subject may not be consciously aware that they are making more informed decisions based on the conditioning of experience with the same problem set, but it can be analysed from their success rate.
In the Iowa contestants were instructed to complete gambling tasks wherein some decks were subtly more advantageous than other. It was shown that on average a person would hone in on the good decks versus the bad decks after 50-60 flips. However, as soon as the 10th flip they would show measurable unconscious apprehension when flipping a bad deck.

There is prevalent misunderstanding that we know the antecedents to our actions and feelings and this has been shown in cognitive psychology to be false. Much of our confidence is a post-hoc rationalisation of why we behave in a certain manner, rather than what is true about the motivation.
In a study participants were presented with seemingly random letters which actually followed a consciously undetectable series of rules. The participants were instructed to continue the series in a similar manner and despite testifying to hold no confidence that what they had written made any sense they were still correct 60-65% of the time.

Ryan Scott aimed to modify this artificial grammar study to show the opposite disassociation to blind sight wherein the subjects metacognition was high despite a no better than chance performance. Effectively the subject would guess truly randomly and yet have an ability to determine whether their random guess was right or not. The speculation as how this could be possible is that the analytical performance at chance was bested by their intuition when asked to rate their confidence level in the answer.