The learner isn't sure after how many correct responses they will be reinforced, so they keep on producing the desired responses.
For example, suppose you're gambling at a roulette or poker machine. You keep thinking "After this one, I'm definitely going to win a fortune." You press the button, but you don't get any money. So you think, "Fine, let me try once more. This one's going to be it, I can feel it!" You press the button, and end up losing money. So basically, you know you're going to be reinforced, but you don't know after how many correct responses you'll be reinforced, so you keep on gambling.
Therefore, variable ratio is the most resistant to extinction, followed by fixed ratio and then variable interval and then fixed interval.