Helpful analogy from Vitalik for data availability sampling: 

Imagine you're an orange farmer with 5k oranges. Instead of checking each one, you randomly pick 25. If all 25 are ripe, it’s likely the whole batch is ripe

Now imagine you use a special pesticide that ensures the entire harvest is ripe if at least 50% are ripe. After picking 25 random oranges and finding them all ripe, you can start to build confidence that a large portion of the harvest is ripe. If your goal is to ensure at least 50% are ripe, the more random oranges you find that are ripe, the higher your confidence grows, eventually reaching near certainty (99.99%) that at least half of the harvest is ripe

This is similar to how data availability sampling will work in Ethereum

/ethereum foundation | control the memes control the minds paragraph.xyz/@chaskin