The link doesn't work, but I imagine he would have figured that there is a 1/10 chance of a 3 being rolled, assuming there are ten different numbers on the die.
Since he rolled the die 20 times, and 4 of those rolls were 3's, the experimental probability of him rolling a 3 are 1/5.
What I think he should do is: try the experiment 3 times, then gather the data of the experiments, and make his prediction.
