We choose the level of $\alpha$ by deciding the probability of a Type I error that we are willing to accept. We do not want to set it too low because we may miss new findings. On the other hand, we do not want to set it too high because we want to be scientifically cautious about incorrectly rejecting a null hypothesis and reporting findings that are suspect.
We also carry out a power analysis before we collect data so that we conduct an experiment that has a reasonable chance of allowing us to reject the null hypothesis and we do not miss new findings.
Putting these two concepts together, which of the following represents a compromise that recognizes scientific caution and the problem of missing new findings?