For example, let's assume that according to our old statistical research average wage of business analyst in US is 68125 dollars per month. Now we wonder if the average is still the same. So our null hypothesis is that the average is still equal to 68125 dollars, while the alternative hypothesis says that it's not equal to this number. We then can take samples and then ...
Wait a minute. Isn't it fool's agenda to try to support the alternative hypothesis when we can see from the start, that it has 100% probability to be true? Even more, if our samples will suggest that the null hypothesis holds, then most likely, such conclusion would be false.
Just think about this way. We can imagine the old average wage as number on number line. If we were to randomly to chose any number near this number on the number line, then what is the probability that we will pick our old average? Zero! There are infinitely many numbers near the old average on the number line. In other words, we have 0% probability that the current average wage will fluctuate to exactly the same value as our old average wage. Consequently it means that there is 100% probability that the alternative hypothesis is true. Yes, new average wage can be close to the old one (like just few dollars more or less), but it would be different nonetheless.