*sigh*
It gives you an idea of what is LIKELY to happen - not what WILL happen.
Your little dice example makes no sense here. On a 6-sided die, we know that the chance of getting a 1 is 1/6. Even if you roll a 1 a hundred times in a row, a generalization can not be made that you will typically get a 1 since we know all the possible outcomes and the likelihood of them happening.
Of course, believing that every P. murinus you encounter will be highly defensive and always try to bite you isn't accurate. However, realizing that the typical temperament of these spiders is rather defensive can help you be prepared to own one, handle one, etc.
Generalizations are made in this context in order to allow preparedness and decision-making.
Until the spider in question has been held, all we can do is give examples of the typical behavior of the species. No one ever said that that particular spider would exhibit those characteristics, but there is a likely chance that it will.
Those who bothered to actually read the words I used will recognize the difference between what I said and what you said I said. I said "the die comes up one most of the time" not "the die will come up one next time". I specifically made the point to distinguish between descriptive and prescriptive statistics, which is why I brought the dice example up in the first place. I don't need a lecture on statistics. I taught and tutored statistics at the college level.
Also, I thought I was pretty clear on the "generalizations" issue. I never said to completely disregard the generalizations. I said they can give you a range of observed outcomes and that they are less useful than direct, personal observation of the tarantula in question.