For language to work in everyday use, it has to be appropriately ambiguous. Too ambiguous, and communication fails: people don't know what you're talking about. Full precision takes a lot of work, though: it's too hard to do on a regular basis. So words like "probably" and "might" exist as a compromise between ease of expression and expression of content: they tell you something, without making the speaker tell you everything (which would bore everybody even more than what we say now).

People in the intelligence business have to think hard about how to express the likelihood of their analysis. You don't want to make stronger predictions than you can back up, and information is generally limited and uncertain. The continued failure to locate weapons of mass destruction in Iraq, despite strong claims about their existence before the way, provides a recent example. Some government watchers believe that intelligence analysts provided guarded assessments with careful caveats, only to have them turned into definitive statements by policy makers.

One interesting study that Richard Heuer presents in his book (see "Review: Richard Heuer's Psychology of Intelligence Analysis.") is the result of asking 23 NATO officers to read a variety of reports and assign a numeric value to their various expressions of probability. 

This graph from Chapter 12 of his book shows that while the phrases at the extremes like "highly unlikely" were quite consistent, others like "probable" were used to express likelihoods ranging anywhere from 25% to 90%. As Heuer points out, the danger here is that such ambiguity opens the door for the hearer, who is already predisposed toward confirmation, to interpret such expressions in line with their preconceptions. (The shaded areas indicate what Sherman Kent, the first director of CIA's Office of National Estimates, recommended as appropriate ranges for these terms. )