I would think 48 suggests quite severe ASD, but it only suggests it.
50 is the maximum possible score on the AQ test. The general population scores 17 on average, but the test's author thinks that 32 is a better threshold. I guess that's because of the difference between actual autism and autism that is deemed worthy of intervention, and I suppose that's linked to money, in the same way as the law usually saves money by pretending there is no crime below a certain arbitrary level of seriousness.
So, assuming linearity, a score of 48 would indicate 94% "real" ASD or 89% "bureaucratic" ASD, which seems quite high by either measure, but as has already been said, the AQ test isn't diagnostic. In my case I can get practically any score I like on the AQ test, without actually lying, because of its forced-choice format and the demand for black-and-white answers to some very grey, subjective questions. You get the choice between agreeing, agreeing strongly, disagreeing, or disagreeing strongly with each statement, but the score calculation ignores the "strength" factor completely.