Of course, this could be part of a sequence that goes
7, 9, __ 13, 15, 17, __ , 21, 23, 25, __ , 29, __ , 33, etc., in which case no number is missing in that portion of the sequence. (In this case it would have been a series of incrementing odd numbers from 7 where we skip odd numbers that match the prime status of the previous number.)
It's for a similar reasons that pattern recognition in AI should never be considered "set in stone" but should only be given a probability. The probability was very high that the missing number was 19, but this should be subject to change on discovery of further information.
Another example happens all the time to engineers. If you happen to sample the amplitude of a 400 Hz sine wave at time increments of once per second you could get a sequence of 0, 0, 0, 0 ... etc. If you think it's a mistake and sample it 200 times a second you could still get 7, 7, 7, 7, etc., or 0, 0, 0, 0, etc., depending on when the wave starts its peaks and troughs. But you could sample it at some odd number of increments, or accelerate and decelerate the time sampling increments and discover it is a 400 Hz sine wave.