You’re right. A guess might be wrong, but that’s not productive. What matters is how wrong the guesses are and are those wrongs relevant to the problem? Uncertainty exists, but you can be certain that uncertainty also obeys specific mathematical laws (such as the laws of probability theory) and thus the uncertainty is usually quantifiable in some manner. Machine learning algorithms and statistics work because this is the case.
The optimal guess after conditioning on known information and a fairly flexible model is not the same as a typical guess by a person.