Differential privacy anonymizes data sets with deliberately noisy data

I saw the complaints about anonymized census data. Is the problem that spoofed data can be rendered unusable for some purposes, or is it just that it becomes harder to use without a data science background?