If these are significance values then they correspond to:
90%, 99%, 95%, 99.9% respectively level of confidence.
Think of the bell-curve and the size of its two tails. The larger the level of confidence the smaller the tails. The null hypothesis is rejected if the investigator finds the result of a statistical investigation lies in the tails. If the test results lie in the main body of the distribution the null hypothesis is not rejected. Type I error occurs if the null hypothesis is wrongly rejected.
90% level of confidence means that in 9 out of 10 tests the null hypothesis would not be rejected, but is likely to be rejected in 1 out of 10 tests. 99.9% level of confidence means that in 999 out of 1000 tests the null hypothesis would not be rejected. Only in 1 out 1000 tests is it likely that the null hypothesis would be rejected.
So Type I errors are less likely to occur the smaller the tails of the distribution are, and this is true when the significance level is very small, in this case, 0.001.