A fallacy in non-deductive reasoning

If we think about the large variation of appearances among donkeys, it is very probable that:

1. Some donkeys look very much like horses.

Now we run this argument:

2. D is a donkey that looks very much like a horse. (Existential instantiation on 1)
3. D looks very much like a horse. (By 2)
4. D is a horse. (Probabilistic inference from 3)
5. Some donkey is a horses.

We might make the inference from (3) to (4) more explicit by making use of:

6. If D looks like a horse, probably D is a horse.

I think this is related to Rowe's argument from evil.