I was recently reading about Langton's ant and the related conjecture which states that for every initial configuration, the ant eventually starts building a 'highway'.
I also read that it has been shown that Langton's ant can be used to simulate any boolean circuit thus the ant is capable of universal computation.
But I am confused by those two facts because if the conjecture is true, then every computation performed by the ant will eventually end up to a highway which is easily detectable thus making any problem related to Langton's ant decidable. But how can this be true if Langton's ant is computationaly equivalent to a Turing machine for which there exist many undecidable problems?
I also read this paper and it specifically says in section 1.2: "If the conjecture is true, then any problem associated to the ant, whose input is an initial configuration with finite support, is decidable" whereas in section 1.3 it states: "There are undecidable problems associated to the behavior of the ant."
I suspect that my confusion on the matter has something to do with the finiteness/infiniteness of the input since in section 4 of the previous paper it says that Langton's ant universality is "a rather weak notion of universality (which requires an infinite - but finitely described - configuration)" which I'm not entirely sure I understand.
So my question is what's wrong with the above reasoning which suggests that the highway conjecture should be false (otherwise there would be no undecidable problems)?