It doesn't seem to be a "might" but a "probably definitely". Couldn't find the link right now but one of the few actually uses of AI that I've seen was to map out animal languages (specifically dolphins and other cetaceans) to develop a translator, Something something if you throw different (human) languages at a space and then dimension reduce it you get quite similar structures even though the languages are vastly different on the surface, and things like dolphins apparently aren't that far off and definitely not less complex.

Or, put differently: Yes, they're actually building a universal translator based on the assumption that because language-capable beings end up speaking roughly similar things you get structural overlaps if you have a sufficiently abstract representation of language (such as a neural net that learned to distinguish it from other stuff).

Aside from that it's been known for a longer time that dolphins are capable of relating complex information to another, e.g. you put one in one pool, the other in another, they can hear but not see another, and they can coordinate pressing buttons in one pool to get at fish in the other.

Also dolphins can recognise that a human gal is afraid of their teeth, disarm themselves with a tennis ball, and thus succeed in their task to get a handjob. That was part of Lilly's programme to teach dolphins English (they really struggle with consonants) which is a book honestly everyone should have read. Don't ask me which book in particular involves an injured dolphin co-habituating with the experimenter (aforementioned gal).