Portia spiders do demonstrate trial and error learning, quoting the abstract of [1]:
> All species from the jumping spider genus Portia appear to be predators that specialize at preying on other spiders by invading webs and, through aggressive mimicry gaining dynamic fine control over the resident spider’s behavior. There is evidence that P. fimbriata, P. labiata and P. schultzi derive signals by trial and error. Here, we demonstrate that P. africana is another species that uses a trial and error, or generate and test, algorithm when deriving the aggressive-mimicry signals that will be appropriate in different predator–prey encounters.
It turns out that the species with more variation in encountered prey types are more likely to rely on search, varying over possible patterns until a response is received. Other papers show their ability to learn generalizes beyond mimicry of vibrational patterns. They are also capable of deriving and maintaining situation specific attack routes and plans.
I'll also argue that a fully instinctual repertoire, even without learning, should count as intelligence if flexibly deployed. Consider: despite an inability to learn or adjust, a Nash equilibrium approximating poker bot or an Alpha Zero neural network can be described as encompassing a deep instinct of the game that enables intelligent action selection.
> All species from the jumping spider genus Portia appear to be predators that specialize at preying on other spiders by invading webs and, through aggressive mimicry gaining dynamic fine control over the resident spider’s behavior. There is evidence that P. fimbriata, P. labiata and P. schultzi derive signals by trial and error. Here, we demonstrate that P. africana is another species that uses a trial and error, or generate and test, algorithm when deriving the aggressive-mimicry signals that will be appropriate in different predator–prey encounters.
It turns out that the species with more variation in encountered prey types are more likely to rely on search, varying over possible patterns until a response is received. Other papers show their ability to learn generalizes beyond mimicry of vibrational patterns. They are also capable of deriving and maintaining situation specific attack routes and plans.
I'll also argue that a fully instinctual repertoire, even without learning, should count as intelligence if flexibly deployed. Consider: despite an inability to learn or adjust, a Nash equilibrium approximating poker bot or an Alpha Zero neural network can be described as encompassing a deep instinct of the game that enables intelligent action selection.
[1] https://link.springer.com/article/10.1007/s10164-010-0258-5