I'm not entirely sure why they needed neural networks to come to this conclusion, "spontaneous number neurons" or not. You'd only have to look at eg. the Pirahã language to come to the same conclusion; it doesn't have numerals and speakers were generally "mathematically illiterate" up until recent times, and teaching any sort of mathematics (ie. including counting) to a monolingual Pirahã speaker was apparently extremely challenging.

Seeing specific neurons or subnets being activated in relation to numerical tasks and assuming that it means numeral sense is somehow innate is honestly a bit of an odd conclusion to make. Yeah, subnets specialize, but that doesn't mean their specialization is necessarily built into us – just that we have the capacity to do so.

[–] Oldmandan@lemmy.ca 4 points 1 year ago

IIRC, there is a bit more complexity than that to the Pirahã understanding of numeracy. Relative quantity is something they're just fine at understanding, (with words for single/less, plural/more and same) it's abstraction of quantities to tokenized values where they struggle. Which, I suppose, also interestingly lines up with the study results; the initial training period resulted in nodes associated with quantity, but those nodes were separate/unrelated to numeracy systems that developed with additional training.

[–] Umbrias 1 points 1 year ago

There are a lot of brain systems and a lot of complexity. Studies confirming something which is otherwise generally understood is par for the course. But until it's tested, it's dangerous to build theories off of it.