Does symbolic manipulation need to be hard-coded, or can it be learned?

Essay by Yann LeCun and Jacob Browning. The title as it appears is: "What AI Can Tell Us About Intelligence", but this title is misleading. The authors say that the central disagreement between skeptics like Gary Marcus and themselves is "Does symbolic manipulation need to be hard-coded, or can it be learned?"

TL;DR: If our goal is human-like intelligence in AI, it helps if we first understand human intelligence, which we don't. How many of our cognitive functions are innate through genetic evolution, and how many are learned through experience? We don't know for sure, but even knowing that would only move the goalposts; the question would then become, "how many functions need to be innately defined, i.e. hardcoded, to achieve human-level intelligence?"

I'm personally more aligned with the empiricist view that our innate abilities are almost entirely the cognitive capability of learning, and specific functions are almost all learned through experience. Take a look at a newborn baby and make note of how few things it's capable of understanding, recognizing, or predicting. In fact, a baby is capable of only a handful of cognitive tasks: "suckle when something touches my mouth and I'm on my side", "that is a human face of some kind", "grip when anything touches my palm", and... well, that's about it. Everything else requires time to gather data, experiences to learn from. A baby doesn't move away from fire because it recognizes fire, it does so because when it comes close, it starts to hurt -- reinforcement learning. It doesn't learn to talk because it innately manipulates sounds into language, it copies sounds it hears and repeats the ones that result in positive results -- self-supervised (and also just regular supervised) learning. Etc. etc.

In fact, even basic numerical skills that you'd think come naturally aren't innate. Make a row of 4 coins and a row of 5 coins, and ask a 4-year-old which has more. They'll likely say the correct answer, the row of 5. But now spread the smaller row out, leaving gaps that make the row larger than the row of 5, and ask the question again. The child will say the row of 4 now has more coins. Even though they just said it had fewer, and the number of coins hasn't changed; the child will not yet understand the difference between "larger" and "more". We don't have an innate understanding of basic symbolic manipulation even of simple objects; we learn it over time.