[meroes]

If we’re neural nets, shouldn’t 12,000 years of post paleolithic experience be enough to have more modern emotions?

All you’re doing is forcing people to choose something other than magic. You could name almost anything and they’d have to choose it over magic since magic doesn’t exist.

This is not convincing at all.

[randomdata]

> If we’re neural nets, shouldn’t 12,000 years of post paleolithic experience be enough to have more modern emotions?

As far as I can tell humans are already able to survive and reproduce to the limits of what the physical human form is capable of. Optimization can only optimize so much. Where do you see "paleolithic" emotions being a limiting factor that leaves room for further optimization? What "modern" emotions do you envision to improve on those metrics?

Human emotions do not appear to be cohesive from person to person, so it seems the generic algorithm is still doing its thing, but if the mutations are no more effective than the "paleolithic" emotions with respect to survivability and reproduction, there isn't much evolutionary pressure to see them become dominant.

[meroes]

Well the parent was saying our emotions are 10k-2.5m+ years old, but I claim society is exponentially more complex since agriculture. Either our emotions have kept pace or they haven’t. If they have, then the Paleolithic nature of our emotions makes no sense. And if they haven’t, it takes away from the neural net idea.

Things like accepting a surgeon and anesthesiologist putting us into suspended animation and opening our hearts we have learned to not freak out about. Or that a single person could nuke the entire planet including our families and we go about our lives. I propose those are recent learned emotional responses. Nevethertheless, it still seems too slow to compare it to a neural net.

[randomdata]

The parent stated that the neural net is trained with a genetic algorithm that seeks to maximize survivability and reproduction. Of which I posit we have optimized as far as it can take it. People have shown to already survive and reproduce to the greatest extent of what appears to be the limits of physical form as we understand it.

So, if we have reached the limits of survivability and reproducibility, only further hindered by other biological processes, what pressure on the generic algorithm would there be to see new traits start to become dominant? If you effectively stop training the NN, you wouldn't expect inference to change.

Perhaps the advent of modern contraception will eventually reveal some mutation that is advantageous enough to overcome the modern setback we've seen in reproduction, bringing to light a different emotion/set of emotions that become dominant. But, realistically, we're only talking a few generations of training on that change to the state of the universe. That's nothing for a genetic algorithm. GA-based training methods are not very efficient with respect to training speed even on computers, and worse in real life when a single generation takes ~30 years to spawn the next.

[exe34]

the medial prefrontal cortex literally reaches into the hypothalamus to inhibit the more primal emotions. people with damage in that area or who are tired and at their wits end, often revert back to earlier, more primitive emotions. it seems like you equate neural nets with artificial neural nets. there is no back propagation in natural selection. it's a genetic algorithm.

[exe34]

> If we’re neural nets, shouldn’t 12,000 years of post paleolithic experience be enough to have more modern emotions

does that mean you believe the Coelacanths should have evolved feet by now?

evolution does not work that way. usually new structures are laid out on top, new behaviours come about that can override older ones at certain times and at other times, "instinct" takes over, and an old program is running again.

[meroes]

That’s sounding much more biological than a neural net now. Neural nets are much quicker to adapt to new parameters. Coelacanths and humans aren’t. Which was my point if 12,000 years isn’t enough whereas neural nets are rapidly changeable, maybe we can’t do this equivalence in calling humans neural nets.

[exe34]

are you maybe mentally inserting "artificial" in front of "neural network"? if so, do you also see GPUs in biology?