Also, evolution is likely not precise enough to have evolved accurate intuitions about the fitness value of all possible risks. We should expect some reasonably broad degree of randomness around the value of risk taking.
>"What gain do colourful feathers on birds really have except to draw attention"
When drawing attention gives you a disproportionate advantage to attracting a mate vs. attracting a predator, it's a very important (and non-neglible) gain.
>We should expect some reasonably broad degree of randomness around the value of risk taking.
I don't think this negates the point. Just because the there is a distribution in the value doesn't mean there isn't a statistically significant directionality of that distribution. I can say there's a distribution of individual player height in the NBA, but that doesn't mean we can't draw conclusions about height having generalized value at the population level regarding the chances of making it to professional basketball.
Humans have no natural predators, so one side of the equation is zero, and the other non-zero.
> Just because the there is a distribution in the value doesn't mean there isn't a statistically significant directionality of that distribution.
Yes, and there is directionality in risk-taking as well. Discoveries, fortunes and high-value mates all require risk taking.
Even risks that appear to have negligible or even zero fitness value, like extreme sports, have netted many people valuable sponsorships or YouTube fame and fortune.
Evolution is not precise and simply cannot capture the full nuance of a concept like "status" in human culture, therefore it has permitted a broad distribution of risk taking.
Of course, you want an adversarial network that's difficult to trick, otherwise the training will produce suboptimal results.
