The only thing that really matters is that the physics are kept consistent over the course of the 10 million or so runs for each scenario. Even if the rules are non-physical with regard to our niche little human environment, the evolution of the agents' tactics can be studied just as effectively. Edit 2: if anyone needs another (arbitrary) example of this, consider imposing the non-physical condition that both seekers and hiders can elect to travel down into the floor, and can then spawn above the map, where they then experience free fall. Let's say they could then move quite a bit laterally as they fall from the ceiling, at least initially, so that the seekers could invade the hiders via a third dimension. So I'm running that scenario in my head, and at some point, it evolves into something like this, where the seekers are just chasing the hiders through the third dimension unless some other constraints get imposed that allow the hiders a defense in the third dimension. And yes, unforeseen defense tactics could crop up again. Let me know if none of this makes any sense to you (YES, YOU!). I don't think that the conditions here incentivized that, though. It's "don't get seen", not "... and also make sure you have the most available space to move around while you're not getting seen". It's tough to summarize the evolution leading up to us humans, but we probably had on the order of 1 billion "generations". Maybe even up to ~1 trillion(? if some of the early phases were particularly quick, like single-celled lifeforms). But now we have eyeballs, this really weird concoction of exclusively organic matter that we're nowhere even close to matching with any technology (except for the NEW Apple iPhone 11, click here to buy one today!). So I think that each generational leap in learning/adapting for the virtual hiders and seekers is very inefficient for the experiments that they're running compared to us animals driven to do pretty much whatever it takes to survive. And good luck modeling how that survival instinct can affect the decision making process. I still don't think Millennials will see "true" artificial intelligence in their lifetimes. Edit: guess I should also state that I think the development of something as complex as the eye is relatively comparable to the development of intelligence. Both are extreme specializations selected for by the environmental stressors, no?I think the box surfing is a modelling bug; how do they move a box while they're on top of it?
The hiders never "learned" to imprison the seekers which highlights a misconception about natural selection. The process doesn't find the best solution, it randomly finds advantages and favours them.
If I haven't recommended Richard Wrangham's Catching Fire I've been remiss; he starts out by pointing out that evolution has had a puny effect on humans as we know them because if you go to Mile High Stadium, with your mother to the right of you and every subsequent row a generation further advanced, and you all stand up and do The Wave, by the time that "wave" gets back to you it will have blown clear through Australopithecus and the person to your left will have lived their life in a tree. If humans were fruit flies our entire evolutionary history would play out in like eight weeks. blackbootz suggested Joseph Heinrick's the Secret of our Success and he was spot on. There's a formidable advantage in culture whereby if you can learn something from yourself, your children will adapt it and by a hundred generations away, your species will do it. But if you can learn something from someone else, everyone's children will adapt it and by three generations away, your species will do it. "Eye" and "intelligence" are both on a spectrum. Is the light-sensing organ on top of reptilian heads an "eye"? it's all a lot of critters have but it's also a lot less than the reptiles themselves have. How about a compound eye compared to the eye of an owl? "true" artificial intelligence in the eyes of everyone isn't a computer that answers questions, it's a golem.The only thing that really matters is that the physics are kept consistent over the course of the 10 million or so runs for each scenario. Even if the rules are non-physical with regard to our niche little human environment, the evolution of the agents' tactics can be studied just as effectively.
It's tough to summarize the evolution leading up to us humans, but we probably had on the order of 1 billion "generations".
So I think that each generational leap in learning/adapting for the virtual hiders and seekers is very inefficient for the experiments that they're running compared to us animals driven to do pretty much whatever it takes to survive.