The canonical game attempting to be computer hardened is https://en.m.wikipedia.org/wiki/Arimaa and it did pretty well... for a while. Admittedly there's less human expertise than chess, but we've been surpassed.
Thanks, this led me down a rabbit hole. Some people now seem to suggest Magic: The Gathering as the last bastion for humanity. But it's not clear if that counts as "simple"...
> An important factor when playing Magic: The Gathering is choosing which cards one should play with. This is an example of a combinatorial optimization problem in a large and bewildering search space.
(There's also a lot of stuff about MTG being Turing complete, which I think is kinda irrelevant to practical AI play.)
Your question would make a good answer for itself. Ask the AI to design a game in which a human would be better than the AI. And then we'll see if your human readers can do better than what the AI comes up with.
Also: how, pray tell, did researchers determine the attractiveness of the primates in that study? Tinder? But seriously... Human standards aren't that relevant, so on what basis do they claim the primates didn't select for attractiveness?
(You'd have to quantify "better" differently - the "best" dnd game is one that's spontaneous and fun. Maybe have the players rate each other at the end of a session and the highest score wins?)
another argument against 100% end-of-term exams from a statistics/econometrics viewpoint: assume the students actual grasp of the topic is the true but unobservable population parameter we seek to estimate using some examination-format. Estimating the true parameter from one observation (i.e. the one-off end-term exam) should yield a lot more variance (i.e. if we we're to repeat the one-off example in countless parallel universes and write down the distribution of exam grades) than a weighed estimation from multiple observations (i.e. a composite grade from homework, mid- and end-term exams). Thus, from an empirical standpoint I'd consider the composite grade a better estimation of the students 'true' grasp of the topic.
Agreed, although I think someone might argue that you're trading off bias with variance in the sense that you'd ideally measure student understanding at the end of the semester, so when you average in measurements from the middle, that increases bias as well as decreasing variance. (Still, from this perspective the optimal tradeoff to minimize total expected error is unlikely to be 100% final...)
Humans are better at soccer
Tyler Cowen links to a book that covers most of what you're asking about the Industrial Revolution today on MR.
Thank you! In case anyone else is wondering, this is the link: https://marginalrevolution.com/marginalrevolution/2022/06/rubin-and-koyama-on-the-industrial-revolution.html
The canonical game attempting to be computer hardened is https://en.m.wikipedia.org/wiki/Arimaa and it did pretty well... for a while. Admittedly there's less human expertise than chess, but we've been surpassed.
Thanks, this led me down a rabbit hole. Some people now seem to suggest Magic: The Gathering as the last bastion for humanity. But it's not clear if that counts as "simple"...
I have moderately bad news for you: https://mobile.twitter.com/rcsaxe/status/1480742617998667785
Okay it's not a world champion and it didn't play but I think drafting is higher skill than playing probably.
As a magic the gathering beyond-enthusiast, I'd like to point out that actually *running* a game of magic would be incredibly easy for a computer...
Why do they think that it'd be difficult for a computer to play Magic??
I'm not an expert at all, but here's a quote from https://ntnuopen.ntnu.no/ntnu-xmlui/handle/11250/2462429:
> An important factor when playing Magic: The Gathering is choosing which cards one should play with. This is an example of a combinatorial optimization problem in a large and bewildering search space.
(There's also a lot of stuff about MTG being Turing complete, which I think is kinda irrelevant to practical AI play.)
Your question would make a good answer for itself. Ask the AI to design a game in which a human would be better than the AI. And then we'll see if your human readers can do better than what the AI comes up with.
Also: how, pray tell, did researchers determine the attractiveness of the primates in that study? Tinder? But seriously... Human standards aren't that relevant, so on what basis do they claim the primates didn't select for attractiveness?
Humans are better at DnD and Calvinball
(You'd have to quantify "better" differently - the "best" dnd game is one that's spontaneous and fun. Maybe have the players rate each other at the end of a session and the highest score wins?)
You know, when you think about it, my "game to make a game" almost *is* Calvinball, isn't it?
another argument against 100% end-of-term exams from a statistics/econometrics viewpoint: assume the students actual grasp of the topic is the true but unobservable population parameter we seek to estimate using some examination-format. Estimating the true parameter from one observation (i.e. the one-off end-term exam) should yield a lot more variance (i.e. if we we're to repeat the one-off example in countless parallel universes and write down the distribution of exam grades) than a weighed estimation from multiple observations (i.e. a composite grade from homework, mid- and end-term exams). Thus, from an empirical standpoint I'd consider the composite grade a better estimation of the students 'true' grasp of the topic.
Agreed, although I think someone might argue that you're trading off bias with variance in the sense that you'd ideally measure student understanding at the end of the semester, so when you average in measurements from the middle, that increases bias as well as decreasing variance. (Still, from this perspective the optimal tradeoff to minimize total expected error is unlikely to be 100% final...)
This post has (some) of what you’re looking for re: Industrial Revolution
https://daviskedrosky.substack.com/p/the-last-crusade?r=rrmvm&utm_medium=ios
Typo: after the paragraph break, the next line starts in the middle of a sentence.
"no one will sue you even if your advice was bad.
doesn’t, they risk being held negligent."
Thanks, fixed!