AI Bots?

[hunkygladiator]

I wonder how Square will be handling the advent of AI bots that outplay players? Not tomorrow of course, but the time is coming when 'rent-an-ai' will be available; or worse, for sale. Neil Tyson put it well in a recent commentary with words to the effect that soon AI will be able to duplicate any photograph, any audible sounds including human voices, any motion pictures, and make short shrift of most passwords in seconds. He said it portends the death of the Internet. So then, to be allowed to play, we have to not be AI. Given that, how do we prove we aren't perfect, against a creation that is perfect at proving it is not perfect?

Shooter

It depends on what you mean by "AI bots".

AI in general, as in artificial intelligence? It's already a thing. Or else what would you call the existing bots...? Some of them do rotations better than 99% of players.

AI as in specifically machine-learning based ones?

1. There is no money to be made making ML/DL based bots since the marginal benefit is extremely low (see the point about bots already doing very well).

2. Exactly how are you going to get the training set? It seems prohibitively costly to get a sufficiently large training set for playing one particular video game compared to just scraping Reddit.

3. How are you going to update the bot for new content? Do you repeat part 2 every single time?

AI bots already exist in Ultimate raiding.

There was a chinese group that made a youtube video a while back that showed their AI bots completing DSR

That video is fake. What isn't fake however is the existence of rotation bots that can do entire rotations for you. There are a lot of fake videos of "cheats" or "bots" in raiding out there, though I'm seriously unsure what is the point of making those fake videos.

Tbf even the Google CEO said he cant explain how AI works, and it is an experiment they unleashed into the world without really knowing what it does, while the ethics team got fired or left/reduced. I can understand why some states will put restrictions, especially when it will destroy a lot of jobs.

But yeah I dont think OP understands how AI usually gets trained and how difficult it would be for a game like this. This isnt even near or midterm future

Why would the Google CEO know how DL models work? It isn't his job and he's not trained in it. And all this brouhaha over AI being a black box is as inane as saying that 358982351 * 8234609182^3527 is a black box. It's literally just a lot of chained regressions exploiting generalizations of Stone-Weierstrass. If someone wants to put in the effort they can examine the weights at every layer.

Furthermore, there's already a lot of research and tools to explain AI models. e.g.: https://github.com/slundberg/shap

The only reason AI will upend society is not because governments ignore the threat but because they literally don't understand it and will therefore likely institute horrible policies.