The first skynet ?

[Novalee]

For the first time, scientists have used artificial intelligence to create complex, three-dimensional simulations of the Universe. It's called the Deep Density Displacement Model, or D3M, and it's so fast and so accurate that the astrophysicists who designed it don't even know how it does what it does.


This means the AI could have a flexibility that makes it suited to a range of simulation tasks - although before that happens, the team hopes to figure out how exactly it has managed to do what it does.


"We can be an interesting playground for a machine learner to use to see why this model extrapolates so well, why it extrapolates to elephants instead of just recognising cats and dogs," Ho said.

https://www.sciencealert.com/ai-sim...t-even-its-creators-know-how-it-s-so-accurate

My question is could this be used to make video games ?

 

[Slapped]

For the first time, scientists have used artificial intelligence to create complex, three-dimensional simulations of the Universe. It's called the Deep Density Displacement Model, or D3M, and it's so fast and so accurate that the astrophysicists who designed it don't even know how it does what it does.


This means the AI could have a flexibility that makes it suited to a range of simulation tasks - although before that happens, the team hopes to figure out how exactly it has managed to do what it does.


"We can be an interesting playground for a machine learner to use to see why this model extrapolates so well, why it extrapolates to elephants instead of just recognising cats and dogs," Ho said.

https://www.sciencealert.com/ai-sim...t-even-its-creators-know-how-it-s-so-accurate

My question is could this be used to make video games ?

I think there are far bigger questions to be asked, both scientifically and ethically, than can it be used to make video games.

 

[Danoff]

This is not a fundamentally unique problem when you have a computer solve a problem for you.

I used to do a lot of spacecraft trajectory optimization. You'd feed the optimizer a set of conditions and a target, and it would give you back the "optimal" trajectory to hit the target with the least fuel. Sometimes the answers were truly wild. You'd sit there scratching your head... why did it do that? What a strange phenomenon. Often this resulted from some understanding that the computer had "discovered" about the dynamics of the simulation that was not previously known by man. Then papers get written explaining what the computer found.

There was no intelligence or life in the optimization algorithm, it was just iterating and calculating the numbers that dictate the next step to take.

 

[Liquid]

Skynet was launched in 1969.

 

[Slapped]

This is not a fundamentally unique problem when you have a computer solve a problem for you.

I used to do a lot of spacecraft trajectory optimization. You'd feed the optimizer a set of conditions and a target, and it would give you back the "optimal" trajectory to hit the target with the least fuel. Sometimes the answers were truly wild. You'd sit there scratching your head... why did it do that? What a strange phenomenon. Often this resulted from some understanding that the computer had "discovered" about the dynamics of the simulation that was not previously known by man. Then papers get written explaining what the computer found.

There was no intelligence or life in the optimization algorithm, it was just iterating and calculating the numbers that dictate the next step to take.

I can completely believe that with orbital mechanics that you got some results that looked well whacky especially with n-body problems (where n>2 of course).

However where you were just feeding an algorithm some parameters and then letting it do it's crunching, this AI algorithm is altering it's own parameters, and even then parameters that it hasn't been trained to alter. Where you were asking why it was doing what it's doing, these guys are having to ask how it's managing to do what it's doing given that it's not been trained to do it.

You're right that it's not unique to have a computer solve a problem for you, but the difference here is that the computer itself is deciding on its own what the problem is and then solving it.

 

[Danoff]

I can completely believe that with orbital mechanics that you got some results that looked well whacky especially with n-body problems (where n>2 of course).

However where you were just feeding an algorithm some parameters and then letting it do it's crunching, this AI algorithm is altering it's own parameters, and even then parameters that it hasn't been trained to alter. Where you were asking why it was doing what it's doing, these guys are having to ask how it's managing to do what it's doing given that it's not been trained to do it.

You're right that it's not unique to have a computer solve a problem for you, but the difference here is that the computer itself is deciding on its own what the problem is and then solving it.

Yea I understand there are differences, but the difference between optimization and "artificial intelligence" is not really that big. You can get some unintuitive results that make you scratch your head and push your understanding of the field. That's normal and good and what the whole point of having the computer come up with the solution really is.

You say I was wondering why instead of how, but part of the question really was how - within physics... how is it doing that? This was particularly applicable in complex gravity fields like you might see with a heavy asteroid.