For the moment I'm going to respectfully disagree with both of you here. I'd think that the state space complexity of battle-mode would be much higher than completing a time trial, as the latter has so many reproduceable elements and the complexity of the former is increased by player input. Computationally wouldn't a time trial be as trivial as starting with a record run and then optimizing? The task isn't even complex enough to bother with an AI implementation. You might call the strategy cheating, but isn't that just what the human is trying to do within the confines of miserable biological reproducibility?
A single study by a researcher with limited resources and motivation is not the final word on AI capability. The AI versus fighter pilot article I linked to did essentially emphasise the complexity of the state space:
Because a simulated fighter jet produces so much data for interpretation, it is not always obvious which manoeuvre is most advantageous or, indeed, at what point a weapon should be fired....
"Here, you've got an AI system that seems to be able to deal with the air-to-air environment, which is extraordinarily dynamic, has an extraordinary number of parameters and, in the paper, more than holds its own against a skilled and capable, experienced combat pilot,"
but you want me to believe that an MK64 time trial is more complex?
For the moment I'm going to respectfully disagree with both of you here. I'd think that the state space complexity of battle-mode would be much higher than completing a time trial, as the latter has so many reproduceable elements and the complexity of the former is increased by player input. Computationally wouldn't a time trial be as trivial as starting with a record run and then optimizing? The task isn't even complex enough to bother with an AI implementation. You might call the strategy cheating, but isn't that just what the human is trying to do within the confines of miserable biological reproducibility?
A single study by a researcher with limited resources and motivation is not the final word on AI capability. The AI versus fighter pilot article I linked to did essentially emphasise the complexity of the state space:
> Because a simulated fighter jet produces so much data for interpretation, it is not always obvious which manoeuvre is most advantageous or, indeed, at what point a weapon should be fired....
> "Here, you've got an AI system that seems to be able to deal with the air-to-air environment, which is extraordinarily dynamic, has an extraordinary number of parameters and, in the paper, more than holds its own against a skilled and capable, experienced combat pilot,"
but you want me to believe that an MK64 time trial is more complex?
Just a couple of points that I feel need to be said:
Top Time Trial strategies are the opposite, where top human strategies do approach perfection, so even though the complexity of Time Trial is lower, that of its top strategies is far higher, so AIs don't discover unassisted the long-range strategies people do, like creative shortcuts and lap skips.
If a script is iterating on a top strategy then it may under certain conditions outperform the top strategy, but I wouldn't call that an artificial intelligence competing with a top player, I'd call it a top player competing with a top player where one side of this self-competition is assisted by a tool.
People don't discover the long range strategies unassisted either. Once one person comes up with a breakthrough that information is available to everyone else. Everyone else copies and optimises it. I think you're suggesting pitting one AI against all players, rather than one AI system versus the realistic performance of one human. I wouldn't be surprised if an AI could outperform the entirety of humanity starting from scratch with zero help anyway, but that is a different test.
Also, every person or institution has limited resources and motivation. If that in some way discredits the Mario Kart AI, it does the same for the fighter pilot AI developed under similar conditions, and thus discredits the stated basis for which you think MK64 Time Trials will soon be within reach of AI.
One would reasonably expect significantly more resources and motivation to apply to a military application like fighter pilot training (and making fighter pilots necessarily redundant) than to playing Mario Kart. When I say 'limited resources and motivation' I mean relative to a research goal that's actually useful.
Just a couple of points that I feel need to be said:
> Top Time Trial strategies are the opposite, where top human strategies do approach perfection, so even though the complexity of Time Trial is lower, that of its top strategies is far higher, so AIs don't discover unassisted the long-range strategies people do, like creative shortcuts and lap skips.
> If a script is iterating on a top strategy then it may under certain conditions outperform the top strategy, but I wouldn't call that an artificial intelligence competing with a top player, I'd call it a top player competing with a top player where one side of this self-competition is assisted by a tool.
People don't discover the long range strategies unassisted either. Once one person comes up with a breakthrough that information is available to everyone else. Everyone else copies and optimises it. I think you're suggesting pitting one AI against all players, rather than one AI system versus the realistic performance of one human. I wouldn't be surprised if an AI could outperform the entirety of humanity starting from scratch with zero help anyway, but that is a different test.
> Also, every person or institution has limited resources and motivation. If that in some way discredits the Mario Kart AI, it does the same for the fighter pilot AI developed under similar conditions, and thus discredits the stated basis for which you think MK64 Time Trials will soon be within reach of AI.
One would reasonably expect significantly more resources and motivation to apply to a military application like fighter pilot training (and making fighter pilots necessarily redundant) than to playing Mario Kart. When I say 'limited resources and motivation' I mean relative to a research goal that's actually useful.
17 comments
For the moment I'm going to respectfully disagree with both of you here. I'd think that the state space complexity of battle-mode would be much higher than completing a time trial, as the latter has so many reproduceable elements and the complexity of the former is increased by player input. Computationally wouldn't a time trial be as trivial as starting with a record run and then optimizing? The task isn't even complex enough to bother with an AI implementation. You might call the strategy cheating, but isn't that just what the human is trying to do within the confines of miserable biological reproducibility?
A single study by a researcher with limited resources and motivation is not the final word on AI capability. The AI versus fighter pilot article I linked to did essentially emphasise the complexity of the state space:
but you want me to believe that an MK64 time trial is more complex?
Just a couple of points that I feel need to be said:
People don't discover the long range strategies unassisted either. Once one person comes up with a breakthrough that information is available to everyone else. Everyone else copies and optimises it. I think you're suggesting pitting one AI against all players, rather than one AI system versus the realistic performance of one human. I wouldn't be surprised if an AI could outperform the entirety of humanity starting from scratch with zero help anyway, but that is a different test.
One would reasonably expect significantly more resources and motivation to apply to a military application like fighter pilot training (and making fighter pilots necessarily redundant) than to playing Mario Kart. When I say 'limited resources and motivation' I mean relative to a research goal that's actually useful.