That’s actually a very interesting question. You’d want a problem which:
is either embarrassingly parallel or large enough to get a decent speedup,
involves a fair amount of complex branching and logic, such that GPGPU would be unsuitable,
cannot be efficiently solved by “shared nothing”, message-passing systems, such as Beowulf clusters and grid computing.
The link also states that the aim should be “to help society, to help others” and to “make the world a better, more interesting place”. Here’s a start; in fact, many of these problems are fairly relevant to AI.
That’s actually a very interesting question. You’d want a problem which:
is either embarrassingly parallel or large enough to get a decent speedup,
involves a fair amount of complex branching and logic, such that GPGPU would be unsuitable,
cannot be efficiently solved by “shared nothing”, message-passing systems, such as Beowulf clusters and grid computing.
The link also states that the aim should be “to help society, to help others” and to “make the world a better, more interesting place”. Here’s a start; in fact, many of these problems are fairly relevant to AI.