Teaching machines to diagnose cancer, because without that, there's no way we are going to solve cancer generally. There just aren't enough pathologists. Harder that going to Mars, probably easier than world peace. Definitely involves an enormous amount of data. Like, a non-trivial fraction of world-wide storage.
There is no one cause of cancer. While there are a number of well-recognized precipitants (HPV, trichloroethylene, UV radiation, etc), cancer is fundamentally a disease of disorder in the genetic code. As entropy invariably increases, the integrity of each of the trillions of individual somatic genomes per human start to degrade, they rarely degrade into a positive feedback loop of replication, and we have not invested enough of our evolution to prevent it. For example: mice don't normally get much cancer because they die before they have time for it. Elephants don't get much cancer because they accumulated enough copies of p53 over their evolution that they don't have as many cancers per N cells.
Humans have 1 copy of p53 from each parent and have extended their lifespan so many ways that their one copy of p53 is no longer enough. Lifelong accumulation of genetic entropy is inevitable. The game is to catch the offending cells and kill them.
