Good point, maybe a better way of phrasing it would have been "It's virtually impossible to predict which tools will _not_ be useful" :)
Fast Fourier transforms, complex numbers, complexity classes, etc. are clearly useful. It's much harder to rigorously claim that a result will _never_ be practical or useful.
Just to be clear, I think it's fair to say that a result isn't _immediately_ useful. But who can say whether or not it will have a more practical application down the road? Only time will tell.
On a long enough timeline, you're right! But if we limit ourselves to, say, the next 5,000 years, it gets a lot easier to accomplish. ;)