to what ? scrypt ?
LTC already use that tech 


If for some reason that's what needed, sure, why not?
That's the point. If the proof of work algorithm were to change, the SHA256 ASICs wouldn't hold any power.
This would be a hard fork, probably with temporarily two competing blockchain version.
But if the Bitcoin's survival depend on it, everyone with a vested interest will just switch to another proof of work algorithm
(Even if some miners don't like it.)
Furthermore, we would have a bunch of Litecoin miners that would gladly use their processing power to mine bitcoins.
Marvellous, isn't it?

If by marvelous you mean blocks that take 100 times longer to complete until difficulty adjusts back to GPU level . . . Which would take maybe 6 years, if we hit 2 Ph/s in in the next few months, which I bet we do. Also, in a competition, how would GPU's ever out compete ASICS?