Even implementing something simple and well known (like a breadth first search for example) on the white board can take an hour, if you are discussing it as you go along, and the algorithm has to be tailored and optimized to fit the interviewer's question.
Sure, you could spend all day or all week talking about it. But the question is, do you need to, and is it a useful thing to do in a job interview setting. But I'm biased here, because I'm generally not a fan of day long job interviews. I don't think they're necessary or practical, at least for most roles.
If it's a whiteboard thing, then don't expect a detailed, working implementation in code. Ask for a sketch / high level overview / pseudo-code that shows the candidate understand it well enough that you can infer that they could implement it with a compiler / google / book / etc.
OR
If you want an implementation, give them a machine, let them use their language of choice, use google, etc., and let them actually code it up, making and fixing errors as they go, etc. I actually like this because it lets you see how someone approaches debugging, etc.
In either case, you could supplement the exercise with some discussion questions "why /when would you use BFS?" etc.