James Cridland (of Virgin Radio) compares last.fm (and similar) with Virgin Radio.
Whether algorithms can pick music choices better than a radio station Head of Music is an interesting question. Obviously, last.fm and similar have the leg up in that they can do personalisation, which broadcasting, by its very nature, can't do.
James claims that last.fm-type systems lack the human contact "to give that personal contact once in a while". But last.fm does have human contact, it's just many-to-many rather than one-to-many. They've recently done some work on the peer-to-peer recommendations engine and the groups are inherently social. However, I'd agree that these features haven't really been made the most of yet (recommendations don't yet directly affect a users personal radio station).
James says that radio stations are good at 'recogniseability' - this is true. To stop listeners flicking stations constantly, big pop stations need to play a high quotient of recognisable tunes. Often, these are songs that a listener would be hard pressed to name many of, since they're often embedded within our collective consciousness without us really being too aware of them. In this way, radio can act as a memory prompt ('oh I remember that tune!').
However, it would seem to me that last.fm is specifically against recognisability. Their stated aim seems to be to help users broaden their musical horizon, with back-catalog, album tracks and obscure artists all part of the growing reportoire of streamable music. Last.fm is perhaps, dare I say it, a long tail business, not in direct competition with popular music radio stations.
Long term, the algorithms for personalised radio stations will improve rapidly. After all, they have a very precise way of measuring success: the percentage of tracks 'skipped', 'blocked' and 'loved' by all of their users. Radio stations have to be content with data from a much smaller subset of their listeners.
Once critical mass has been achieved, personalised radio will be able to overtake broadcast radio in being able to provide people with continuous, high-quality background music, a function which many broadcast radio stations currently fulfil.
The only two stumbling blocks I can see for personalised radio are music licensing (currently a big thorn as the music industry hasn't yet adapted to new distribution mechanisms) and distribution (which might be solved with ubiquitous low-cost wi-fi networks or somehow syncing personalised radio data back to the iPod).
This isn't to predict the death of radio, however. Radio is more than just the distribution of music. It's also a way of building an intimate connection between broadcaster and listener, and of delivering high-quality non-music content. This kind of listening, which isn't background-driven but focused, attention-holding content, will continue to thrive. People don't listen to radio just for the music, but also to hear personality, stories and local interest. And whilst this kind of radio will continue to experiment with distribution (podcasts, etc) and feedback channels (e-mail, texts, websites, etc), the central essence of this radio will live on.
In short, traditional radio will become more like Radio 4, local and community radio, and personalised, socially-driven radio will take over from Radio 1.
'A VC' compares Pandora vs Last.Fm and concludes that Last.Fm is better because "I don't want a computer recommending music to me. I want other people, people who share my taste in music, recommending music to me."