Algorithmic Recommendations

Until recently I was quite annoyed by the fact that Spotify's "similar artists" did not help me much in expanding my musical comfort zone. This is a quite common flaw with many "peopole who bought this also bought these" recommendations as well. This fact seems to have lead some people to believe that algorithmic recommendations in itself will imprison its users contrary to the more natural exploration that one make through traditional means (friends, music stores, radio, TV, magazines, etc).

I don't subscribe to this fear. Quite on the contrary really.

Even though many recommendation systems today absolutely exhibit this flaw, there is nothing in the idea of software driven recommendations that impel such a shortcoming. It is more likely that the flaw is in the assumptions of the user. If a service or product provider assumes its users will best appriciate consumption within their existing likings, the nature of the recommendations will bind the users within their circles. If one, on the other hand, make the assumption that the user instead likes to explore the unknown, algorithms may still be of great help. The difference is that the assumption of the user will demand a different design of the software and the underlying rules will not be as simple and obvious as for case based recommendations.

I believe the use of exploration friendly recommendation systems will just keep growing. As Spotify now incorporates third party applications, the likelyhood of finding better means of exploration has improved a lot.