That may be true, but Negroponte was proselytizing streaming media back in the early 90s, long before anything like Spotify or Netflix were actually practical. Did all this hot air have any effect? I have no idea. But from where I was sitting, he was definitely pushing hard for something that others were not at the time.

Also, the first music recommendation service I ever heard of was started at the Media Lab. It was called Ringo, and you'd email it a list of your favorite bands, and it would then email you back recommendations for other music you might like. Ringo was implemented using some sort of statistical correlation amongst other people's lists of favorite bands.

Ringo was eventually spun off into a company called Firefly which ran a music community website that used this technology. It didn't do so well, though, and was eventually bought by Microsoft, which shut down the website about a year later.

> Like isn't streaming music based on preferences pretty easy?

Sure, but I don't think that doing it very well is. I mean they typically do an okay job, but I've yet to find one that just blows me away with its ability to find music for me that I will love.

Yes, I'm introduced to some good new music this way. But I also have to listen to a lot of dreck too.

#1 is "It must have all of the music", which is really a lawyer problem, not an engineering problem.

(#2-#11 are just variations of "No no, I mean all the music", "Does it have really old music, like X?" "What about really niche music like Y?" "Oh, oh, what about Z? I bet it doesn't have Z.")

Well, maybe that's why they generally are not so great.

On the other hand, Pandora has traditionally existed with playing music that you'll like, but didn't explicitly select, as its reason for existence. Maybe that's why it seems to do a better job at this than other streaming services.

That must be why Spotify constantly fills my once useful “discover weekly” feed with what I would describe as “Ambient Indian Buddhist Trip-Hop”. Not really what I wanted at all. Though this week it seems to be “Ambient Irish Bagpipe”, which is also not at all what I wanted.

So, no... I think Spotify has a long ways to go to learn how to give me what I want...

No, I don't think so.

Millions upon millions have been spent pouring money into recommendation engines (which is an advanced "based on preferences"). Companies like Netflix know that they can keep customers that much longer if they're good at finding the needles in the haystack for each specific customer, and they famously offered a million-dollar bounty to anyone that could improve their algorithm by even a few percentage points (and, just as famously, everyone recognized that it would be worth far more than a million dollars to them).

Pandora tried to do it by creating a "music genome" for every single song in its database, with dozens of attributes per song, thus linking disparate songs though much more than just "genre."

Nowadays it's pretty much "throw a massive database at a ML system, and pump out 'people who like these 10 songs tend to like this song'", but for a long time those systems were crappy. And, even now, its hard for those systems to find a balance between showing you something that matches what you've listened to, and playing you ten nearly-identical songs in a row.