Alan Turing once wrote on a postcard “Science is a differential equation. Religion is a boundary condition”. A deep and enlightening statement.
Differential equations are beautiful, because they show the relation between (mathematical) objects that interact in a system, and there may be many different resolutions. More than science, I consider life to be a differential equation : many variables, many relations, many solutions, some you can make yours.
This afternoon we had a great meeting with a media tech company, looking at using AI for content recommendation. With Franck, my business partner, we have worked on metadata for pretty much our whole career, and we know what to expect from data. He’s worked in e-commerce, and he knows the algorithms you want to use to get people to buy stuff. On my side as a music lover I’ve grown frustrated with recommendation algorithms used by all music players : they either rely on poor metadata, like release year and music genre, the most advanced may include tempo. There’s an exception to that, which is Pandora : it uses the Music Genome Project that is a super rich database of music annotations that is a work of love and science by crazy musicologists. So you have tons of information to pick from, from lyrics theme to overall mood, music scales and instrumentation. That thing made me buy a lot of CDs back in the days.
On the other side you have playlists : that works pretty much for everything, if John Doe liked this tune and associated it with the tune you just listened to, you will like it too. And in most case it will work, because music is a social experience : the link you make between songs is heavily dependent on context : you heard that song on the radio while camping in 2003, that radio was playing that genre of music with mostly releases of that year, so the next song will probably be a song of the same genre and same decade. Nobody cares if it was a I IV V chord progression with minor harmonies and a Prophet doing the bassline.
There is a strong social component in music. It took me a long time to understand what people liked in genres such as techno or smooth jazz. As a not very social music nerd I used to enjoy exclusively the richness and originality of the arrangement and harmonies, but one day I understood the point I was missing : as I was walking in Manly Beach, there was a band playing what I call “sunny jazz”, slightly upbeat versions of jazz standards and jazzy covers of pop tunes : not incredibly original, but great with drinks and friends. Music that was not loud or interesting enough to force everyone to listen to it, but you could switch easily between singing with the singer, talking with friend and order some finger food. Great combination.
That social experience that is associated with music has a lot of interaction with the music itself : music is representative of the trends of the time, not only through the clothes of the band, but also gimmicks, instrumentations, language. Some bands have been through the ages and navigated the trends, like the Rolling Stones who went from blues purist to swinging London, then to psychedelic and blues again, adding some disco at some point before a long line of pop-rock surfing on self caricature, all with matching guitars, hairdos and fancy clothes. That actually is a problem with typical, bad recommendation algorithms : you may start in 60s Beat with the Graham Bond Organization, Long John Baldry and Screaming Lord Sutch, and suddenly you hit the Stones and the whole thing shifts to Start Me Up, Yes’ Owner of a Lonely Heart and the horrible pop music of that era.
Every experience is a differential equation : it’s a subtle mix of sensible and social experience that interact and it’s generally very difficult to tell them apart. Some specialists can arrange things on explicit parameters, and they could influence a population : a film critic can explain the relation between movies, there are equivalents for paintings or music. Then for social representation, you may want to look like a Parisian intello by repeating what you read in Les Cahiers du Cinéma, pretty much the same way as you have to listen to the latest album from that band on the cover of the New Musical Express. Netflix has tons of data about their movies and series, but there recommendation algorithm has become a joke. It’s not such a big problem, because the only important thing socially is to watch the same content as your colleagues to have something to talk about at the coffee break.
So I like Michel Legrand’s “les Moulins de Mon Coeur” because of its subtle harmonies, but also because it makes me look like an intellectual and I somewhat need to differentiate on that parameter probably because of some kind of insecurity… What a crazy equation !
PS : Looking for the Turing’s quote I also found this article that talks about differential equations, love and beer. Check it out !
PS 2 : in some cases, it’s not even social there are associations that are purely statistical, but repeatable. Reminds me of that crazy website on correlations.