Machine Learning And Choice

To follow up from the streaming music post: something I have been thinking about is the consequences of a machine learning algorithm on musical taste.

There are these ideas out there that you go through musical stages in life. I am not really convinced by the methods used in that study. And I don’t think it asks the right question. What causes musical preference?

I am out of my depth on this one, so I have more questions than answers.

Here is a theory that I couldn’t find in searching, but I know I heard from somewhere (a podcast, a ted talk…not sure). The theory is: Music tastes are path dependent. Meaning that they do not converge the way that the “stages” model describes…or at least, I think their approach imposed too many preconceptions about what music is–their definitions are something along the lines of: “Intense” (punk/metal) for teenagers.

OK…if they say so.

I would go more micro than that. When do (if) you accept synthesized music over instrumental? I think that dimension was excluded entirely. What types of accents do you like in singers?

There is also a finding out there that music becomes likeable simply via repetition. So songs that you hear over and over eventually become good.

Machine Learning

So we now have a thing that analyses some dimensions of .mp3 meta-data. It looks at the data of what we have listened to, and it suggests similar songs–similar is defined in the black box. The algos essentially try to establish links in your preferences (as if listening to a song is a signal of preference at all) to other music. You can train this algo with ‘likes’, etc…

BUT! What if you get caught in some sort of infinite loop? Like the algo defines your sweet spot so well that you have no need to browse elsewhere, to expose yourself to other types of music.

AND! What about when you are in the mood for a different type of music…the algo has almost no data on mood (YET!(1984 man! NSA! Snowden! Like…this aggression will not stand man.))

I have heard other theories about the way music evolved to ‘fit’ the environment it is played in. So chamber music had pianos and Cellos and shit, stadium rock was all about being louder, so big amps and electric guitars, and now most music is mastered for iTunes, and Kanye somehow is the best.

So what if there is value in being a traveller across music genres (btw wtf is a genre?), across mediums, and across soundscapes (different instruments producing different frequencies, at different intervals).

If your music tastes are evolutionary, what does this tool–that evolves to collect and suggest music that mimics your past tastes–do to you?

I mean, you will be listening to ‘new’ music. But it’s the same old shit, right?

Where is that WTF!? factor that makes art good? And good here is the idea that there is value in being exposed to shit you have not been exposed to, because it makes you think more broadly, and your broader tastes will help you solve riddles or some shit.

Also neuroplasticity fights alzheimers…Like…so Algos cause Alzheimers. Bro Science, bro.

Also. I don’t. know. where… to begin. But. Cheap shot!!! Protect your balls! Metal is a bunch of loud sounds, and some people think its real manly, but like what makes loud and angry manly? Nothing. Grow up.

How do magnets work?

Happy listening bro-dude-girl-baby-lady-boy 😉

Spotify vs. Google Music Play (GPM)

[UPDATE 7/10/2015]
This guy wrote a more clear (better) article on the same topic:
 http://laurencekellett.co.uk/spotify-vs-google-play-music-rather-inconclusive-comparison/
Now that I have learned Spotify a little better, there are certain things I am missing from GPM, one of them is the radio. The automatic radio stations on GPM are not an option in Spotify, like I had initially believed. I might be switching back for this…but I don’t know yet. Spotify is my app for the time being.
————-Old Below————————————————————————————–
I was keeping this note in my Evernote, and figured it might be useful to someone else.
Overall
I am still in a trial period with Spotify, and I won’t really make a full move until that is over…so in a couple months. But if you want a quick answer, I recommend Spotify over GPM, but it isn’t an outright winner, GPM is better for certain things.
 
Price: Equal $9.99/mo.
 
Search:
  • Google wins. I don’t have a rigorous method, but I don’t need one. I was searching for an artist and a song. Putting the words in random order still gets you to the thing you are looking for in Google play. Spotify returns the artist, but not the song. And when I searched for both at the same time, nothing came up.
    • Search querry: “Boyz Noize Avalanche” stumped Spotify, Google didn’t even blink.
Sound Quality:
Winner:spotify
  • At first listen I thought that Spotify might sound better, but when I made sure the volumes were the same and ran through a number of tracks, I concluded that if there are differences, I can’t distinguish.
  • At second glance, there is a feature to turn up the quality in spotify, so my test to compare was gpm  to the lowest quality stream in spotify, so spotify has higher quality sound.
Liking music: TBD. 
  • Spotify has a starred playlist, but it is hard to get to, and it doesn’t seem like you are able to even star a song in the Android app. It’s too much effort, one way or the other.
  • I have really enjoyed using the GPM to create a “moving playlist”. I simply hit the thumbs up, and voila, it goes to my playlist, thumbs down takes it off. The mechanism is so simple  that I have really grown to appreciate this feature.
  • Spotify has artist tracking…which is great, I currently have songkick to do this for me, so GPM is a bit harder to like an artist without actually adding something of that artists’ work to your library…which then makes your library harder to navigate because it is larger.

Curated Playlists

  • Spotify seems to have this one. Spotify simply has more stations that show up when you start looking. It also does a better job of laying them out by mood or genre. GPM has a tonne of radio stations, and no great way to interface with them. They seem to give you some stock ones that are good, but there isn’t a great catolouge of something to really explore.
  • More on playlists in GPM. COM allows you to save any playlist you are listening to for offline use, Spotify doesn’t let you save radio stations.

Random Playlist

  • TBD the randomly generated playlists by GPM are very impressive, they don’t mix genres, so the flow is good. And I have discovered a lot of great music with this.
  • Since this is a machine learning thing, I have to give Spotify more data before I can fairly compare these two features…
    • but my gut tells me GPM will win this, just because they are likely better at programming machine learning algos…they already have better search, after all. But who knows. If Spotify has more users, which I believe it does, they might have better data to program from…how well that translates to the individual is questionable. And ultimately it will be a hard category to evaluate…since it depends on my emotional reaction to the playlist anyway.
  • As previously mentioned, Gpm lets you save radio playlists, and download them for offline use, not so with spotify
Selection
  • Friends (local musicians) are in gpm but not spotify, could be a function of how new spotify is to Canada.
  • Spotify in theory has more music, but in practice both libraries are so large that this feature just doesn’t matter in a shootout.

Organizing music

  • Neither service is a clear winner.
  • GPM  As I mentioned has a really simple thumbs up/down tool, and I love this feature
  • Both let you save albums, artists, songs
    • Adding songs automatically adds them to your albums
    • Adding albums automatically adds all tracks to your songs
    • I don’t like this, as I don’t want to browse through single-song albums nor do I want my songs playlist to be more about singles than the less “stand alone” tracks that populate many albums
  • Both let you create as many playlists as you want, but doing so adds songs to your library, which then makes it harder to navigate
  • Biggest issue these streaming sites both have is the lack of a smart (programmed) way to keep your weaker interests out of the search of your favourites. But Google’s thumbs up is a genius step in the right direction

Navigation

Google Has Better Site Navigation

  • Spotify has a more complex menu system, GPM only gives you that card display to go through your music
    • Btw the classic iTunes triple list system is still the gold standard for me…but it’s not a streaming service yet
  • It’s a bit tedious to navigate from the artist page to the album page to the song page in Spotify, GPM is much more simple.

An excerpt from one of my ongoing debates

Quick and dirty: The other person is arguing from a stance that I categorise this way: favours individual decision making in the hands of the individual, and strongly rejects the notion that policy makers have a paternalistic role. While I think there is a lot of merit in this position, I weakly reject it. Disclaimer: I can’t actually speak for the other debater, it’s better if you read in between the lines, and you know, generally think for yourself. I generally believe thinking for yourself is good.

The blog post is going to start at the numbered list, not with anything written prior. The reader will have to infer what the other person’s position is. This is part of a dialectic, because I have been learning about Socrates.

In response to an email, I wrote the following:

  • We are part of a swarm, with distributed intelligence.
    1. This is actually the fundamental insight of Adam Smith and the basis of many arguments put forward by Hayek/von Mises/Friedman (the Austrian/Chicago economics schools).
    2. Consider a machine learning algorithm. It can make great predictions, without understanding causation.
      1. You should try to read some material on this stuff. It’s the biggest and least well understood disruption in modern business. A lot of people talk about ‘Big Data’ but only a handful of people even know what it means. Half of my individual intellectual endeavors today are focused on to use machine learning techniques because I (an economist(ish)) am about to be made obsolete.
        1. So many tangents! The other half of my intellectual efforts are the source of these many debates, as I feel the only other value add economist have is the philosophical side…it really is a branch of philosophy more than it is a science.
      2. This theoretical framework you put forward, suffers from the informational load required by the individual.
        1. Education is a necessary step for an individual to have the ability to:
          1. Recognize they are ignorant:
            1. The more you learn, the less you realize you understand. Put differently, education raises more questions than it answers. I think this is a fundamental law. For, in practice, there is an infinite amount of knowledge.
          2. At the same time, I offer a (albeit incoherent) philosophical alternative that doesn’t address (yet) the faults of micro-political battles, and mechanisms in which people can game the system for personal gain.
            1. I feel that you have a much better “reason” based footing, and have spent more time on the problem of political philosophy, but I believe your arguments advance a position that, ultimately, is either ill-prepared or not-yet communicated to offer guidance on such nuanced implications of aspects outlined in this response…but that is why there is utility in this dialect.
  • In a practical sense, modern research offers us a lot of answers as to where and how individuals are able to make certain decisions.
    1. The study and cataloging of cognitive biases has provided people with the tools to either exploit the uneducated or to act paternalistically
      1. There is a natural/practical binary here, since the only moral pathway you have left open in your theoretical framework is one in which it is only moral to educate the individual, but since there is a cost associated with education…this is not a practical solution.
        1. We can go and find a million case studies to show that there is a natural way that this game ends, and it isn’t with education. I took a whole graduate level course that focused on these problems, and every time I bring them up I fail to impart the importance of this insight.
  • Prominent Evolutionary Biologists (See E.O. Wilson) have described the human being as one that is internally conflicted, and also not *naturally* reasoning creatures.
    1. We have conflicting internal reward systems for pro and anti social behavior
      1. The history of the homo sapiens, in evolutionary terms, is one in which strategies of survival at various times were individualistic (the most primitive regions of the brain) and collective (the regions of the brain that only humans and a few other species even have…very rarely occurs)…and yet, there were also oscillations between individualism/collectivism that continue to this day, in terms of evolutionary development.
        1. The new regions of the brain, reason dictates, had to serve some role for survival
          1. Language/communication allowed for better survival or fitness
            1. This is a collective strategy à(implies) language improves collective fitness
              1. Language is a collective tool, it’s hard to imagine what role it would occupy as something other than a collective tool. Ex: Talking to myself, out loud, with specialized vocal patterns helps me accomplish what? To provide better fitness in what way?
            2. (hotly debated) The language/communication part of the brain predates reasoned thinking à(implies) reason is not a survival function, but rather a byproduct of evolutionary history
          2. *naturally* this is a loaded/imperfect term. Don’t read too much into it. Essentially, there are no evolutionary forces that modern researchers have discovered (that have shown up in what I have read) that support reason as the mechanism of fitness in homo sapiens. Reason is a by-product. It is cognitively expensive, and our brains don’t naturally use this path of thinking unless posed with significant incentives/stimulus to do so.
            1. For me to say something dumb/but more or less what I think: THIS IS WHY LIBERTARIANSIM WILL NEVER WORK It relies on people utilizing a function of their biology that only ever kicks in when the incentives are strong enough.
            2. For this reason governments or corporations or any institution, really, can exploit any homo sapiens by using environmental factors that allow for the path of least resistance, cognitively speaking…This the most fundamental reason I believe marketing to be incredibly immoral. It is willful use of another individuals’ weakness to extract profit/rents.
              1. When it comes to judging people as moral/immoral, somehow, ignorance is an acceptable excuse for me. I don’t believe the typical person employed in marketing is aware, nor would be equipped to correct for this feature in their work.
              2. Nor do I believe it is fair to single out marketers, as my own work has many of the same issues, and I would like to not be judged harshly, either.
              3. Somehow the only correction for these problems is institutional, since it is systems that allow us to best utilize reason as a species. Institutions are guaranteed to be flawed in terms of libertarian doctrines, because by default they limit individual power (I am aware this is a leap in logic).
  • (Emphasis that this is a Belief) Ignorant conservatism does a decent job of stabilizing the most radical effects of ignorant progressives, but does not offer improvements to the status quo
    1. A point of agreement in our philosophy is that society should not throw out the distributed intelligence that has accumulated in the form of customs and culture just because you have a fancy model that tells you something else will work better
    2. In fact, reliance on new models has a long record of failure. Those who claim to be more reasonable or more enlightened are always somewhat wrong. (Ex. The residential school system in Canada, or the soviet union, etc…)
    3. At the same time it is a mistake to say that the current arrangement is a just or good
      1. Distribution of wealth is not merit based, which is a problem for stability. (Current liberal elite academic research) People are finding clever ways to show how anti-social behavior is caused by individual disenfranchisement.
        1. It is possible that this is not a failure of libertarian ideology, but rather a failure to practice libertarian ideals
          1. It is fundamentally still an unknown assumption, and probably will always remain such that society functions better or worse with more freedom at the individual level, and I suspect that even the way this is worded is terribly problematic. Especially since it is likely, not at all a linear path of causation.
  • Reason is a flawed tool for public policy, because it requires assumptions of cause and effect.
    1. Again, since a lot of human society functions, and nobody knows why, perhaps a better approach to society building is one that sets goals, and experiments on how to achieve those goals rather than one that tries to start from the principles of justice.
      1. Isn’t this already how it works?
    2. There is incredible moral hazard, admittedly, if this were a mainstream political ideology, in the wrong hands, could lead to fascism
      1. Can’t define “wrong hands”
    3. Libertarians would run this country into the ground because they are too wedded to the idea of individual responsibility, without accepting that individuals are not biologically nor socially prepared or able to make every decision.
      1. (Something you can maybe clear up for me) Libertarianism fails to go into how the individual should govern him/herself…which makes it a nice relativistic doctrine, but ultimately why the Kantian ethics don’t necessarily lead to libertarianism.

Mutually assured erections

The title has nothing to do with the post. I’m just immature.

Actually, I wanted to call it Mutually Assured Destruction, because I nuked a ten year friendship over the weekend. But I’m more or less fine. So it wasn’t my destruction. And I’m sure my friend is fine, or will be.

On friendship. On bullshit, onomatopoeia…uhhhhhh. So what the fuck happened?

Long story. The easiest starting point is that friendship is a loosey goosey term, and people throw it around because they are too embarrassed to be honest about their relationships.

I think a true friend is someone who wants the best for you, even at the expense of themselves. In other words, someone who loves you. Sometimes love leads to erections. mostly not tho, sadness.

But there are all these other people that don’t fit that definition, that we all call friend, guy. So lets try some categories. Key word: try.

1) People you share interests with:

This is easy, you like that person, in some limited fashion, largely because there are few truly individual pursuits, and a buddy that can share your passions is great…Do I need to say more? You know this.

2) People you have chemistry with:

Let me just struggle through this one. The personality is a real(ish) thing, and some personalities gel better with other personalities. Real competitive people like to compete with other real competitive people, and sometimes they like admirers, and sometimes people like to admire people. There are bazillions of combinatrixs of personalities. But sometimes its just easy to be around someone, and that works for a while. But it doesn’t get you to love.

3) People you just know, because you spend forever with them because of circumstance

Time has a funny way of revealing the nuances of someones’ character. I think we naturally just like familiarity. And nostalgia for past gives us warm glow. So people have a tendency to think this means friendship. It doesn’t.

4) People you don’t really like, but are usefull

This happens. I think we are really embarrassed to admit this one, and it is culturally punished to reveal this stance to another person. These are people we use. I of course don’t do this.

5) People we admire, want to emulate / People we feel we can teach

This one is maybe just people we share passions with, but it seems a little more specific.

There are easily more categories…but why keep categorizing?

The categories are situations, and all of them can be the foundation of a real true friendship, but none of the necessarily get you there.

It really is about love, dude, bro, man, girl, sweetie.

What’s love got to do, got to do with it? Ok. It ultimately is about whether you want that other person to be their most fulfilled person. And you really have to want it. Because you can sit there and posture all day on how you possibly might want this for everyone, hence everyone is yo fwiend. But you don’t want this, not at the expense of yo fwiend time with that other person, do you?

This is where is gets tough. Why did I nuke a friendship? Because I didn’t feel that anything mattered but form, substance was lacking. Pick one of the categories, or make up a new one. Ultimately, there is a lot of expectation that gets built up over time. You share a passion, and when those passions diverge…if you aren’t true friends, that feels like a huge loss, and you want to pull that other person towards you, even at some cost to that other persons’ passions, and so there is some damage done here to yo fwiend.

To be fair, you might do that, pull your friend in the wrong direction, if you love the person, people fuck up, but ultimately you hopefully correct that mistake. See the bird for the feathers.

And ultimately there is a whole dimension of depth. You can love people to varying degrees, but there is one fundamental truth that has to be maintained…you have to put the other person before yourself, in your own mind, to some degree. You have to enjoy seeing the other person flourishing. Despite your own sense of loss that you may not be the most central figure in that flourishing. You take joy in that shit. It’s like a mutually assured erection that way. They get a success boner, and you get a remote-success boner (I’m not aware of any women actually reading this, but you get lady versions of boners. What is that? Like really wet? Forgive me I’m clueless. Everyone is hot to trot, glory).

What you don’t get is resentment. Or the sadz. You sure as hell don’t interfere and try to make the other person feel guilt. People are responsible for their own happiness, and it takes a unique set of circumstances for a given person to be happy. Whatever the fuck happiness is. So friends help friends be happy.

Now in closing, it is a long life, and there is no need to waste one of those 5 categories…or whatever categories. If you have a relationship that can be a friendship…don’t be butthurt if it isn’t the case today. And be honest with yourself. If you want it. Great. If not. O well. If on second thought, you actually love somebody, and you be being a dinkus, you got time, just own the mistake and make it better.

Some Politics

Something I thought about on the long drive is what I would do with power.

  1. I would structure every policy based on rigorous vetting, and if no data exists, we go to random assignment of policies to see which ones work. The metrics would perhaps not be traditional ones.
  2. The ideological guiding principle would be the enablement of people, not maximizing social welfare.
    1. Enablement of people would have to control for a multitude of factors, but it is ultimately expanding each individuals’ potential production frontier to some unknown limit — which we would hope correlates with satisfaction with life. Economic models fail against the data in this regard, and utility theory should be thrown out.
    2. failure to achieve this task is understood, it’s simply the ideal that drives micro-decisions, and the principle question that needs to be asked in all policy debates
  3. If people take on risk, they get to enjoy the benefit or eat the consequences. This means rich people can get wiped out, and if they use other peoples’ money or well being without properly communicating risk and obtaining consent, it is a criminal charge.
  4. SWF are messed up utilitarian garbage that force people to use shitty data to measure quality of life, but if measuring money, law makers would be biased away from concentrated losses for distributed gains, on one end, and biased away from distributed losses for concentrated gains at the other end. These two principles are foremost after individual enablement.
    1. We would not be biased towards maximizing social welfare, because Pareto efficiency is an amoral principle.
  5. Politicians would run in vocational races, ex) The minister of health is an election for health administrators.
  6. Taxation is shifted away from punishing desirable activities like earning an income, and more aligned with punishing anti-social behavior. Activities that have concentrated gains, and distributed losses, for example are taxed heavily.
  7. Subsidies more or less go away, unless there are distributed gains and concentrated losses. Subsidies must be Pareto neutral or as best as can be determined. Experimental treatments will determine more detailed policy decisions, not precedent.

This is a work in progress, but political camps such as libertarianism and collectivism are both elements somehow, capitalism w.r.t property rights still seems to work well, would not be abandoned, but property rights would be expanded to include nuanced and non-market goods. Let this serve as a spontaneous breakdown of my political biases. Is it a rightist or leftist position? How useless is our current political language and classification system?

Best Death

I thought about the best time to die. I concluded that it depends on whether you have given up on goals or not.

In the case of giving up, you would spend all of your resources on increasingly enjoyable experiences, and you would have a false tooth with a lethal poison or explosive to bite down on. You wait until the most thrilling experience is at hand, and just before you do it, with the highest level of anticipation, you bite the tooth. Because anticipation is at a better time to go than disappointment, the actual climax is the point just before doing something exciting.

In the case of not giving up, you die unexpectedly or as a martyr for the cause. Julius Caesar style.

Thus are the best deaths.

Worst death is WWI pointlessness, where your life was made obviously worthless to the people who control your death.