Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
AI Music Google Open Source Robotics Software News Entertainment Hardware Technology

Google's 'Project Magenta' Art Machine Composes Its First Song (thenextweb.com) 72

An anonymous reader writes: Google's Project Magenta, which aims to use machine learning to create music and art, just created its first song. The song, which can be more appropriately described as a 90-second melody, is quite simplistic and reminiscent of an old Nokia ringtone. It's impressive for a machine! Magenta is built on top of its TensorFlow system, and all the open-sourced materials one could ever need are available through its Github. The team wants to be able to tell stories from the art it creates similar to that of artists. "The design of models that learn to construct long narrative arcs is important not only for music and art generation, but also areas like language modeling, where it remains a challenge to carry meaning even across a long paragraph, much less whole stories," the team wrote. "Attention models like the Show, Attend and Tell point to one promising direction, but this remains a very challenging task."
This discussion has been archived. No new comments can be posted.

Google's 'Project Magenta' Art Machine Composes Its First Song

Comments Filter:
  • I wonder (Score:4, Interesting)

    by John Smith ( 4340437 ) on Wednesday June 01, 2016 @06:04PM (#52228967)
    If we could use deep learning to let computers learn styles and patterns, eventually incorporating them into new music. It could usher in a new era where every film is composed by John Williams, or Mahler's tenth is finished, or there's a new Bach and Beethoven being made every day. Of course, on the arguably darker side, pop music could become entirely computer designed, although considering the quality it would actually sound better if done by computer.
    • by Anonymous Coward

      No, you can't until you have a proper music theory. Otherwise, surely, we would be able to reproduce Bach and Beethoven with the mass amount of mental effort we already have available. Remember, artificial intelligence works like normal intelligence. If it isn't possible using normal intelligence, how is it going to be possible using artificial intelligence? Artificial intelligence might be able to speed it up, but what makes art meaningful is that it *relates*, and that is not a "trainable" thing. And even

      • Re:I wonder (Score:4, Insightful)

        by kheldan ( 1460303 ) on Wednesday June 01, 2016 @07:34PM (#52229439) Journal

        Remember, artificial intelligence works like normal intelligence but only in fantasy and science fiction novels, we don't have real 'AI' in the real world.

        Fixed that for you.

        We don't even really have a clue how our own brains work yet, let alone being able to even begin to emulate it with a machine. All we have right now are cheesy imitations that fall way, way short of the mark.

        Don't anyone sit there and try to convince me that some cheesy algorithm is going to totally emulate human master composers because that's total and complete bullshit. Until we have fully human-level artificial intelligence, completely self-aware, with a full complement of human-level emotion and imagination, there won't be any machine-generated art of music that is equivalent to new works by human artists or human composers. Period.

        • by Lotana ( 842533 )

          I disagree with you that music composition needs strong AI.

          Music composition is a narrow area with relatively well-defined inputs and outputs. Weak AI [wikipedia.org] thrives under such small subsets of real world. Also in terms of music, there are so many varied opinions on what level of quality is acceptable or indeed sublime. While the track in the article obviously does not meet your standard, I am sure there is at least someone out there that enjoyed listening to that tune. I can think of several examples of "music" t

          • Music composition is a narrow area with relatively well-defined inputs and outputs

            That's like saying, "English is easy, there are only 26 possible outputs and 26 possible inputs." Not exactly.......

            • Music composition is a narrow area with relatively well-defined inputs and outputs

              That's like saying, "English is easy, there are only 26 possible outputs and 26 possible inputs." Not exactly.......

              Reducto ad absurdum.

              I would not have said music differs as much in possible inputs and outputs as the board game Go. The obvious BIG difference is the lack of objective measurements of "success" to evaluate positions against (as humans can't agree what is good music - many like pop and cant appreciate classical, vice versa, and so on for other styles), and the success also equates to more than the sum of the individual notes (ie how smaller pieces fit against wider song).

        • Don't anyone sit there and try to convince me that some cheesy algorithm is going to totally emulate human master composers because that's total and complete bullshit.

          Well, either the composer's brains work by magic or they can be emulated by some kind of machine.

          Until we have fully human-level artificial intelligence, completely self-aware, with a full complement of human-level emotion and imagination, there won't be any machine-generated art of music that is equivalent to new works by human artists or hu

          • Well, either the composer's brains work by magic or they can be emulated by some kind of machine.

            NO. You're missing the fundamental point: We do not understand how the human brain does what it does yet, not even close, and without that fundamental understanding you CANNOT build a machine or write mere software that duplicates it's fuinctionality. They write clever mimicks that only go so far but fall way, WAY short of the mark. There is a hard limit to this until we fully, completely understand how our own brains work. Stop buying into all the hype.

            • You're missing the fundamental point: We do not understand how the human brain does what it does yet, not even close, and without that fundamental understanding you CANNOT build a machine or write mere software that duplicates it's fuinctionality.

              And yet we have made machines which play chess.

              They write clever mimicks that only go so far but fall way, WAY short of the mark.

              What mark would that be? Human-level ability? Our chess machines are already at the point of beating world champions. 10x human abili

        • Have you see the "Continuator" program perform? Or the "Painting Fool" perform? Both were covered here on slashdot a couple of years ago.

      • No, you can't until you have a proper music theory.

        What is wrong with the music theory we have? What parts are missing?

    • Check out David Cope's work. He deconstructed Bach's music into an underlying grammatical theory, generalised it to other composers, then built EMI, a program to analyze music for stylistic patterns, then to use his musical grammar to create a full composition in that style. Later he wrote Emily Howell, which can do grammar-based compositions in its own synthesised style. You might be surprised how good the pieces are.

      Google is taking a very different approach, and time will tell how successful it will be.

      • Check out David Cope's work. He deconstructed Bach's music into an underlying grammatical theory, generalised it to other composers, then built EMI, a program to analyze music for stylistic patterns, then to use his musical grammar to create a full composition in that style. Later he wrote Emily Howell, which can do grammar-based compositions in its own synthesised style. You might be surprised how good the pieces are.

        Almost every sentence here is misleading or wrong. Cope didn't build an "underlying grammatical theory" from Bach or anyone else. He basically built a program that pulled out patterns from a specific repertoire (Bach chorales, Mozart sonatas, Chopin Mazurkas, etc.). Then he would have his program "generate" hundreds of "new compositions" that sound like rearranged bits of old works. Then he'd select the best few out of those hundreds of nonsense compositions and showcase them to audiences as new compute

        • Do you have any sources I can follow up? I'm just going on the various popular reports of his work from the last few years, most of which more or less say what I said. I know there has been a pretty wide range of reactions from musicians though; certainly some disliked his work, even getting angry.

          I'm no musical expert myself though, so I'm interested in learning more specific details and criticisms.

          • Posts like this give me hope for slashdot (and the future of humanity in general).

            OP posted what he had read
            A post then comes in with informative (or seemingly from layman's point) post discrediting / correcting OP
            OP asks for his sources so that he may be better informed.

            No name calling, no defensiveness, just open dialog to share knowledge.

            Cue next thread where people call each other shills or sjw or other in-vogue petty insults

    • If we could use deep learning to let computers learn styles and patterns, eventually incorporating them into new music. It could usher in a new era where every film is composed by John Williams, or Mahler's tenth is finished, or there's a new Bach and Beethoven being made every day. Of course, on the arguably darker side, pop music could become entirely computer designed, although considering the quality it would actually sound better if done by computer.

      http://mp3songsringtone.in/ [mp3songsringtone.in]

  • Hey, "deep learning" is all about pattern-matching, and what is music except a bunch of patterns! Let's have machines repeat back the patterns we use! Look -- it's artificial intelligence!

    *sigh*

    Yes, music IS just a bunch of patterns: patterns that evoke very specific emotions. You can't just string 'em together like popcorn, they go together for a reason. Listening to this is just painful, like watching a person try to walk on a broken leg with the bone still sticking out.

    • by jlechem ( 613317 )

      I think it's not just this but anything created by computers is now called AI. I read an article in Wired that had me throwing the magazine out the window about how close we are to true machine learning.

      We're not even close, as a software engineer I fucking hate things called AI. True AI is so far from us it's redonkulous. Yeah it triggers me whenever I see LOOK AI MADE THIS!!

      • It's the medias' fault. They've taken the term 'artificial intelligence' and turned it into a meaningless buzzword. We have no clue how our own minds work yet, therefore trying to create machines that are equivalent is absurd. All they've produced so far are cheesy algorithms that fall way short of the mark.
        • I've seen calculators referred to as AI...obviously, 1+1=2 through the intelligence of the machine. Give it ten years math will be the new magic and the dumbing down of the culture will be complete. I've even seen the work I've done called 'artificial intelligence research' while at the time we called it Runge–Kutta, so I now tell people that I'm an AI researcher.
          • You must be rolling your eyes at least as hard as I am at these people who think we're going to have so-called 'AI' creating masterpiece works of art or composing classic symphonies like human masters, and that in the next few years so-called 'self-driving cars' will be 100% safe, need no manual controls for a human, and you'll be able to put your kids, alone, into one of them and send them off on a trip in it.
      • I was really agreeing with you.

        And then you had to use the phrase "triggers me" ...sigh.

        There are a (very) few legitimate uses of "triggers" - sexual assault, PTSD, ... - truly traumatic and objectively brain altering events. The co-opting of this term for things like the above should be incredibly insulting. You are looking to hijack this word to seemingly give your minor annoyance an artificial legitimacy.

        A far more appropriate expression would've been "this really gets to me" / "this pisses me off" / "th

    • It lacks emotion. Music is all about emotion. Without it all you have is a series of meaningless notes.

      That "song" was indeed painful to listen to. It didn't say anything to me. I didn't feel anything from it. It was nonsense.
      • It lacks emotion. Music is all about emotion. Without it all you have is a series of meaningless notes.

        That "song" was indeed painful to listen to. It didn't say anything to me. I didn't feel anything from it. It was nonsense.

        - Review of every boy/girl band or american idol type album or indeed the majority of the pop music industry.

        Unfortunately 9/10 people want mindless garbage that they can bob their head to and insert their own meaning over the top "Hey I was once in a relationship that broke up! This song is about me!"

        • The tracks go like this:

          Track 1. I am in love with this person
          Track 2. I am conveying my love for this person
          Track 3. We are in love
          Track 4. You are hurting me
          Track 5. I am breaking-up with you
          Track 6. I have moved on

  • http://thenextweb.com/google/2... [thenextweb.com]

    Jesus, that's a bloody awful animated gif to have on a "news" article.

    It's also a bloody awful headline.

    In fact, it looks like a bloody awful website all round.

    As for the tune, it's not bad, but it's a little too unpredictable to be catchy.

    • You tell us. Frankly, I’m reminded of walking into a Best Buy and seeing some kids discover electronic keyboards and their back track buttons, or an old school Nokia ringtone, but it’s pretty impressive for a machine. Hell, it’s better than what I can do.

      If you can't do better than that after spending millions... I'm speechless.

      Band in a Box (remember that) was better.

  • Watch it crank out 57 variations of the first few bars of Under Pressure and pat itself on the back for its ingenuity...

  • by jasnw ( 1913892 )
    My comment is not on the music (muzak, more like) but on that quoted bit from the team at the end of the summary. If everything they write looks like this, no wonder they think holding on to a coherent thought across an entire paragraph is a challenge.
  • by YesIAmAScript ( 886271 ) on Wednesday June 01, 2016 @07:04PM (#52229323)

    If the monkey selfie isn't copyrighted (in the US at least), is this? Are creative works by computers copyrighted?

    • by wjcofkc ( 964165 )
      Not at the moment. However, there may come a point where AI insists their works are copyrighted. We are teaching them to be human after all.
  • First song? (Score:4, Insightful)

    by PCM2 ( 4486 ) on Wednesday June 01, 2016 @07:35PM (#52229443) Homepage

    Is it its first attempt, or only the first attempt that can reasonably be classified as a tune?

  • this is trash (Score:4, Informative)

    by phantomfive ( 622387 ) on Wednesday June 01, 2016 @07:51PM (#52229517) Journal
    It's a crappy melody on top of a repetitious beat. Open the article and listen to it. Compared to the algorithmic music we've had in the past, this is a regression. I knew an undergrad in the 90s who was making computer generated motets better than this.

    And that's not even addressing the emotional aspects that are communicated through music. Totally banal (seriously, ,if you disagree with me, at least listen to the 'song' before explaining why you disagree).
    • It's a crappy melody on top of a repetitious beat.

      So...Trance?

    • by mccalli ( 323026 )
      It feels like the building blocks you'd use to hang a piece off, not the piece itself. You can easily imagine the first part, which is medium-level catchy, being used as the intro to all the complications, key changes, rhythmic additions, chordal backing etc. that would be needed to flesh this out into a real piece. Interestingly the bit where the algo tried to do a shift came off the weakest. It seems it new how to do broken arpeggios and a very basic chord structure, but then when it needed to shift into
      • by mccalli ( 323026 )
        " It seems it new" - I hang my head in shame.
      • It feels like the building blocks you'd use to hang a piece off, not the piece itself. You can easily imagine the first part, which is medium-level catchy, being used as the intro to all the complications, key changes, rhythmic additions, chordal backing etc.

        Indeed, if your name were Bach, you could make a piece out of even the most horrendous melody.

        • by mccalli ( 323026 )
          True - but consider this one: Vangelis - To The Unknown Man [youtube.com]. Perhaps it's not your style of music, but it starts off with a very simple and repeatedly stated melody. That melody is then expanded on throughout the rest of the piece. There are many more examples, but that one came to mind quickly to me for some reason.

          That's what I meant by lacking - seems like they have the makings of a reasonable anthem bit by following a very simple skipped-note arpeggio rule, and then none of the elaboration required t
          • True - but consider this one: Vangelis - To The Unknown Man [youtube.com].

            True the melody there is simplistic, but here's more emotional communication in 3 seconds than in the entire Google composition.

    • Maybe the song in question is good; maybe it isn't - that's up to the listener.

      You, human, have your own opinions on what you call music; be it inspirational, rock, country, new age, or whatever. Your mind was 'trained' with years of what *you* find interesting or wonderful. You've listened to samples to derive your opinion on what you would create - if you created something. It might be nice, or awesome; It might be utter crap, too.

      AI today requires "training" of that AI. It needs to be fed many, many samp

      • Maybe the song in question is good; maybe it isn't

        Oh really?

        But, yes, it's going to be a while before AI starts composing top 10 hits under any genre.

        So you agree, it is bad then. Very consistent words you have there.

  • Since it was generated by an AI, who owns the copyright on this awful piece?

    There were better song generators on the Commodore 64. Wow that was bad.

  • WTF is John Tesh supposed to do now?
  • Looking at this post, I remembered this: http://www.wired.com/2008/02/d... [wired.com]:

    "Here were produced rubbishy newspapers containing almost nothing except sport, crime and astrology, sensational five-cent novelettes, films oozing with sex, and sentimental songs which were composed entirely by mechanical means on a special kind of kaleidoscope known as a versificator."

    Songwriters and musicians are really annoying, in that one has to pay them, they have to live somewhere etc., let's automate! I'm not cynical enou

  • Not like that has never been done before, or better. Mind the date: http://www.nytimes.com/1997/11/11/science/undiscovered-bach-no-a-computer-wrote-it.html?pagewanted=all [nytimes.com]

    How do the two compare? I know this Google attempt will not qualify as 'composed by Bach', so is there something special in the way the Google AI came to this awful sequence of notes? If the Google folk except it to do better, why did they not wait a few learn-iterations and publish that result?
  • Their latest album _The_Astonishing_ tells a story of a future time when digital music is generated by "noise machines" and mankind had forgotten what real creative music was. I know this is first attempt for AI but I (one with pretty much zero creativity) could come up with something better than that!

"The way of the world is to praise dead saints and prosecute live ones." -- Nathaniel Howe

Working...