@@johnprice3341 I would love to see you actually try out and see how it goes. Cause that's probably not gonna happen, would be similar but they would be a lot different too
BP Thành Vinh I didn’t understand “it would love to see you...” did you mean “I?” And the less data you give it, the more repetitive and less musical it would be because it guesses bad patterns from the lack of a large data set.
@@johnprice3341 yea, It was "I..." well, it depends on the symphony itself. Sometimes the third/fourth movement has no correspondence with the first two/three. Sometimes, the second movement act as a bridge and third is a transendence movement, there's too much variety, I might have heard it doesn't mean it's common, it is classical music afterall
Smooth brain: resting your violin on your knee with your bow pointing upwards Wrinkly brain: tucking the violin under your arm with the bow somewhere down low Galaxy brain: putting the violin and bow on the floor under your chair, eat a snack while chatting to your desk partner, throwing the wrapper at the French horns, and missing your entry
@@TAP7a i know a guy who'll literally hang his violin on the back of his chair during tacet or breaks... it's the most risky shit i've ever witnessed and somehow in my three years of knowing him i've never seen it get knocked off. i reckon gravity can be explained due to the mass of his balls.
An AI scientist here. I haven't read the article that you were talking about in the beginning, but yes, as it said, the AI (or rather, "Machine Learning") algorithms can process the available data, find patterns in it, and then use that to either predict what comes next, or perform a prognosis of what new data might look like, or somehow classify the new data samples. So, as an extremely simplified example, if you give it a sequence "1, 2, 3, 4, 5...", then it will learn the pattern that there's a sequence of numbers, and each new number is bigger than the previous by 1, and then will give a prediction: "...6, 7, 8, 9, etc". And if you give it something like "1, 3, 6, 10, 15, 21, ...", then it might learn that it's again, a sequence of numbers, next one is always bigger by an increasing number, and it will continue the sequence with "...28, 36, 45, etc." The sequences and the patterns in them can be much more complex, of course. Incidentally, more sophisticated Machine Learning algorithms can do this with things like stock market prices, climate/temperature data, strings of text (maybe you've heard of AI algorithms "writing" books, movie scripts or fanfics?), images (e.g.., generating human faces or entire photographs, see www.thispersondoesnotexist.com, or, given a picture of a dog, say that "it's probably a dog"), and yes, with music data. Because sheet music can be easily encoded with a sequence of numbers or symbols, and then the algorithms can analyse it and give predictions. My guess, again, without reading the article, is that the algorithm only generated the base of the musical piece, and then this guy just tweaked it, cleaned it, smoothed it out and fancied it up a bit, to sound prettier and more human-like. And yes, it can't quite get the "human element". The inspiration, the "soul", so to speak. So if you give the algorithm a sequence like "1, 2, 3, 69, 420...", the algorithm will just go "u wot mate?". You should've posted the link in the description to the original article. EDIT: yeah, I removed the bit about the degree. That wasn't a good place to mention it, I agree. I don't know why that rhymed.
My 91 year old neighbour, who is like my grandmother said that you guys are very attractive young men with very soothing voices. And that you are very funny too. She really like your videos and always want to watch them with me. So much love from Switzerland, the other side of the globe😂❤️
AI robot voice: "And for this next movement, we shall delve into and explore the deepest emotions sadness, anger, and pain. It is titled ctrl+alt+del."
I think people are forgetting that the only thing the AI did was write a couple of melodies that resembled Shubert's style, that's it, no harmony, no orchestration, nothing else. The movements were written by a composer imitating Schubert and using the AI's generated melodies. That being said, I would love to listen to an actual completely AI generated movement. Orchestrated and all... Needless to say they'd need an entirely different AI learning setup. Anyway, remember people, the only thing the AI did was write the melodies.
Yep, the smartphone site clearly say that it just wrote the melodies: consumer.huawei.com/uk/campaign/unfinishedsymphony/ I’ve heard when AI is used for orchestration and it usually sounds off. It just seems like a useful tool for composers to get inspiration… for now.
Eddy: It would be interesting to see an AI concert Me: *VOCALOID* edit. I'm not saying that the Vocaloids can make their own songs like the AI twoset mentioned in this video. But they do indeed put up holograms for concerts. I also agree that they shouldn't explore vocaloid by themselves but someone else that is familiar with vocaloid should recommend to them. Sorry if I caused any misunderstandings ;A;
I wouldn’t necessarily call Vocaloid AI software tho as people generally do all the music and tuning of the voices themselves. Honestly it’s more work than just singing in making a voicebank, making an original song with a voicebank, or making a MMD. Every single detail down to each syllable needs to be fine tuned and it’s just as much work as normal music producing or maybe even more. As for MMDs, if you don’t have experience the dancing is just gonna come out looking like those crappy ragdoll games. But I do agree that since we already have hologram concerts, concerts with AI generated music using voicebanks and production software don’t seem too far fetched.
As an AI researcher (specifically from the Natural Language Processing area) the reactions from you guys are feeding my heart hahaha But, let's take it easy as mentioned by my colleagues already AI in general are pattern finders most of the time, and we may have to wait a little longer for really creative AIs to appear. Although I can totally see them coming with Reinforcement Learning and GANs. Well these are my 2cents ☺️ Your work is awesome, please continue. Btw I've just bought a Violin and will go for my practices as well.
I wonder what an AI would do if it composed a symphony, used a survey to analyze whether it counted as a "good" piece, and got a 50/50 like/dislike response. Would it conclude that, based on the nature of statistics and the subjectivity of art, that this was technically the most ideal piece it could make?
@@johannsebastienbach Is it really? Because music is at the end of the day and endless repetition of the same patterns. Now days coming up with a new melody that also sounds good is rather difficult. And what your brain process as "creative" in music is exactly those patterns (that why they are used constantly). I have to find a composer that do not inspire himself to other previous composers, and inspiration is the pure definition of what the IA does, look at patterns and try to replicate them. The actual weakness of the AI is that it cannot define what it sounds good. You have either to train it by telling it "this is beautiful learn this" or by giving it a set of parameters to look at and saying "this are the parameter for the perfect composition". In any case, we have the entire western culture complaining about how role modeling works around everything. We claim that the concept of beauty is defined by the TV and News papers, meaning that as human we choose to consider what looks "good" by looking at what other people tell us to consider good. So can you please state where is the difference between AI learning from humans and humans learning from humans?
@@GabrieleNunnari I don’t think it’s that simple. I find it fascinating that even with the limited amount of notes we haven’t reached the limits. The Batman dark knight theme sounds completely unique and hundreds and thousands of songs have been made before it. Also music evolves. What we consider today as classic was the metal of the medieval times, don’t forget that. Styles emerge and create new atmospheres. Music is 100% a transformation of information of emotions. The notes are the tools, which is why we haven’t run out of melodies imo
@@codecampbase1525 after stumbling upon "music similarities" compilations on youtube where random anime game music, classical music, indie rock and who knows else have pretty much the same motif, and then starting to notice it outside of youtube videos myself, i'm almost sure that we already did and all of the melodies are either same, with slight variations or a mix of several others with the saving grace being the fact that melody is not the only part of the musical piece.
As an AI practitioner, I can try to explain what happened. The most likely algorithms that generate the piece is called LSTM (Long-short Term Memory), or it also could be Attention Model. Both algorithms learn the pattern based on all the "training data" that they are provided, namely all Schubert symphonies, or could be all symphonies that are ever written in the similar period. The learned algorithm will try to predict the "future data", based on the "input data", which is Mov 1st and 2nd, and predict the next note, and next note, all the way to the end when it predicts the "end note". Here are some explanations to your questions: 1. How does it remember the proper sequence of instruments coming in? Because it learns the PATTERN of all symphonies, and according to some rules of symphony composing (or "code" as you guys mentioned in the video), they all happen in certain order. Once AI decides the proper timestamp that each instrument comes in, it will predict the starting notes and rhyme for each instrument, and lay them on top of each other, one by one. 2. Why is it repetitive? This also happens to AI speech generation, which uses very similar technique with AI music composition. The model is predicting the "next note" in sequential order. For example, when the current predicted note is C, the next most likely note will be predicted at G. Then after several rounds of iteration, when some future note is predicted at C again, the one after future C will be predicted at G again. In other words, if the music never ends, you will see endless rounds of same passages occurring, just like seeing repeating decimals 2.123412341234...... 3. Why does it sound more like Tchaikovsky in certain parts? Easy to think about, that AI tries to learn the pattern based on all "training data", and the algorithm developer must have provided Tchaikovsky's pieces as part of training data. Throughout the AI music generation process, AI is trying to predict the next passages by giving a mixture of all composers that it learns from, such as 40% Tchaikovsky + 50% Schubert + 5% Dvorak + 5% Beethoven. 4. Why does 3rd movement sound like 1st movement? Because it learns the pattern of all musics. Not sure if it's true in Symphonies, I would say it's very likely the case in piano concertos, that 3rd movement paraphrases the 1st movement quite often, such as Rachmaninoff piano concerto 2 & 3.
@Tianqi Yang This is by far a very good explanation from a technical perspective. I was hoping someone would explain the things here. As a part of AI community, I would want to shed some light on this too. To those who don't follow AI actively and to those who are afraid of this. Here are some facts? Take it with a pinch of salt because these are something the community is saying(not my personal opinion) 1. AI -- > AI is, essentially to date, a pattern-matching machine. It is very good at finding a very very complex pattern and has already defeated humans on this topic. But it is a machine/algorithm at the end of the day. AI works primarily from memory(which it has already learned) 2. AGI --> AGI is supposed to be a system that has the human-level intelligence. Intelligence is not defined yet. AI model was inspired by human brain neurons but doesn't mean it is fully mimicking the human brain. There are lots of things besides a neuron that controls the human system(cell for an example). 3. If somebody says AGI is already there they are lying (some business strategy). Nevertheless, AGI will come in the long run maybe. 4. somebody could argue saying that historically any new invention creates new jobs and kills the older ones (Telephone invention killed a lot of jobs but created a lot of jobs too) 5. Here is a good argument, In a world where if AI automates all the boring jobs then people will have time to do creative thinking. My understanding is at least in my life the skills like ART will have much more value than the ones which can be automated. Nobody cares(at least from the point of music intellectuality) about AI generating art(a lot of deep learning models can easily generate paintings). Also, people wouldn't mind if bots like google duplex automate a lot of call center jobs.
I've been closely following the Google Coconet Ai and Magenta that was presented last year in a doodle... it is able to finish or create simple melodies with AI. It dissected Bach's compositions to compose harmonizations in the style of Bach. They trained it by omitting certain notes from original compositions and created a rating system for the AI rewarding solutions that closely resemble the original compositions. Through the accumulation process they gradually increased the amount of missing notes. Harmonization was the goal but with a few more rules and exceptions added they made it compose entire symphonies from scratch. Even the data evaluation process for rating predictions can be automated because google, through 'google music' and 'youtube music' is able to create popularity charts based on airtime numbers, gender, country, special interest, etc....
I bet their project was actually a disaster, and there was a bunch of human intervention with the AI’s choices, even editing the music afterwards. It’s easy to lie and say something is AI; Sofia’s public conversations were rehearsed beforehand.
Plot twist: Ling Ling is an AI from the future. That’s why he/she can practice 40 hours a day and be perfect at everything. *X-Files theme intensifies*
11:33 Miguel's comment is spot-on. It's not like they're claiming they created a Schubert AI that can write an infinite number of symphonies in the style of Schubert. It's not much different from any composer attempting to finish the symphony based on their understanding of Schubert. It's not perfect but it shows that technology can do and how it can help compose great music (albeit with a human composer cleaning up the finished product)
@@RobinsMusic Just like how a human can no longer best a computer at 'Chess' or 'Go' or 'Tic-Tac-Toe', or any game, soon the best human composer will not be able to compose at the level of 'A.I. composers". Sorry Schubert.
FYI the piece was actually finished with the help of a human composer, here’s what they said on their website: “This wouldn’t have been possible without pairing the technological innovation of Huawei’s AI with human expertise, so Emmy award-winning composer Lucas Cantor was brought onboard to arrange an orchestral score based on the melody that the Mate 20 Pro smartphone composed to compete the symphony and perform it.” Don’t get me wrong, the Ai is still really impressive though.
@@rogerio4039 Exactly, remember that "interesting" prodigy? She only used four notes as a motif to create a complete piece. Does that count as "writing melodies" too?
@@nianyiwang actually they can, i've read an article before that says movie script writers uses AI to analyze past good selling movies and provide relevant keywords from them as a baseline to write a new movie. i'll see if i can find that article again
ACTUALLY AI DID NOT COMPOSE THIS ON IT'S OWN . The AI actually only composed a melody that was latter turned into a full score by a human composer. Huawei hired composer Lucas Cantor to arrange an orchestral score based on a "melody" the AI wrote . The following is from Huawei own press release. "This wouldn’t have been possible without pairing the technological innovation of Huawei’s AI with human expertise, so Emmy award-winning composer Lucas Cantor was brought onboard to arrange an orchestral score based on the melody that the Mate 20 Pro smartphone composed to compete the symphony and perform it live."
1:22 Lucas Cantor composed the movements... with the HELP of AI. That's really different from AI composing a symphony, and probably accounts for the Pixar/Disney vibe.
Schubert would be turning in his grave to hear that attempt at completion. What is worse, is that a really successful completion has already been made many years ago by Brian Newbould, who knew everything there is to be known about the music of Schubert. Lucas Cantor is a bloody fool as is his AI. If you would rely on Lucas Cantor, you might as well eat your head.
@@CSRgamer I think the algorithm didn't do much, it probably just generated a set of notes as the article in the video said, then the compositor just started his work from there. Whenever I work with AI it always feels way too mechanical, and the learning process make it feel like a random generator (random until you get something right), even if you give them huge datasets these algorithms have an insignificant "brainpower" compared to the human's, we eighter need a breakthrough in the design or much more processing power to make, for example, bigger neural networks.
It sounds scarily good but it does say in the article that Lucas Cantor composed it "with the help" of AI. I wonder how much he changed the AI suggestions. Maybe the raw AI material is not THAT good?
Now I want to see David Bruce interview Cantor about how this *actually* worked, especially the orchestration, with commentary on the "how to sound like" aspects by Nahre Sol 😆
This shit right here. Most tech-journalism around AI can be super fallacious, thanks in part to the programmers trying to drum up interest. You need to check how much the AI "wrote a symphony" and how much "a human picked what the AI was to focus on and replicate". Even in the case of a 'true AI' what suggestions it will form are based on a system designed by a human, so it's always tricky to say the AI did anything on its own. Counter-point though, a lot of people in art would argue you could say the same about human composers/artists/writers. The conventions of different genres and mediums are learned and developed from past humans' techniques and teachings, so one could say we're all "programmed" to perceive specific forms of arts in specific ways. Value judgements about an AI's ability to be creative is thus a bit confusing.
I think the AI made the repetitive stuff we can hear in the 3rd mvt. There are ideas made by the AI, but the human person chooses how articulate these little patterns. I don't know if an AI can wrote an entire symphony without too much repetitions. In the case of Schubert, there are not enough datas : the AI can only see little patterns, but not a whole pattern we could find in all Schubert's symphonies. So I think (and hope) human composers are always much better than a good AI.
5:07 what every you say against AI, this brass swell was so unexpected for me and hit me hard! really emotional moment, a lot of expression and art in this single brass hit
What I thought is: it would very fun to give AI an existing finished piece and let it compose the last movement, and then compare the real one and the artificial one. I'm particularly interested in what Sibeluis Violin Concerto's last movement would turn out like
@@a.baciste1733 It is, most machine learning (one of AI branch, so to speak) validate the result generated by the model with real data to see whether the model is good or not good enough. (CMIIW)
This is pretty much a principle in machine learning: after training the machine (AI) on a dataset, you give it some new input it hasn't seen, observe the generated output, and compare against the true expected output to see how the machine performed
I work as an AI scientist and have a PhD in music theory (specializing in machine learning/AI for music analysis). While I haven't seen the code to this particular example, so many of these hyped "AI composers" are really fake clickbait to get attention for a company (Huawei in this case). My guess is that they had a computer go over the MIDI or xml file of Schubert's symphony and perhaps other similar works. You then train a model to give a probabilistic output of what came next. Example: given the notes G, Bb, A, Schubert would then go to G 40% of the time, D 20% of the time, Bb again 5%, etc. I made those probabilities up, but that's the basic idea. Same with rhythm. You can do this with simple Markov chain probabilities or something more complex like an LSTM or Transformer model. But the end result is that you can produce rhythms and melodies that have a similar probability as Schubert and any other music the model was trained on. But you typically have to get the melodies started (seeding) and end them also. However, current AI has a hard time making motifs, large-scale cohesion, and form. I suspect that this "AI" made a few melodic fragments, and the composer: finished the melodies, arranged the form, made harmonies, dynamics, orchestration, developed motifs, etc. Also, humans performed, recorded, mastered, and produced the music. There are of course very interesting and groundbreaking developments in music AI, such as the Magenta project where their Transformer model approach is starting to show signs of learning motifs and phrase structure. Sure, perhaps in 10 years a lot of commercial music might be made with AI, but we as humans get to choose whether we want our music (not the background muzak that is made to sell you stuff) to be a pure human-to-human pursuit, sit back and let computers make all our music, or integrate AI into our musical world as we see fit much as we have for other technologies. I think music will always be a human-first expression simply because now we will pay lots of money to see people perform music when we could have saved the money and had our computers play a recording of that same music.
"I think music will always be a human-first expression simply because now we will pay lots of money to see people perform music" Sharon Apple has entered the chat
@@MrWhangdoodles the commenter owes you nothing. they took the time out of their day to write a professionally-backed explanation; and regardless of whether or not he formatted it like a polished essay, thats pretty badass if u ask me. badass mfs dont need no paragraphs. peace and love, brother.
"It really challenges what it means to make music and what it means to be human." I am working in an AI-related field; this quote is very important and refreshing for me. Thank you.
Nah. The AI is reflecting what Schubert composed according to human generated rules on pattern matching. I'd say, AI shows us more what it means to be human and to make music. Note how quickly Eddy and Brett detected it was like Schubert, but not exactly like Schubert. Even I, an amateur musician, thought the third movement was very much like the first, and essentially a slightly changed copy. A human composer wouldn't do that. There's a story to tell in the music. You might have some phrase repeating or evocative of the earlier movements, but there'd be surprising differences as the story progresses. I think of Beethoven's 9th and his 1st, 2nd, and 3rd movements and how they refer back to each other, but lead inevitably to the fourth.
An AI does need hundreds of hours to be able to do something. I once made one that played minesweeper and it took it 3 days to learn how to play ok. And that is hundreds or thousnads of cycles per hour.
Doesn't that depend on your computational power? I mean, I think that Alpha Zero, one of the best chess engines in the world, trains for around four hours and defeats Grand Masters.
@@mariopalenciagutierrez4318 What did you do to overcomplicate an AI so badly that it can't play minesweeper properly for days? It should be easy to make one that plays perfectly without any machine learning involved at all, meaning no learning time. Even machine learning based a standard computer should learn minesweeper in seconds, not days. Don't base technical limitations on your own bad code. "Hundreds or thousands of cycles per hour"? It should be per second if you implemented it correctly.
@@gronbuske i did it through machine learning. Yes, i know it was hundreds of thousands per hour. Took a couple million games for it to learn by itself.
@@mariopalenciagutierrez4318 I do hear what you say, but even a few million games should take seconds or minutes, not hours and days. Post your github or wherever you put your code and I can probably fix the issue for you. A minesweeper round done by a computer should take a few milliseconds, you're saying hundreds per hour meaning every round took seconds instead. That only shows you have some serious bottleneck in your program, not that "AI does need hundreds of hours to do something"; that is absolutely false and misguiding.
I think one or two years ago Google created a Doodle which was something like a tribute to Bach. We can input notes/ our own melody to that doodle, and the AI system would compose, within a few seconds, a 4-part polyphonic melody.
@@heyytheree Why? Humans want to listen Beethoven because its Beethoven. Will you replace someone's entire musics because someone's a little better musician? AI won't be interested in replacing Beethoven. More likely to be interested in creating uncharted novel classical music industry market (by its creators).
Aw hell yeah, Schubert appreciation! Literally the first exposure to classical music that I remember was the theme of the Canadian cartoon Little Bear (I obsessively watched the reruns on UK kids TV), which was an arrangement of the Allegro vivace from Schubert’s Sonatina in D Major, and I absolutely fell in love with as a toddler/little kid 🥺. I definitely wouldn’t have picked up a violin several years later without it, so I owe an awful lot to one extremely talented Austrian dude lmao, thanks Schubert 😌
watching this made me realize... it would totally be amazing to have a twoset podcast where they just discus and dissect classical music and maybe even have classical music friends / experts on the show
Be the first OR be the best. If you're into coding, you must know making all the money oftentimes just comes down to taking a trending app, improving on it and releasing that. If you want to be a bonafide inventor/creator that feels like a whole other mindset, and keep in mind the "13 failures for every 1 success" parable. Don't stop.
Actually this would be a very interesting way to test it: feed it most of a composer’s body of work, give it only a taste of one completed piece, and then let it try to finish that piece. I wonder how different it’d end up compared to the real thing.
It was very nice to see them acknowledge both sides/perspectives. I like it when people say "I am conflicted" because most of the time it means they've tried looking at it from multiple perspectives. Saying that also means that rather than committing to on side too soon, they would rather gather more information to make a better decision. Edit: Personal thoughts. The idea of AI composing music is pretty cool and intimidating! Music is still a human thing though.
That tragic and soaring climax after the buildup at 8:00 , literally right when Brett said, "Fly!" was really quite spectacular and unexpected. I wonder if anyone can recognize a similar passage from other classical works. This has actually inspired me to begin a listen through of Schubert's and Beethoven's complete symphonies and more, outside of my Romantic piano comfort zone.
The first four bars of eine kleine nachtmusik is nothing but Tonic and Dominant chords. No tensions, no passing notes, just simple arpeggio. If Mozart didn't exist, it would never happened. The simplest example of striking genius. ua-cam.com/video/MeaQ595tzxQ/v-deo.html Beethoven's 5th symphony also has a similar four bar opening. So So So Me──。Fa Fa Fa Re──。Only four notes. Can AI do that? I doubt it. A machine doesn't recognize potential beauty mined deep inside the simplicity. To understand that, a machine has to analyze human being itself. In another words, a machine has to be capable of enjoying music to create one...lol A machine also have to be a genius to produce a good one. Letting mother nature do the job is far easier way.
@@ttwiligh7 It's not too far though. AI is a learning software, it can learn to do so. Like any master composer, they first learned theory in order to create beautiful symphonies. They were taught, trained and became masters of their craft. No toddler can play Beethoven without being taught first. This sounds very elitist, in my opinion.
@@kotarodesu_23 You can easily end up with garbage just by following rules, unless you have a talent....something AI lacks most. If it was that easy as you say, everyone can become a great composer by sheer effort. Let's say you study Bach. Will you be able to write something as deep as Chaconne? Not a chance. How do you expect a machine to express those emotions when it has none? A machine is less than untalented people in that sense, to recognize beauty. How do you judge if a symphony is beautiful or not? Is there any methodical way? Only if we know how to become genius, and can break down talents into zeros and ones, yes, it might be possible. And this symphony, is mostly human's work....Not like AI came up with orchestrations. They say only melodies. Not even close to composing a complete piano piece. If AI could make something like Chopin's piano piece entirely by its own, then I will believe.
@@ttwiligh7 AI is still in its infancy. And like all infants, aren't capable of much. I'm not arguing on the side of AI nor on the side of this elite perspective. You choose to believe in its fallacies and shortcomings whereas I believe in the potential, and that says a lot about you than you think.
@@kotarodesu_23 It says a lot about how you see music. Even most talented musicians can't become Bach. Then why you think that is possible for a machine? Yeah, it's like a infant, but not every infant has a talent. My question is, how do you program a talent? And You don't seems to have an answer. Just believing? You think you can learn a talent? That's why we circulate the same argument. If somebody can program AI to be like Bach, he could be Bach himself. There are just so many things you can't just learn methodically, by zeros and ones. Still in a dark, still a mystery. As I say before, if you can break down creativity into computer language, it might be possible. But that require astronomical bytes, and when it's done, AI must understand human emotions to sense beauty and judge its value like we do. Not going to happen in near future. That's what I'm saying.
I think it would be *more* interesting to have AI finish a work that *is* finished by only giving it a partial score. Then we can see/ hear how close it comes to what the composer *actually* did.
You guys say that LingLing will come and save us (from AI)... But what if LingLing IS some sort of AI...? Would we need to be saved from LingLing? If that were to happen... Who then would be saving us now? Are we even safe anymore? Btw, lovely video. Refreshingly different. Hope to see some more
Ooh, whoever did the recording of that orchestra actually recorded in stereo with decent separation and did little blending when mastering. Very nice. It gives the recording a lot more spatiality to the sound stage and enhances instrumental imaging on that stage. I really appreciate that.
Honestly, there is something about every person's art- writing, paintings, music- being unique and the fact that no two people can create the exact same piece of original art from scratch that has always been strangely comforting to me. But with the years there have been many a time when I came across a line of writing or a piece of music and was shocked by the similarity it had with some of my works and some of others'. Sometimes whole lines of writing or phrases of my original music that I thought were truly "mine" were the same as works I'd never seen before composing or writing. This has only lead me to believe even firmly in the philosophy of art for the sake of art. Because, thankfully, there is no winner in art. Sorry for the unrelated rant aahh.
The thing a lot of people don’t know is it’s going to imitate explicitly what it hears. Brilliant, shattering ideas that take us in new directions can’t be improvised by a system which produces the same pattern over and over. That and I want to watch people struggle through the art of composing and performing to understand their life through their art. It’s like kind of the point. Edit: also tho it’s damn impressive.
I'd love to play the whole 'unfinished' symphony with the AI movements included. That big melody that is high up in the strings sounds like it would be so much fun to play.
No, but it would be funny if you knew that the AI might've thought that coughs were an "essential" part of the movement. The generated movement weren't probably a sequence of sheet music but rather an actual audio sequence.
I love how they listen to Schubert's music. Also, I think the AI composition is impressive from a technological perspective BUT definitely not Schubert.
Honestly that might be a matter of pumping more resources into the training of the AI: beefier hardware than a smartphone + longer training time with more human experts involved in guiding the AI would likely yield more impressive results
@@LordOfFlies True, which is why it's not actually a useful approach to try to directly define subjective qualitative characteristics. Instead a more plausible appraoch may be to let the AI learn through numerous examples and through constant human feedback. Over time it might learn certain examples of patterns that invoke the intended emotions within the average listener.
"Give it 20 years" AI: "Hold my beer" People find it almost impossible to grasp the exponential rate of AI growth. Software of this kind will be 100 times better in 18 months.
Not to be rude but watching 2-Minute Papers doesn't make you an expert on machine learning. There's always a ceiling with a given set of technologies and eyeballing a problem like composing a symphony from afar doesn't tell you where it is.
Something important to note ... the AI did not orchestrate the music. The orchestration was completed by Lucas Cantor ... which is probably why the final two "movements" sound like film soundtrack. The AI generated the "melodies" for the two movements which Cantor drew from (it would be "interesting" to hear those melodies). This is nothing but a stunt from Huawei.
If you look at it as AI assisted composition, I'd say it's a bit more than just a stunt. We're seeing a glimpse of the future whether we wish to acknowledge it or not.
4 роки тому+381
It is VERY close to a fake ! It reads: "The Mate 20 Pro smartphone listened to the first two movements of Schubert’s Symphony No. 8, analysed the key musical elements that make it so incredible, then generated the melody for the missing third and fourth movement from its analysis." And then: "This wouldn’t have been possible without pairing the technological innovation of Huawei’s AI with human expertise, so Emmy award-winning composer Lucas Cantor was brought onboard to arrange an orchestral score based on the melody that the Mate 20 Pro smartphone composed to compete the symphony and perform it live." Summary : a computer has produced (we dont' know how many) melodies ; and a human expert has done ALL THE HARD PART ! Plus : where is the scherzo ? This 3rd movement is clearly a 1st movement !
I agree, it was the human element that made it what it is and not purely out of the model. But as a ML and CNN practitioner myself, it might be possible, feeding the model with enough data can possibly result with a piece written by an AI itself.
4 роки тому+5
Zeamoose Akino - indeed. The brain is a NN so (if you exclude transcendent properties such as souls or spirits) why not? But how many parameters? 10^18 ? Because your NN will need to have enough memory (forget LSTM ...) to maintain the consistency of the piece (provided it has learned the Sonata form, rondo, scherzo-trio etc.). Maybe you can tweak the topology to that end. But still, I am not sure that’s a smart way of spending terawatts.
@ Musical composition is a teeny tiny fraction of our brain, and music itself is simple AF. The 25 minutes of part 3 and 4 are expressible in less than 3KB before applying PCA. Lets say that the solution space is
4 роки тому+1
@@Idiomatick - the point is not to contain the information of this symphony, or even be able to reproduce the symphony, but be able to compose a new one that sounds like Schubert. The ML should learn : how Schubert composes melodies ; how he builds a sonata form (after learning the principle of sonata forms) ; how he conducts a development (harmony progression, theme variation, etc.) ; how he treats each instrument in the orchestration, including usage of registers and combinations of several instruments ; etc.
i'm late to this but HOLYYYY this is giving me an existential crisis... I really like how in the ending they briefly brought up the topic of what it means to create music (or just art in general) and what it means to be human and I really wish they could talk more about that in another video. This is mindblowing.
Here is how this works: the more data, the more accurate the model. This test was a bit unfair. Make it digest the entire Schubertian corpus and see what happens.
Honestly, I think AI is very people extremely underestimate AI’s future in music. Music boils down to a format of patterns, and AI is exceptionally well at picking up on patterns… I think the music industry will be heavily effected by AI by 2030.
Music is not pattern, music is based on pattern which is created by human. Based on does not mean strictly follow. Music like Jazz and Flamenco often have variations of the basic theory and evolves overtime, don't play by score, play by feel. AI can only follow, cannot create.
@@wonghow actually, an that is what an AI is best at, it is best at following and creating indirect patterns. Afterall, creating is literally what an AI is made to do. Following a strict pattern is what a traditional programming script is best at.
@@redthunder6183 AI only learn from certain patterns, and rules, does not create. Humans don't need to use patterns or rules, that's why you don't understand music. A player can create based on feel. "El Duende" That's why AI means AI, a robot, can't feel, can't think, can't listen, not suited for everything.
Aphrontic Alchemist what I mean is...it sounds just like the first movement but altered. What the first movement might sound like in a parallel universe. Not literally “upside down”.
I kinda feel this on a visual arts level, to an extent. You know, people using filters and photoshop on their pictures to make it look like an oil painting that they then print out on a canvas and hang on their wall, instead of paying an artist to paint their portrait and thinking they are still getting "the same effect." I mean, on the one hand, I can understand financially it makes more sense, but artists really do get the shaft all the time. And, personally (though I could be heavily biased), I think those "filter paintings" look really tacky nine times out of ten.
Practising a centuries old art, don´t be surprised if you are left behind. Imagine masons or glasscrafters being pissed off because machining makes their craft obselete. As an artist, shouldn´t you be glad of getting more tools to express yourself?
@@avenderiel Of course it hurts people because they mastered and practiced it for most of their life and they'll get replaced by an AI which only trained for hours or days. Sure, they could express themselves more but their former passion to do what they do will be snuffed out. Imagine some kind of Ling ling that'll more often be better, faster, and more popular than you.
I am a Computer Science PhD student and this article is just inflating the whole AI thing. It is still quite far from actually being able to replace a human. But we are trying 😂
I was just waiting for you to say it sounds like film music. :D When I heard the first tone of that AI music, all I thought was "it sounds way more modern than your standard classical piece, and definitely more modern than the passages from Schubert's symphony that you showed us". You gave it a name. It is repetitive... I'd also say there's this air of... sterile cleanliness, if you know what I mean? It sort of lacks the depth and dynamics that I would expect from a classical piece. Mind you, I am by far no expert, but it just sounds like the modern style that has lost some of the classical flamboyance (is that the right word? I dunno...). And it feels like the tempo doesn't change at all (that's probably what you meant by abrupt). So in my opinion, AI still has a long way to go. Thank god, your jobs are saved. :) Interesting part about music being just a code - well, in essence, everything is just code. A bunch of ones and zeroes if you strip it down to the basics, because if you get down to it, our thoughts and biological processes are just a bunch of electrical impulses. So the thing that we call our soul is just code and everything we produce is code as well. But do we really want to think like that? :D
This reminds me of when Vocaloids came out and the voice synthesizers became a thing. A lot of people were angry, claiming that only humans can sing, that they were trying to make "robots" take the singers place, but it really depends on how you see it. In the case of vocaloids, instead of comparing them to singers they should be compared to instruments. By using them, composers were able to make their songs without having to rely on someone else to sing. They could express exactly what they wanted, its not different from expressing yourself through a piano piece. I believe the AI would work like that. It is not there to substitute composers, but as a tool to maybe help them. I hate this mentality that people cant use technology for their creation or else its "cheating". Like when people claimed that digital art is not art compared to a canvas paiting. They are all art creatd with different tools and have their own particular dificulties, theres no such thing as one being easier or superior to the other. The AI could be used to help people, specially younger composers. Not to be just "ok AI make the symphony for me" but to help with inspiration. Don't composers sometimes find inspiration in other pieces of music? It could be like that. People need to stop thinking technology exists to substitute things, technology is supposed to help use, make things work easier and quicker. Like if a digital paint takes less time to make compared to an oil paint why is that a bad thing? That means that digital artist now can make their creation with less time and thus have time to create more.
As an artist, I'm afraid I disagree. Good things take time, work and care. It's not about speed. There is a certain satisfaction that comes with learning how to master and manipulate the medium, be it oils, watercolors or pen and ink. It's the feel of the brush on the canvas, the smell of the turps, the effect of the light on the pigments. There is a more personal quality to a painting on a canvas which digital art can't quite replicate, though style and composition may vary. If I may, the comparison of the vocaloid to the man playing the piano doesn't hold up: human voices, like pianos, are both instruments, and in both cases it is a human being who is manipulating them to produce the music. Twoset have shown two episodes now where they displayed the talents of new composers who submitted their works. They seem to be doing just fine without any assistance from machines. Where technology helps, I'm all for it, but where it seeks to take from us those things which are intrinsic to being human, it defeats its purpose.
@@studiodensmore4979 Well agree to disagree then. I didnt meant to say that digital art is something you can do fast, artists take hours even days to finish a single image. And yes its not a physical brush or paint, but that doesnt devalue their work or make it easier to make than an actual paint. I dont think I expressed myself well in that regard so thats my bad. Thats why I said both digital and non-digital art is still art. I personally dont see digital art replacing traditional art at all. They are different, they require different techniques, both have their artists soul and heart in it and are both beautiful in their own way. And I never said composers are having a hard time at all, I meant that it could be an extra tool for some not all. That was just an example I definitely dont think anyone should or would use the AI, but I also dont think that if someone did, that their work should be undervalued or compared to those that dont. I just dont believe in this old mentality that everything must be classic, old-style, made by hand to have value or to be considered art. Specially because art appeals to people in different ways. I know people who claimed digital artworks are empty, but I myself look to know pieces like Mona Lisa and other famous paiting and I feel absolutely nothing from it. That doesnt mean any of these pieces are bad, they simply appeal differently to each person
While I'll admit that is does take some practise to master digital "brushes" etc, it doesn't take nearly as much skill as actually painting. Digital Does make it easier-- you don't have to think about the drying times of pigments, of glazes, of the water retention of the paper, of having south facing windows, and a myriad other factors. I have seen some beautiful images rendered with digital media. But the same painting done twice, once digitally and once in real life, by the same artist, will have more life to it in the natural- and be more valued by the artist himself for the work it took. As for a composer's works being devalued for using AI, this is a slightly different argument. For digital vs natural paint, the question is about medium, but in either case it is still something used by a human being, who dictates the outcome. For a composer using AI, we are talking about a machine doing the work For you. In such a case as this, yes, I think the work would have less value, as the composer would simply have sat back and let the machine do all the work. (Though this is just speculation, as I don't believe a machine could ever compose; how could something with neither heart nor soul be inspired to make music?) You don't think something made by hand should have more value than something machine-made!? Have you never made anything by hand yourself? And if you did, and offered it to someone, and were told, " oh, I have a machine that can do that," how would you feel? Part of the value of learning to sing, to paint, or compose, is the learning process itself: of the confidence gained in overcoming difficulties, not bypassing them. Of the joy and satisfaction of mastering a skill, and sharing it with others. In short--part of being human.
Satsu Jin yea, digital art has its own problems with the creation process, it takes just as much skill as traditional art with paintbrushes and stuff, I like traditional and digital art but I kinda like digital art more, because to me it just has that kinda life to it that traditional art kinda is, missing I wanna say? But yea, I like digital art better than traditional it takes just as much skill as traditional art to master, art in any form is still art.
Brett: Hey siri, compose me a symphony, please Siri: Okay, I found this on the web for compose me a symphony please Brett: Nah, Siri is back in 1990s Siri: So you dare me to compose a symphony that will ruin your future, Checkmate!
i’m actually playing the unfinished symphony in my orchestra currently and when i was telling my teacher about it she mentioned that she played in a recording of an ai completed version… so i’m pretty sure my teacher is one of the bassoonists in this
AI after releasing its 9th symphony:
A fatal error occurred.
😂
Kiyo Yo LMFAOOOOO
Good one XD
if you can die after 9th symphony humanly you can die after 9th symphony digitally 😂
It's not a bug. Just a feature
It would be INTERESTING to 'feed' this AI two movements of a finished symphony and see how it compares to the original 3rd :)
It would be too similar. 3rd movement needs contrast, so you should feed it other things on purpose
@@johnprice3341 I would love to see you actually try out and see how it goes. Cause that's probably not gonna happen, would be similar but they would be a lot different too
BP Thành Vinh I didn’t understand “it would love to see you...” did you mean “I?” And the less data you give it, the more repetitive and less musical it would be because it guesses bad patterns from the lack of a large data set.
@@johnprice3341 yea, It was "I..." well, it depends on the symphony itself. Sometimes the third/fourth movement has no correspondence with the first two/three. Sometimes, the second movement act as a bridge and third is a transendence movement, there's too much variety, I might have heard it doesn't mean it's common, it is classical music afterall
That's a good idea.
AI releases its 9th Symphony
*Windows XP shutdown sound effect*
Nice 😂😂😂
lmaoo u won
underated lol
This made me laugh out loud.
I like how whenever you're not playing your violin you just kind of... cuddle it
Our instruments are our babies ❤️
Smooth brain: resting your violin on your knee with your bow pointing upwards
Wrinkly brain: tucking the violin under your arm with the bow somewhere down low
Galaxy brain: putting the violin and bow on the floor under your chair, eat a snack while chatting to your desk partner, throwing the wrapper at the French horns, and missing your entry
@CatsKalambur lol yea, unlike that thumbnail from one of Twoset's reacting video on movie Brahms I think?
@@TAP7a i know a guy who'll literally hang his violin on the back of his chair during tacet or breaks... it's the most risky shit i've ever witnessed and somehow in my three years of knowing him i've never seen it get knocked off. i reckon gravity can be explained due to the mass of his balls.
@CatsKalambur F-hole sounds so wrong lol
An AI scientist here.
I haven't read the article that you were talking about in the beginning, but yes, as it said, the AI (or rather, "Machine Learning") algorithms can process the available data, find patterns in it, and then use that to either predict what comes next, or perform a prognosis of what new data might look like, or somehow classify the new data samples.
So, as an extremely simplified example, if you give it a sequence "1, 2, 3, 4, 5...", then it will learn the pattern that there's a sequence of numbers, and each new number is bigger than the previous by 1, and then will give a prediction: "...6, 7, 8, 9, etc".
And if you give it something like "1, 3, 6, 10, 15, 21, ...", then it might learn that it's again, a sequence of numbers, next one is always bigger by an increasing number, and it will continue the sequence with "...28, 36, 45, etc." The sequences and the patterns in them can be much more complex, of course.
Incidentally, more sophisticated Machine Learning algorithms can do this with things like stock market prices, climate/temperature data, strings of text (maybe you've heard of AI algorithms "writing" books, movie scripts or fanfics?), images (e.g.., generating human faces or entire photographs, see www.thispersondoesnotexist.com, or, given a picture of a dog, say that "it's probably a dog"), and yes, with music data. Because sheet music can be easily encoded with a sequence of numbers or symbols, and then the algorithms can analyse it and give predictions.
My guess, again, without reading the article, is that the algorithm only generated the base of the musical piece, and then this guy just tweaked it, cleaned it, smoothed it out and fancied it up a bit, to sound prettier and more human-like.
And yes, it can't quite get the "human element". The inspiration, the "soul", so to speak. So if you give the algorithm a sequence like "1, 2, 3, 69, 420...", the algorithm will just go "u wot mate?".
You should've posted the link in the description to the original article.
EDIT: yeah, I removed the bit about the degree. That wasn't a good place to mention it, I agree. I don't know why that rhymed.
+1 to this. I am learning ML and this is what it is.
Came here to say this, but u did better
it izzzz wat it izzzz
"1, 2, 3, 69, 420, ..."
You are trying to imply something over here aren't you?
@@AnnXYZ666 next number is 504.
I am sure that even the most intelligent AI can't compose a lo-fi that is similar with Brett's.
That made my day. Thank you
Nah... maybe Brett is an AI which cannot function right with a viola. sCiENcE
You mean AI can't be programmed to such low standards ?
@@immanueldah2684 Probably because a AI programmed to do lo fi relaxing song would not reach a terror like music as Brett's
of course not, no one can match the flavor of Brett's Lo-Fi
My 91 year old neighbour, who is like my grandmother said that you guys are very attractive young men with very soothing voices. And that you are very funny too. She really like your videos and always want to watch them with me. So much love from Switzerland, the other side of the globe😂❤️
awwww so niceee
Oof my Mom has been single for a while and she said that she thinks Brett is cute and hot. I told her he's 28 while she's 34. She said f it 😂😂😂
@@arss2788 Brett is cute!!! Brett is underated imo!! I love Brett
@@hoichan5378 well me too but Im asexual but sure yeah
Aaawwww 💕
AI robot voice: "And for this next movement, we shall delve into and explore the deepest emotions sadness, anger, and pain. It is titled ctrl+alt+del."
And the next movement: “alt+f4”
B^U
Ctrl + Power + Refresh
Task manager?
@@henrykwieniawski7233 Ctrl+Shift+Esc
I think people are forgetting that the only thing the AI did was write a couple of melodies that resembled Shubert's style, that's it, no harmony, no orchestration, nothing else. The movements were written by a composer imitating Schubert and using the AI's generated melodies. That being said, I would love to listen to an actual completely AI generated movement. Orchestrated and all... Needless to say they'd need an entirely different AI learning setup. Anyway, remember people, the only thing the AI did was write the melodies.
Yep, the smartphone site clearly say that it just wrote the melodies: consumer.huawei.com/uk/campaign/unfinishedsymphony/ I’ve heard when AI is used for orchestration and it usually sounds off. It just seems like a useful tool for composers to get inspiration… for now.
I thought that might be the case. Thanks for confirming it
Someone does the test on youtube ua-cam.com/video/nA3YOFUCn4U/v-deo.html. Fully created music thanks to machine learning.
If you just want computer generated melodies... use an Arpeggiator 🤣
@@hokutodecuisine3855 thanks, I'll check it out
How my 2020 was: discovering twosetviolin and getting addicted
:0 twoset actually hearted it!
they haven’t done this in a while
honestly same
same
@@heyytheree thanks!!!!
@A.H someone else did it i copied it
eddy: im not a music expert
the entire fandom: YOU ARE A MUSIC EXPERT
Yep, he definitely is!
In terms of theory, not really.
Do not read my profile picture Yeah agree. Could say that’s one of their weaknesses.
05:04 The AI is so good, it simulated someone coughing durin the performance.
and again at 7:27 lol
well the music itself is being performed by real musicians and instruments, an ai simply composed the score
@@awkwardsaurus wysi
This AI has a real long way to go
@@mrtoast244 wysi
Damn it’s way better than expected.
hi! i love your piano arrangements!
i like turtles
@@Tokimisuna turtles like me
it's the first movement variation... no creation occurred. reiterate and change intervals
Are you serious??? This is generic elevator music. Completely soulless.
AI needs to stick to Go.
Eddy: It would be interesting to see an AI concert
Me: *VOCALOID*
edit. I'm not saying that the Vocaloids can make their own songs like the AI twoset mentioned in this video. But they do indeed put up holograms for concerts. I also agree that they shouldn't explore vocaloid by themselves but someone else that is familiar with vocaloid should recommend to them.
Sorry if I caused any misunderstandings ;A;
Ah yes, Miku.
Hahaha true
Yes this one hahahaha
I wouldn’t necessarily call Vocaloid AI software tho as people generally do all the music and tuning of the voices themselves. Honestly it’s more work than just singing in making a voicebank, making an original song with a voicebank, or making a MMD. Every single detail down to each syllable needs to be fine tuned and it’s just as much work as normal music producing or maybe even more. As for MMDs, if you don’t have experience the dancing is just gonna come out looking like those crappy ragdoll games.
But I do agree that since we already have hologram concerts, concerts with AI generated music using voicebanks and production software don’t seem too far fetched.
Petition for TwoSet to react to Hatsune Miku
AI still isn't smart enough to stop the audience coughing.
Sleeping dart turrets in the concert hall is something that can be arranged.
and them from clapping between movements
@@juliannap9452 i loooove when a movement ends and you get that awkward scattered applause from people who didn't get the memo 😂
It raises realism.
That's how you get SkyNet.
As an AI researcher (specifically from the Natural Language Processing area) the reactions from you guys are feeding my heart hahaha
But, let's take it easy as mentioned by my colleagues already AI in general are pattern finders most of the time, and we may have to wait a little longer for really creative AIs to appear.
Although I can totally see them coming with Reinforcement Learning and GANs.
Well these are my 2cents ☺️
Your work is awesome, please continue. Btw I've just bought a Violin and will go for my practices as well.
Ai will never compose any lengthy work as it lacks logical processing just like a general ai bottleneck incpabale of abstract thinking
I wonder what an AI would do if it composed a symphony, used a survey to analyze whether it counted as a "good" piece, and got a 50/50 like/dislike response. Would it conclude that, based on the nature of statistics and the subjectivity of art, that this was technically the most ideal piece it could make?
@@johannsebastienbach Is it really? Because music is at the end of the day and endless repetition of the same patterns. Now days coming up with a new melody that also sounds good is rather difficult. And what your brain process as "creative" in music is exactly those patterns (that why they are used constantly).
I have to find a composer that do not inspire himself to other previous composers, and inspiration is the pure definition of what the IA does, look at patterns and try to replicate them.
The actual weakness of the AI is that it cannot define what it sounds good. You have either to train it by telling it "this is beautiful learn this" or by giving it a set of parameters to look at and saying "this are the parameter for the perfect composition".
In any case, we have the entire western culture complaining about how role modeling works around everything. We claim that the concept of beauty is defined by the TV and News papers, meaning that as human we choose to consider what looks "good" by looking at what other people tell us to consider good. So can you please state where is the difference between AI learning from humans and humans learning from humans?
@@GabrieleNunnari I don’t think it’s that simple. I find it fascinating that even with the limited amount of notes we haven’t reached the limits. The Batman dark knight theme sounds completely unique and hundreds and thousands of songs have been made before it.
Also music evolves. What we consider today as classic was the metal of the medieval times, don’t forget that. Styles emerge and create new atmospheres. Music is 100% a transformation of information of emotions. The notes are the tools, which is why we haven’t run out of melodies imo
@@codecampbase1525 after stumbling upon "music similarities" compilations on youtube where random anime game music, classical music, indie rock and who knows else have pretty much the same motif, and then starting to notice it outside of youtube videos myself, i'm almost sure that we already did and all of the melodies are either same, with slight variations or a mix of several others with the saving grace being the fact that melody is not the only part of the musical piece.
As an AI practitioner, I can try to explain what happened. The most likely algorithms that generate the piece is called LSTM (Long-short Term Memory), or it also could be Attention Model. Both algorithms learn the pattern based on all the "training data" that they are provided, namely all Schubert symphonies, or could be all symphonies that are ever written in the similar period. The learned algorithm will try to predict the "future data", based on the "input data", which is Mov 1st and 2nd, and predict the next note, and next note, all the way to the end when it predicts the "end note". Here are some explanations to your questions:
1. How does it remember the proper sequence of instruments coming in? Because it learns the PATTERN of all symphonies, and according to some rules of symphony composing (or "code" as you guys mentioned in the video), they all happen in certain order. Once AI decides the proper timestamp that each instrument comes in, it will predict the starting notes and rhyme for each instrument, and lay them on top of each other, one by one.
2. Why is it repetitive? This also happens to AI speech generation, which uses very similar technique with AI music composition. The model is predicting the "next note" in sequential order. For example, when the current predicted note is C, the next most likely note will be predicted at G. Then after several rounds of iteration, when some future note is predicted at C again, the one after future C will be predicted at G again. In other words, if the music never ends, you will see endless rounds of same passages occurring, just like seeing repeating decimals 2.123412341234......
3. Why does it sound more like Tchaikovsky in certain parts? Easy to think about, that AI tries to learn the pattern based on all "training data", and the algorithm developer must have provided Tchaikovsky's pieces as part of training data. Throughout the AI music generation process, AI is trying to predict the next passages by giving a mixture of all composers that it learns from, such as 40% Tchaikovsky + 50% Schubert + 5% Dvorak + 5% Beethoven.
4. Why does 3rd movement sound like 1st movement? Because it learns the pattern of all musics. Not sure if it's true in Symphonies, I would say it's very likely the case in piano concertos, that 3rd movement paraphrases the 1st movement quite often, such as Rachmaninoff piano concerto 2 & 3.
@Tianqi Yang This is by far a very good explanation from a technical perspective. I was hoping someone would explain the things here. As a part of AI community, I would want to shed some light on this too.
To those who don't follow AI actively and to those who are afraid of this. Here are some facts? Take it with a pinch of salt because these are something the community is saying(not my personal opinion)
1. AI -- > AI is, essentially to date, a pattern-matching machine. It is very good at finding a very very complex pattern and has already defeated humans on this topic. But it is a machine/algorithm at the end of the day. AI works primarily from memory(which it has already learned)
2. AGI --> AGI is supposed to be a system that has the human-level intelligence. Intelligence is not defined yet. AI model was inspired by human brain neurons but doesn't mean it is fully mimicking the human brain. There are lots of things besides a neuron that controls the human system(cell for an example).
3. If somebody says AGI is already there they are lying (some business strategy). Nevertheless, AGI will come in the long run maybe.
4. somebody could argue saying that historically any new invention creates new jobs and kills the older ones (Telephone invention killed a lot of jobs but created a lot of jobs too)
5. Here is a good argument, In a world where if AI automates all the boring jobs then people will have time to do creative thinking. My understanding is at least in my life the skills like ART will have much more value than the ones which can be automated.
Nobody cares(at least from the point of music intellectuality) about AI generating art(a lot of deep learning models can easily generate paintings). Also, people wouldn't mind if bots like google duplex automate a lot of call center jobs.
wow, computer science student and twosetter and Chinese. Can we be friends?
@@tannyxie8328 I was electronics engineering student and I'm Indian. Can we be friends? 🙈.
I've been closely following the Google Coconet Ai and Magenta that was presented last year in a doodle... it is able to finish or create simple melodies with AI. It dissected Bach's compositions to compose harmonizations in the style of Bach. They trained it by omitting certain notes from original compositions and created a rating system for the AI rewarding solutions that closely resemble the original compositions. Through the accumulation process they gradually increased the amount of missing notes. Harmonization was the goal but with a few more rules and exceptions added they made it compose entire symphonies from scratch. Even the data evaluation process for rating predictions can be automated because google, through 'google music' and 'youtube music' is able to create popularity charts based on airtime numbers, gender, country, special interest, etc....
I bet their project was actually a disaster, and there was a bunch of human intervention with the AI’s choices, even editing the music afterwards. It’s easy to lie and say something is AI; Sofia’s public conversations were rehearsed beforehand.
Eddy’s and Brett’s expression going from like “Okay, not bad”, “Interesting...”, “OH SH*T”
Plot twist: Ling Ling is an AI from the future. That’s why he/she can practice 40 hours a day and be perfect at everything.
*X-Files theme intensifies*
Maybe Ling Ling is a cyborg. Robotic perfection with human emotive capabilities.
11:33 Miguel's comment is spot-on. It's not like they're claiming they created a Schubert AI that can write an infinite number of symphonies in the style of Schubert. It's not much different from any composer attempting to finish the symphony based on their understanding of Schubert. It's not perfect but it shows that technology can do and how it can help compose great music (albeit with a human composer cleaning up the finished product)
The reality is, pretty soon, a human composer will not be sufficient to "clean up" the works of a computer. They will lack the skill.
@@TheJerad1 ?? why would that be
@@RobinsMusic Just like how a human can no longer best a computer at 'Chess' or 'Go' or 'Tic-Tac-Toe', or any game, soon the best human composer will not be able to compose at the level of 'A.I. composers". Sorry Schubert.
@@TheJerad1 I'd put a lot of money on you being wrong about that one!
FYI the piece was actually finished with the help of a human composer, here’s what they said on their website:
“This wouldn’t have been possible without pairing the technological innovation of Huawei’s AI with human expertise, so Emmy award-winning composer Lucas Cantor was brought onboard to arrange an orchestral score based on the melody that the Mate 20 Pro smartphone composed to compete the symphony and perform it.”
Don’t get me wrong, the Ai is still really impressive though.
I thought the same. How can we know how much this music was really written by AI?
@@rogerio4039 Exactly, remember that "interesting" prodigy? She only used four notes as a motif to create a complete piece. Does that count as "writing melodies" too?
I ruined the 100 likes hehe
@@kyota9712 Well, Beethoven was excellent at that. Taking a motif and developing it in different ways to produce a work of stunning beauty.
C K Yess
AI: * exist to take over TwoSet's job *
TwoSet: * make a monetized content using the AI *
Uno reverse card.
AI cannot genereate topics.
@@nianyiwang yet
Nianyi Wang wait a couple of years
😄👍
@@nianyiwang actually they can, i've read an article before that says movie script writers uses AI to analyze past good selling movies and provide relevant keywords from them as a baseline to write a new movie.
i'll see if i can find that article again
ACTUALLY AI DID NOT COMPOSE THIS ON IT'S OWN . The AI actually only composed a melody that was latter turned into a full score by a human composer. Huawei hired composer Lucas Cantor to arrange an orchestral score based on a "melody" the AI wrote . The following is from Huawei own press release. "This wouldn’t have been possible without pairing the technological innovation of Huawei’s AI with human expertise, so Emmy award-winning composer Lucas Cantor was brought onboard to arrange an orchestral score based on the melody that the Mate 20 Pro smartphone composed to compete the symphony and perform it live."
Bruh
why is huawei everywhere lol
The spy Mawsta so in other words big deal. A not exceptionally bright high school student could write a program that generates a melody randomly.
That makes more sense
Fefe H Well if they really are a front for Chinese intelligence that would explain why
1:22 Lucas Cantor composed the movements... with the HELP of AI. That's really different from AI composing a symphony, and probably accounts for the Pixar/Disney vibe.
Yeah I really want to know what exactly the AI actually did and what our boy Luke did.
Schubert would be turning in his grave to hear that attempt at completion. What is worse, is that a really successful completion has already been made many years ago by Brian Newbould, who knew everything there is to be known about the music of Schubert. Lucas Cantor is a bloody fool as is his AI. If you would rely on Lucas Cantor, you might as well eat your head.
@@CSRgamer I think the algorithm didn't do much, it probably just generated a set of notes as the article in the video said, then the compositor just started his work from there.
Whenever I work with AI it always feels way too mechanical, and the learning process make it feel like a random generator (random until you get something right), even if you give them huge datasets these algorithms have an insignificant "brainpower" compared to the human's, we eighter need a breakthrough in the design or much more processing power to make, for example, bigger neural networks.
Beethoven:
“I sense someone is challenging my throne."
lmao-
Lmao
Lmao
Lmao
lmao
“Even the most beautiful music is just code”
Never has Ling Ling been so disappointed in the world.
Shubert in Heaven be like:
“Am I a Joke to you?”
Dani Par I’m gonna pretend I didn’t see how you spelled Schubert
@@kryger4840 lol
@Dani Par Schubert*
Schubert is in Hell.
@delphin bringsby calm down with your conspiracy theory. It’s not like you are a messenger from god or something. So how would you know?
6:08 this chord progression in particular, is very Schubert-esque. major props to the AI
Don’t worry TwoSet, your jobs are safe😂
For now.
Remember coppa
i wish their jobs are safe, i wouldn't want an ai telling me that i should practice instead of them
Comedians huh
RonnieWM Mok UA-camrs
It sounds scarily good but it does say in the article that Lucas Cantor composed it "with the help" of AI. I wonder how much he changed the AI suggestions. Maybe the raw AI material is not THAT good?
Now I want to see David Bruce interview Cantor about how this *actually* worked, especially the orchestration, with commentary on the "how to sound like" aspects by Nahre Sol 😆
He cudda just gotten ideas for a melody or something. Pretty vague lol
This shit right here. Most tech-journalism around AI can be super fallacious, thanks in part to the programmers trying to drum up interest. You need to check how much the AI "wrote a symphony" and how much "a human picked what the AI was to focus on and replicate". Even in the case of a 'true AI' what suggestions it will form are based on a system designed by a human, so it's always tricky to say the AI did anything on its own.
Counter-point though, a lot of people in art would argue you could say the same about human composers/artists/writers. The conventions of different genres and mediums are learned and developed from past humans' techniques and teachings, so one could say we're all "programmed" to perceive specific forms of arts in specific ways. Value judgements about an AI's ability to be creative is thus a bit confusing.
I think he orchestrated the melody that was made by the AI
I think the AI made the repetitive stuff we can hear in the 3rd mvt. There are ideas made by the AI, but the human person chooses how articulate these little patterns. I don't know if an AI can wrote an entire symphony without too much repetitions. In the case of Schubert, there are not enough datas : the AI can only see little patterns, but not a whole pattern we could find in all Schubert's symphonies.
So I think (and hope) human composers are always much better than a good AI.
The irony is that only a computer could practice 40 hours a day...
TRAINING is the new PRACTICE
Golden
Ling Ling does it since birth. Loo
Being a quantum physicist or living on Mercury can make you practise 40 hours a day.
@@fryderyk_chopin_sir_newton maybe but in either case there will be no one else there to listen!
5:07 what every you say against AI, this brass swell was so unexpected for me and hit me hard! really emotional moment, a lot of expression and art in this single brass hit
What I thought is: it would very fun to give AI an existing finished piece and let it compose the last movement, and then compare the real one and the artificial one. I'm particularly interested in what Sibeluis Violin Concerto's last movement would turn out like
I was thinking the same ting. Hope some one will do that.
To be fair, I would suspect this is how the AI is actually trained so... it could be quite close? No idea, would love to hear that too!
@@a.baciste1733 It is, most machine learning (one of AI branch, so to speak) validate the result generated by the model with real data to see whether the model is good or not good enough. (CMIIW)
This is pretty much a principle in machine learning: after training the machine (AI) on a dataset, you give it some new input it hasn't seen, observe the generated output, and compare against the true expected output to see how the machine performed
Eddy: Says "Interesting" normally
Us: panik
My 2020: discovering TwoSetViolin and becoming addicted
unoriginal comment🤣
Spending my 2020 with these guys 😂👌 especially for their reaction ❤️
AI comment? ;)
Welcome to the fold
My 2018 finding twoset and becoming addicted
I work as an AI scientist and have a PhD in music theory (specializing in machine learning/AI for music analysis). While I haven't seen the code to this particular example, so many of these hyped "AI composers" are really fake clickbait to get attention for a company (Huawei in this case). My guess is that they had a computer go over the MIDI or xml file of Schubert's symphony and perhaps other similar works. You then train a model to give a probabilistic output of what came next. Example: given the notes G, Bb, A, Schubert would then go to G 40% of the time, D 20% of the time, Bb again 5%, etc. I made those probabilities up, but that's the basic idea. Same with rhythm. You can do this with simple Markov chain probabilities or something more complex like an LSTM or Transformer model. But the end result is that you can produce rhythms and melodies that have a similar probability as Schubert and any other music the model was trained on. But you typically have to get the melodies started (seeding) and end them also. However, current AI has a hard time making motifs, large-scale cohesion, and form. I suspect that this "AI" made a few melodic fragments, and the composer: finished the melodies, arranged the form, made harmonies, dynamics, orchestration, developed motifs, etc. Also, humans performed, recorded, mastered, and produced the music. There are of course very interesting and groundbreaking developments in music AI, such as the Magenta project where their Transformer model approach is starting to show signs of learning motifs and phrase structure. Sure, perhaps in 10 years a lot of commercial music might be made with AI, but we as humans get to choose whether we want our music (not the background muzak that is made to sell you stuff) to be a pure human-to-human pursuit, sit back and let computers make all our music, or integrate AI into our musical world as we see fit much as we have for other technologies. I think music will always be a human-first expression simply because now we will pay lots of money to see people perform music when we could have saved the money and had our computers play a recording of that same music.
cool～
"I think music will always be a human-first expression simply because now we will pay lots of money to see people perform music"
Sharon Apple has entered the chat
i love you
Jesus Christ, dude. Ever heard of paragraphs?
@@MrWhangdoodles the commenter owes you nothing. they took the time out of their day to write a professionally-backed explanation; and regardless of whether or not he formatted it like a polished essay, thats pretty badass if u ask me. badass mfs dont need no paragraphs. peace and love, brother.
The 5 dislikes were virtual assistants: Siri, Cortana, Google Assistant, Alexa, and Google Home
6 now. So... bixby? 😅
Google assistant and google home aren't the same?
IBM watson
Now 19. They must be those dozens of "assistants" from Google play store's garbage side
😂😂
"It really challenges what it means to make music and what it means to be human." I am working in an AI-related field; this quote is very important and refreshing for me. Thank you.
The fact that UA-cam recommended me this while working on my thesis delving in AI is very concerning yet very insightful.
They even added a few coughs lmao
it is at 5:05
Stop using weird slang terms please, call it what it is, it's the viola
@@haoya7688 nice one
@UChfHIK-IOlptzROMWE6W0cgoh no you commented it before me
#UChfHIKIOlptzROMWE6W0cg oh no you commented it before me
"It really challenges what it means to make music, and what it means to be human, I guess." -Eddy
You perfectly worded it, Eddy. Such wisdom.
@@heyytheree Eddy's quotable quotes shall live forever
Nah. The AI is reflecting what Schubert composed according to human generated rules on pattern matching. I'd say, AI shows us more what it means to be human and to make music. Note how quickly Eddy and Brett detected it was like Schubert, but not exactly like Schubert. Even I, an amateur musician, thought the third movement was very much like the first, and essentially a slightly changed copy. A human composer wouldn't do that. There's a story to tell in the music. You might have some phrase repeating or evocative of the earlier movements, but there'd be surprising differences as the story progresses.
I think of Beethoven's 9th and his 1st, 2nd, and 3rd movements and how they refer back to each other, but lead inevitably to the fourth.
LING LING GRANGER Definitely! Maybe remove the I guess though... to make Eddy sound like the true genius he is!
Musicians: *practice and struggle to get better, 40 hours every day*
AI: cha cha real smooth
An AI does need hundreds of hours to be able to do something.
I once made one that played minesweeper and it took it 3 days to learn how to play ok. And that is hundreds or thousnads of cycles per hour.
Doesn't that depend on your computational power? I mean, I think that Alpha Zero, one of the best chess engines in the world, trains for around four hours and defeats Grand Masters.
@@mariopalenciagutierrez4318 What did you do to overcomplicate an AI so badly that it can't play minesweeper properly for days? It should be easy to make one that plays perfectly without any machine learning involved at all, meaning no learning time. Even machine learning based a standard computer should learn minesweeper in seconds, not days. Don't base technical limitations on your own bad code. "Hundreds or thousands of cycles per hour"? It should be per second if you implemented it correctly.
@@gronbuske i did it through machine learning.
Yes, i know it was hundreds of thousands per hour.
Took a couple million games for it to learn by itself.
@@mariopalenciagutierrez4318 I do hear what you say, but even a few million games should take seconds or minutes, not hours and days. Post your github or wherever you put your code and I can probably fix the issue for you. A minesweeper round done by a computer should take a few milliseconds, you're saying hundreds per hour meaning every round took seconds instead. That only shows you have some serious bottleneck in your program, not that "AI does need hundreds of hours to do something"; that is absolutely false and misguiding.
“If it sounds good it is good”. Duke Ellington
I think one or two years ago Google created a Doodle which was something like a tribute to Bach. We can input notes/ our own melody to that doodle, and the AI system would compose, within a few seconds, a 4-part polyphonic melody.
Afaik they already review it
"Hey Siri, compose me a symphony, please."
Siri: "Okay..."
Their faces for that split second was like it was all over.
Year 3050, the study of the history of music
“This legendary symphony was composed by AI number 4040. It’s a really good symphony”
@Maybelle Lee well hello again
@Maybelle Lee I see you a lot
It's sacrilegious
@@heyytheree Obviously
@@heyytheree Why? Humans want to listen Beethoven because its Beethoven. Will you replace someone's entire musics because someone's a little better musician?
AI won't be interested in replacing Beethoven. More likely to be interested in creating uncharted novel classical music industry market (by its creators).
Aw hell yeah, Schubert appreciation! Literally the first exposure to classical music that I remember was the theme of the Canadian cartoon Little Bear (I obsessively watched the reruns on UK kids TV), which was an arrangement of the Allegro vivace from Schubert’s Sonatina in D Major, and I absolutely fell in love with as a toddler/little kid 🥺. I definitely wouldn’t have picked up a violin several years later without it, so I owe an awful lot to one extremely talented Austrian dude lmao, thanks Schubert 😌
watching this made me realize... it would totally be amazing to have a twoset podcast where they just discus and dissect classical music and maybe even have classical music friends / experts on the show
Oh nooooo !!!! I was working on similar project.........
F 😞
ahhhhh, what a horrible feeling.
F :/
Be the first OR be the best. If you're into coding, you must know making all the money oftentimes just comes down to taking a trending app, improving on it and releasing that. If you want to be a bonafide inventor/creator that feels like a whole other mindset, and keep in mind the "13 failures for every 1 success" parable. Don't stop.
It was my project toooo..noooo😭😭😭😭😭
David cope wrote an entire opera using AI years ago. This is not a new thing.
I'm actually interested to see how it would finish Mozart's lacrimosa starting from bar 8
same-
Same too here
Actually this would be a very interesting way to test it: feed it most of a composer’s body of work, give it only a taste of one completed piece, and then let it try to finish that piece. I wonder how different it’d end up compared to the real thing.
@@mrsb50 exactly what i was thinking
It was very nice to see them acknowledge both sides/perspectives. I like it when people say "I am conflicted" because most of the time it means they've tried looking at it from multiple perspectives. Saying that also means that rather than committing to on side too soon, they would rather gather more information to make a better decision.
Edit:
Personal thoughts. The idea of AI composing music is pretty cool and intimidating! Music is still a human thing though.
That tragic and soaring climax after the buildup at 8:00 , literally right when Brett said, "Fly!" was really quite spectacular and unexpected. I wonder if anyone can recognize a similar passage from other classical works. This has actually inspired me to begin a listen through of Schubert's and Beethoven's complete symphonies and more, outside of my Romantic piano comfort zone.
The ones that come to my mind are Sibelius concertos, it sounds like late romantic/contemporaneous
I got so many chills with that
“Even the most beautiful music is, ultimately, just code.”
THEM’S FIGHTIN’ WORDS.
The first four bars of eine kleine nachtmusik is nothing but Tonic and Dominant chords. No tensions, no passing notes, just simple arpeggio. If Mozart didn't exist, it would never happened. The simplest example of striking genius.
ua-cam.com/video/MeaQ595tzxQ/v-deo.html
Beethoven's 5th symphony also has a similar four bar opening.
So So So Me──。Fa Fa Fa Re──。Only four notes.
Can AI do that? I doubt it. A machine doesn't recognize potential beauty mined deep inside the simplicity.
To understand that, a machine has to analyze human being itself. In another words, a machine has to be capable of enjoying music to create one...lol A machine also have to be a genius to produce a good one. Letting mother nature do the job is far easier way.
@@ttwiligh7
It's not too far though. AI is a learning software, it can learn to do so. Like any master composer, they first learned theory in order to create beautiful symphonies. They were taught, trained and became masters of their craft. No toddler can play Beethoven without being taught first.
This sounds very elitist, in my opinion.
@@kotarodesu_23
You can easily end up with garbage just by following rules, unless you have a talent....something AI lacks most.
If it was that easy as you say, everyone can become a great composer by sheer effort.
Let's say you study Bach. Will you be able to write something as deep as Chaconne? Not a chance. How do you expect a machine to express those emotions when it has none? A machine is less than untalented people in that sense, to recognize beauty. How do you judge if a symphony is beautiful or not? Is there any methodical way?
Only if we know how to become genius, and can break down talents into zeros and ones, yes, it might be possible.
And this symphony, is mostly human's work....Not like AI came up with orchestrations. They say only melodies. Not even close to composing a complete piano piece.
If AI could make something like Chopin's piano piece entirely by its own, then I will believe.
@@ttwiligh7
AI is still in its infancy. And like all infants, aren't capable of much. I'm not arguing on the side of AI nor on the side of this elite perspective. You choose to believe in its fallacies and shortcomings whereas I believe in the potential, and that says a lot about you than you think.
@@kotarodesu_23
It says a lot about how you see music.
Even most talented musicians can't become Bach. Then why you think that is possible for a machine? Yeah, it's like a infant, but not every infant has a talent. My question is, how do you program a talent? And You don't seems to have an answer. Just believing? You think you can learn a talent? That's why we circulate the same argument.
If somebody can program AI to be like Bach, he could be Bach himself. There are just so many things you can't just learn methodically, by zeros and ones. Still in a dark, still a mystery.
As I say before, if you can break down creativity into computer language, it might be possible. But that require astronomical bytes, and when it's done, AI must understand human emotions to sense beauty and judge its value like we do. Not going to happen in near future. That's what I'm saying.
it kinda scares me that AIs are able to do things like this.
Same. I just feel like they are taking over the human population slowly, without letting us know.
Lmao there's even a hypothesis/theory that the universe is an AI simulation. Even made 3 movies about it.
Btw, the AI only wrote the melody, the orchestra score was made by a real composer
@@noo4449 how did you know
@@mynamemylastname5620 it said so in the description of the video they reacted to
I think it would be *more* interesting to have AI finish a work that *is* finished by only giving it a partial score. Then we can see/ hear how close it comes to what the composer *actually* did.
Bach's last fugue. Mozart's Requiem, and more importantly, the Amen fugue
Schu-Schu: *Exists*
Ling-Ling: *Finally! A worthy opponent! Our battle will be legendary!*
Bruh how do you compare those ?
Kung fu panda?
I liked this comment but quickly unliked it. It should stay at 69.
alphago flashbacks
The John Henry story of music lmao
You guys say that LingLing will come and save us (from AI)...
But what if LingLing IS some sort of AI...?
Would we need to be saved from LingLing? If that were to happen... Who then would be saving us now?
Are we even safe anymore?
Btw, lovely video. Refreshingly different. Hope to see some more
Lmao bro we on the same frequency, i wrote the same comment word for word before reading urs. High five!!
@@oo6127 haha yeah man! 👏👏
LingLing was always a transcended concept/being, so i wouldn't put past it
@@bakedmomo5693 but, what IF
shushu and lingling are gonna get us D:
Schubert be like: "I'm composing and reincarnated as an A.I."
I would see this movie
*I’m decomposing
I'd watch the anime
This is an isekai through and through. I bet if you search hard enough, you can find a light novel with this exact premise
Hello Classicaloid
Ooh, whoever did the recording of that orchestra actually recorded in stereo with decent separation and did little blending when mastering. Very nice. It gives the recording a lot more spatiality to the sound stage and enhances instrumental imaging on that stage. I really appreciate that.
Honestly, there is something about every person's art- writing, paintings, music- being unique and the fact that no two people can create the exact same piece of original art from scratch that has always been strangely comforting to me.
But with the years there have been many a time when I came across a line of writing or a piece of music and was shocked by the similarity it had with some of my works and some of others'. Sometimes whole lines of writing or phrases of my original music that I thought were truly "mine" were the same as works I'd never seen before composing or writing.
This has only lead me to believe even firmly in the philosophy of art for the sake of art. Because, thankfully, there is no winner in art.
Sorry for the unrelated rant aahh.
The thing a lot of people don’t know is it’s going to imitate explicitly what it hears. Brilliant, shattering ideas that take us in new directions can’t be improvised by a system which produces the same pattern over and over.
That and I want to watch people struggle through the art of composing and performing to understand their life through their art. It’s like kind of the point.
Edit: also tho it’s damn impressive.
Eddy: the day has finally come....
Me ... to play the Sibelius concerto?
Eddy: ...classical musicians are going to lose their jobs
Me: oh :/
SIMP SIBELIUS!!
same
In like 10 years, they might. But even 10 years isn't accurate enough. A.I. is still at its toddler age. It will take a lot of time before it stands.
He never promised to play it but if he does, the earliest he would do it is for the 3M subscribers, which they didn't reach yet.
I thought it was going to be the unpublished Sibelius symphony.
I'd love to play the whole 'unfinished' symphony with the AI movements included. That big melody that is high up in the strings sounds like it would be so much fun to play.
5:05 The AI is so op it included coughs as in real performances
I heard that, too! Was that really AI-generated? It had an echo sound as well, as if someone had coughed in an auditorium...
Maybe the AI only wrote the notes, and actual people were still playing the piece
I’m pretty sure it was only written by AI it was performed by real people
No, but it would be funny if you knew that the AI might've thought that coughs were an "essential" part of the movement. The generated movement weren't probably a sequence of sheet music but rather an actual audio sequence.
I wouldn't be surprised if it puts many coughs during the "performance"
Schu-Schu is a sibling of Ling Ling
But they're the disappointment
@@lalalalais Ouch I felt that one
The "jumping off cliff" part around 8:00 actually sounds like Ghibli films.
Omg YES
I imagined an off-the-cliff wingsuit glide
Can anyone link me what are they listening too? Thanks mates
YES
When I heard that part I got flashbacks to Ni No Kuni
it is just amazing to see you guys enjoying music so much and feeling touched through time and space !
Phone of AI Schuschu: *writing a masterpiece*
My phone: *still stuck at the Nokia ringtone*
The Nokia ringtone came from a classical guitar piece tho
I think it's spelled like "Schu schu" since the AI are trying to finish Schubert's unfinished symphony.
@@gauripande9273 oh yeah right i forgot about that😂😂
Imagine when you get called your phone starts composing a symphony on the spot
@@andresbourgonjon6935 that would... actually be quite pleasing 😂
I love how they listen to Schubert's music. Also, I think the AI composition is impressive from a technological perspective BUT definitely not Schubert.
Honestly that might be a matter of pumping more resources into the training of the AI: beefier hardware than a smartphone + longer training time with more human experts involved in guiding the AI would likely yield more impressive results
btcprox the hardest part is programming in things to analyse. Explaining in code passion is quite hard.
@@LordOfFlies True, which is why it's not actually a useful approach to try to directly define subjective qualitative characteristics. Instead a more plausible appraoch may be to let the AI learn through numerous examples and through constant human feedback. Over time it might learn certain examples of patterns that invoke the intended emotions within the average listener.
Plot twist: Ling Ling is ACTUALLY an AI
thats why he can practice 40 hrs a dau
Yep with his multicore processors, he can practice multiple sections at the same time.
The AI is Ling ling's student
Distributed computing y'all
"Give it 20 years"
AI: "Hold my beer"
People find it almost impossible to grasp the exponential rate of AI growth.
Software of this kind will be 100 times better in 18 months.
There didn't even exist computers 100 years ago
@@homailot2378 nobody said 100 years
@@MrSpacelyy in the video they did, 1000 years in fact
Not to be rude but watching 2-Minute Papers doesn't make you an expert on machine learning. There's always a ceiling with a given set of technologies and eyeballing a problem like composing a symphony from afar doesn't tell you where it is.
@@homailot2378 really? I was responding to the comments. Could you remember at what time it was said?
From the same creators of Ling Ling, Ding Ding, Lang Lang...
here it comes: Schu-schu
Something important to note ... the AI did not orchestrate the music. The orchestration was completed by Lucas Cantor ... which is probably why the final two "movements" sound like film soundtrack. The AI generated the "melodies" for the two movements which Cantor drew from (it would be "interesting" to hear those melodies). This is nothing but a stunt from Huawei.
The bass line movements and counterpoint are too good.
If you look at it as AI assisted composition, I'd say it's a bit more than just a stunt. We're seeing a glimpse of the future whether we wish to acknowledge it or not.
It is VERY close to a fake !
It reads: "The Mate 20 Pro smartphone listened to the first two movements of Schubert’s Symphony No. 8, analysed the key musical elements that make it so incredible, then generated the melody for the missing third and fourth movement from its analysis."
And then: "This wouldn’t have been possible without pairing the technological innovation of Huawei’s AI with human expertise, so Emmy award-winning composer Lucas Cantor was brought onboard to arrange an orchestral score based on the melody that the Mate 20 Pro smartphone composed to compete the symphony and perform it live."
Summary :
a computer has produced (we dont' know how many) melodies ; and a human expert has done ALL THE HARD PART !
Plus : where is the scherzo ? This 3rd movement is clearly a 1st movement !
Your's is the most important and relevant comment I have read in response to this video. And it's exactly what I was expecting was the case.
I agree, it was the human element that made it what it is and not purely out of the model. But as a ML and CNN practitioner myself, it might be possible, feeding the model with enough data can possibly result with a piece written by an AI itself.
Zeamoose Akino - indeed. The brain is a NN so (if you exclude transcendent properties such as souls or spirits) why not? But how many parameters? 10^18 ? Because your NN will need to have enough memory (forget LSTM ...) to maintain the consistency of the piece (provided it has learned the Sonata form, rondo, scherzo-trio etc.).
Maybe you can tweak the topology to that end. But still, I am not sure that’s a smart way of spending terawatts.
@ Musical composition is a teeny tiny fraction of our brain, and music itself is simple AF. The 25 minutes of part 3 and 4 are expressible in less than 3KB before applying PCA. Lets say that the solution space is
@@Idiomatick - the point is not to contain the information of this symphony, or even be able to reproduce the symphony, but be able to compose a new one that sounds like Schubert.
The ML should learn : how Schubert composes melodies ; how he builds a sonata form (after learning the principle of sonata forms) ; how he conducts a development (harmony progression, theme variation, etc.) ; how he treats each instrument in the orchestration, including usage of registers and combinations of several instruments ; etc.
i'm late to this but HOLYYYY this is giving me an existential crisis... I really like how in the ending they briefly brought up the topic of what it means to create music (or just art in general) and what it means to be human and I really wish they could talk more about that in another video. This is mindblowing.
AI taking your jobs?
Lamentable.
Here is how this works: the more data, the more accurate the model.
This test was a bit unfair. Make it digest the entire Schubertian corpus and see what happens.
I read these as corpses 😂
Ok, now I have been convinced not to drop my additional AI course at uni... Thanks Twoset and Schu-Schu...
funny because xiu xiu is actually a music band already lol
@@TheHmmer4 that came off as really bitter. OP was just thanking twoset for motivating them to learn something new. Listen to some Schubert and chill
TwoSet unironically roasts Lucas Cantor's phrasing and mixing whilst trying to critique an AI
Maybe it should turn into a roasting channel.
Shushu, Lingling, Chingching. We unlocked three of the musical apocalypse horsemen.
It’s nice to see twoset just listen to music once in a while
Lucas Cantor: I used an AI to finish my symphony
Symphony Composers from 1600-2000: Coward you couldn’t finish it on your own
I imagine them buying tickets to see a ballet, getting too wrapped up in the music and just pulling out their violins and playing along.
Compared to the child prodigies, *this is an advengers level threat*
tf are advengers
Lol
@@iandugger1168 YOUR USERNAME IS AMAZING
@@iandugger1168 How do you not know what the Avengers are?
@i play the piano so yeah 😮
I do wonder if we feed the AI some Paganini, would it spit out a piece so difficult that it's impossible to be played by human?
Humans with talents: **Exists**
AI: Hi
Honestly, I think AI is very people extremely underestimate AI’s future in music. Music boils down to a format of patterns, and AI is exceptionally well at picking up on patterns…
I think the music industry will be heavily effected by AI by 2030.
I have been making ai music last year, there is a video on my profile you can check, its extremely good and takes a split sec to make! So yes.
AI will change everything. Not only our music.
Music is not pattern, music is based on pattern which is created by human. Based on does not mean strictly follow. Music like Jazz and Flamenco often have variations of the basic theory and evolves overtime, don't play by score, play by feel. AI can only follow, cannot create.
@@wonghow actually, an that is what an AI is best at, it is best at following and creating indirect patterns. Afterall, creating is literally what an AI is made to do.
Following a strict pattern is what a traditional programming script is best at.
@@redthunder6183 AI only learn from certain patterns, and rules, does not create. Humans don't need to use patterns or rules, that's why you don't understand music. A player can create based on feel. "El Duende" That's why AI means AI, a robot, can't feel, can't think, can't listen, not suited for everything.
"oh unfinished. it shouldnt take long"
Homer im sorry to tell you but-
Sad when the competition in classical music has become 5 year-old asian kid VS a robot
The AI is even worse
That “3rd movement” was just an upside down version of the first movement
This
Upside down how? Inverted, retrograde, what?
Aphrontic Alchemist what I mean is...it sounds just like the first movement but altered. What the first movement might sound like in a parallel universe. Not literally “upside down”.
Like they spun the record backwards and wrote down the notes. Who cares though really, it sucked
But the AI sounds more exciting
Will smith: Can a robot write a beautiful symphony?
Robot: can you?
I kinda feel this on a visual arts level, to an extent. You know, people using filters and photoshop on their pictures to make it look like an oil painting that they then print out on a canvas and hang on their wall, instead of paying an artist to paint their portrait and thinking they are still getting "the same effect." I mean, on the one hand, I can understand financially it makes more sense, but artists really do get the shaft all the time. And, personally (though I could be heavily biased), I think those "filter paintings" look really tacky nine times out of ten.
LOL...give it time, my friend.
I work in IT, and seeing how tech progress over time is a bit scary.
It will get perfect faster than you ever thought.
Practising a centuries old art, don´t be surprised if you are left behind. Imagine masons or glasscrafters being pissed off because machining makes their craft obselete.
As an artist, shouldn´t you be glad of getting more tools to express yourself?
@@avenderiel Of course it hurts people because they mastered and practiced it for most of their life and they'll get replaced by an AI which only trained for hours or days. Sure, they could express themselves more but their former passion to do what they do will be snuffed out. Imagine some kind of Ling ling that'll more often be better, faster, and more popular than you.
@@digimitesh2246 AI killed the Concerto star.
@@avenderiel Yeah you're right. But our only choice from here is to go forward.
To all those saying early or first..
Go practice
Damn it! You caught me!
I did. Practice being early...
I agree,what about you 🙂🙂🙂🙂🙂
The truth
I am a Computer Science PhD student and this article is just inflating the whole AI thing. It is still quite far from actually being able to replace a human. But we are trying 😂
Don't try!!
Why? What for?
Exactly lol. Everytime I see the hype about AI composer I get super skeptical.
Art is supposed to reflect HUMAN experience. There is no meaning behind AI "artwork" as algorithms don't have human experience.
My name is Connor, I am the violinist send by Cyberlife
I was just waiting for you to say it sounds like film music. :D When I heard the first tone of that AI music, all I thought was "it sounds way more modern than your standard classical piece, and definitely more modern than the passages from Schubert's symphony that you showed us". You gave it a name. It is repetitive... I'd also say there's this air of... sterile cleanliness, if you know what I mean? It sort of lacks the depth and dynamics that I would expect from a classical piece. Mind you, I am by far no expert, but it just sounds like the modern style that has lost some of the classical flamboyance (is that the right word? I dunno...). And it feels like the tempo doesn't change at all (that's probably what you meant by abrupt). So in my opinion, AI still has a long way to go. Thank god, your jobs are saved. :)
Interesting part about music being just a code - well, in essence, everything is just code. A bunch of ones and zeroes if you strip it down to the basics, because if you get down to it, our thoughts and biological processes are just a bunch of electrical impulses. So the thing that we call our soul is just code and everything we produce is code as well. But do we really want to think like that? :D
No worries, the AI only practices 20 minutes a day when the phone isn’t too busy watching cat videos.
This reminds me of when Vocaloids came out and the voice synthesizers became a thing. A lot of people were angry, claiming that only humans can sing, that they were trying to make "robots" take the singers place, but it really depends on how you see it.
In the case of vocaloids, instead of comparing them to singers they should be compared to instruments. By using them, composers were able to make their songs without having to rely on someone else to sing. They could express exactly what they wanted, its not different from expressing yourself through a piano piece.
I believe the AI would work like that. It is not there to substitute composers, but as a tool to maybe help them. I hate this mentality that people cant use technology for their creation or else its "cheating". Like when people claimed that digital art is not art compared to a canvas paiting. They are all art creatd with different tools and have their own particular dificulties, theres no such thing as one being easier or superior to the other.
The AI could be used to help people, specially younger composers. Not to be just "ok AI make the symphony for me" but to help with inspiration. Don't composers sometimes find inspiration in other pieces of music? It could be like that.
People need to stop thinking technology exists to substitute things, technology is supposed to help use, make things work easier and quicker. Like if a digital paint takes less time to make compared to an oil paint why is that a bad thing? That means that digital artist now can make their creation with less time and thus have time to create more.
As an artist, I'm afraid I disagree. Good things take time, work and care. It's not about speed. There is a certain satisfaction that comes with learning how to master and manipulate the medium, be it oils, watercolors or pen and ink. It's the feel of the brush on the canvas, the smell of the turps, the effect of the light on the pigments. There is a more personal quality to a painting on a canvas which digital art can't quite replicate, though style and composition may vary.
If I may, the comparison of the vocaloid to the man playing the piano doesn't hold up: human voices, like pianos, are both instruments, and in both cases it is a human being who is manipulating them to produce the music.
Twoset have shown two episodes now where they displayed the talents of new composers who submitted their works. They seem to be doing just fine without any assistance from machines.
Where technology helps, I'm all for it, but where it seeks to take from us those things which are intrinsic to being human, it defeats its purpose.
@@studiodensmore4979 Well agree to disagree then. I didnt meant to say that digital art is something you can do fast, artists take hours even days to finish a single image. And yes its not a physical brush or paint, but that doesnt devalue their work or make it easier to make than an actual paint. I dont think I expressed myself well in that regard so thats my bad. Thats why I said both digital and non-digital art is still art. I personally dont see digital art replacing traditional art at all. They are different, they require different techniques, both have their artists soul and heart in it and are both beautiful in their own way.
And I never said composers are having a hard time at all, I meant that it could be an extra tool for some not all. That was just an example I definitely dont think anyone should or would use the AI, but I also dont think that if someone did, that their work should be undervalued or compared to those that dont. I just dont believe in this old mentality that everything must be classic, old-style, made by hand to have value or to be considered art. Specially because art appeals to people in different ways. I know people who claimed digital artworks are empty, but I myself look to know pieces like Mona Lisa and other famous paiting and I feel absolutely nothing from it. That doesnt mean any of these pieces are bad, they simply appeal differently to each person
Well said
While I'll admit that is does take some practise to master digital "brushes" etc, it doesn't take nearly as much skill as actually painting. Digital Does make it easier-- you don't have to think about the drying times of pigments, of glazes, of the water retention of the paper, of having south facing windows, and a myriad other factors.
I have seen some beautiful images rendered with digital media. But the same painting done twice, once digitally and once in real life, by the same artist, will have more life to it in the natural- and be more valued by the artist himself for the work it took.
As for a composer's works being devalued for using AI, this is a slightly different argument. For digital vs natural paint, the question is about medium, but in either case it is still something used by a human being, who dictates the outcome. For a composer using AI, we are talking about a machine doing the work For you. In such a case as this, yes, I think the work would have less value, as the composer would simply have sat back and let the machine do all the work. (Though this is just speculation, as I don't believe a machine could ever compose; how could something with neither heart nor soul be inspired to make music?)
You don't think something made by hand should have more value than something machine-made!? Have you never made anything by hand yourself? And if you did, and offered it to someone, and were told, " oh, I have a machine that can do that," how would you feel?
Part of the value of learning to sing, to paint, or compose, is the learning process itself: of the confidence gained in overcoming difficulties, not bypassing them. Of the joy and satisfaction of mastering a skill, and sharing it with others. In short--part of being human.
Satsu Jin yea, digital art has its own problems with the creation process, it takes just as much skill as traditional art with paintbrushes and stuff, I like traditional and digital art but I kinda like digital art more, because to me it just has that kinda life to it that traditional art kinda is, missing I wanna say? But yea, I like digital art better than traditional it takes just as much skill as traditional art to master, art in any form is still art.
Brett: Hey siri, compose me a symphony, please
Siri: Okay, I found this on the web for compose me a symphony please
Brett: Nah, Siri is back in 1990s
Siri: So you dare me to compose a symphony that will ruin your future, Checkmate!
i’m actually playing the unfinished symphony in my orchestra currently and when i was telling my teacher about it she mentioned that she played in a recording of an ai completed version… so i’m pretty sure my teacher is one of the bassoonists in this
Really love how they got lost in the music while listening to Schubert's unfinished symphony.