this is SCARY as a composer…because people who we will work for might not be able to see the difference between AI and human…this is a fear becoming more valid by the month
I was 6-0, and it came down to a few things: 1. AI doesn’t understand cadences or ends of phrases. Especially in the film and pop rounds the AI felt a real lack of musical direction and definitive « end points. » The Baroque AI also had a lot of weird out of place chords. 2. Certain notes. In the romantic AI the violin hits a B naturaly against a Bb in the orchestra melody that was super jarring and felt like something that a real composer wouldn’t do. 3. Recording artefacts: In the contemporary human recording, there was a slight echo from the hall they recorded in that the AI recording didn’t have. There were other things too (Violin tone quality, etc.) that tipped me off but the big thing is that while the AI can really closely emulate the sound of an orchestra, they don’t yet have the knowledge of how composers actually write music, structure chord progressions, etc. To make it sound truly human.
I would say the major thing AI does not have is a multilayered idea within the music. Each peace of art has several layers which you can discover during your journey. But AI has only one layer. That’s it.
Yeah... The thing is, most of listeners don't understand any of that, either. I can only assume that this new tool will find its place in the ecosystem kind of the way synthesizers did, as another possibility to do things.
I, too, got 6/6 right, and mostly for similar reasons, even though (or maybe: because) I intentionally ignored the sound and only assessed the compositions. However, to nitpick, I both agree and respectfully disagree with your second point. I agree because this was also the exact point where I realized this was the AI one. I disagree however because first of all, it's rather a c flat over b flat than a "b natural" over "b flat", and this suspended none over a a dominant chord is actually omnipresent in most romantic composers' music. The reason, in my opinion, that it feels "out of place" here is that it does not occur in a "natural" or "organic" context: It's an element that would usually mark a point of high tension - but melodically, harmonically, and dynamically the tension is already decreasing/being released when this minor none suspension occurs: We have already been in the dominant for one entire bar (no harmonic progression), the melody reached its apex in the previous bar and reaches this suspension from above rather than below, and the dynamic goes back. However, to circle back to your main point, I, too felt that the AI compositions clearly lacked what I would call "musical dramaturgy". It's as if, when tasked with writing something in the style of, say, Shakespeare, it would produce a series of sentences that, on their own, sounded completely Shakespearian, but, together, did not actually tell a Shakespearian story.
Yes, I have noticed similar things... It is even more evident when you feed it with an audio source that the AI can use as a starting point for generating the music. AI is very lacking especially in harmonic progressions: Its music always conveys something static as if it were assembling material that is aesthetically related but without a clear destination to head towards.
the only thing I got from your comment is that it's PhD level people like yourself who can even tell a difference anymore. I've been playing piano for 40 years, I can't tell. Ask my T Swift listening wife and she'd just say, it's easy listening classical good for reading!
What is scary to me is not that AI might surpass human music in quality or soul, but that it will be just passable enough and so easy to produce that it will be all that the majority of people hear.
This is the danger of modern music and consumerism. We've made human produced "art" so mechanical and soulless that people aren't trained to seek out authentic art
@@sivad1025 And it is like with performing music. Did anyone hear of anybody who booked a whole professional big band or choir-and-orchestra (incl. conductor) for, say, a wedding? --- Mostly, they book a one-person-plus-midi-keyboard-band. On booking platforms, those people write that they are willing to pay 500 EUR maximum. --- And of course you don't hear of any of those who only play already recorded music at their parties. Then, if AI music of a certain quality will be so low cost and produced in no time, who will care about human composers, studio engineers, musicians, instrument makers etc? --- Already, most people are happy with a low quality sound from recordings and electronic devices. A mobile phone is enough for them, they will never know how a really good, or even: excellent, hi-fi sounds like! That's so shockingly dumb: someone walking down the streets while listening to music from a squeaking device while good bass speakers would be vital for that genre!
@birgits.3702 Totally agree! I'm so glad you said that though. I hope to get married this year and I've been playing with the idea of investing in an orchestra accompaniment. My church occasionally has their orchestra accompany mass. For Mother's Day, they played Mozart's Coronation mass. We just need to be deliberate in supporting real music
The popular music is so bad quality that it does not matter if many listen to AI classic. Creative minds will alway stay with true human music from the past.
This is exactly the thing. Normal people do not hear these fine nuances, they literally are completely unaware of what we are discussing here. It will be cheap and easy.
I got all of them correct, the way I could tell is that AI doesn't seem to care for the beat. Sometimes it feels like it just randomly switches time signatures, or 2 different parts don't exactly match up.
@@sarahyeoh0111 I think he means the meter (the time signature). That's what I reacted to. It will seem to be in 3/4 and then in 4/4 and then in something where you can't even really be sure what meter it is. (In the music from the periods where this was not commonly done.)
I think the difficulty in the film music round speaks more to the declining quality of modern television and film music, which tends to act simply as "sonic wallpaper".
Film music is declining by choice, not because of some crisis. Sadly film directors now reject the idea that film music should be significant enough to stand alone as an independent voice. They think this creates a double narrative, treating viewers as naive and disturbing them. Of course I don’t agree, and I refused to write music for films.
Early Classical-era DOES sound a bit more Baroque. It's technically "Galant" music - CPE Bach and early Haydn are good examples. Late Beethoven would present a similar problem - too "out there" to represent Classical and also not quite like any Romantic-era music.
Where things break down is... e.g. J.S.Bach. A large data model based on essentially age-old Markov chains does not step outside its fundamental constraints. Easy example again, ChatGPT never produced an original text that argued in new ways a certain point. It, per definitionem, can't. Move that into the sphere of music... the limitations are obvious. But clueless tabloids and then investors going for hyped BS about AI fundamentals they don't understand completely take over.
"Does it go to somewhere or not" - this was a good recipe, helping me to get all. For my ears, the important difference is a meaningful composition vs. a senseless composition - more than the "sound" of the audio. To add here: the selection was REALLY GREAT! It allowed the composers to shine, we heard really hidden masterpieces in all genres. Great-great selection, congratulations to your team, Ray!
I went 5-0, "guessed" correctly. I don't think I can say my assessment is better than Ray's, but it did came down to "feeling". The 3rd one, the 2nd romantic felt well structured, human touch indeed. And the last one, the 2nd song felt uplifting and a bit more distinguinshable.
Agreed, the AI generated music will play the right notes in their own independent bits, but it doesn't feel like there is a 'purpose'/'idea' to it, which makes sense... since there isn't, it's just a generative algorithm. Just like pretty much all AI generated stuff, it seriously needs an editor. Also I think pop music could be the hardest one by far. The tracks are generic enough already
Yeah, Ray seemed to be more focused on the technical aspect of the playing, whereas I thought most of these were pretty obvious from the composition side. The AI tracks can never put together a cohesive melody.
@@Tweeteketje I think the commenter meant the 0 as the one they did not guess right. Definitely the human touch! I guess they pass as not AI for the time being.
Got them all. I was applying the same rules I use when either reading or listening to verbal content. AII has weird repetitions. Inconsistent "pronunciations" or throwbacks to something earlier in the content in a way that just doesn't fit - that's why a human editor is needed to polish the content. I had to laugh about your reaction to the Bridgerton theme.. I'm sure I will get lots of negative comments here, but that's the wsy I feel about the whole series - it is artificial and tries too hard to be accurate to/reproduce something it isn't. It's very funny that is reflected in the scoring as well.
Hi Ray Chen! It’s FamousePianist from Tonic. I know you probably won’t see this message, but I have to say I love your playing and I am so thankful for learning about Tonic. Tonic motivates me to practice and I really enjoy the opportunity to listen to you and other professional musicians practice. ❤ Thank you so much! 😊
I think it is quite easy to understand the difference. Especially in old classical music where music tends to be derived from speech and tends to have a rhethorical speaking quality such ML based systems cannot reproduce. With contemporary music it is harder, as there often you need to get the big picture to understand the design. The systems behind this are good at reproducing the generic aspects of a style, which also explains why it works so well for pop music ...
I'm an opera singer - my performance repertoire is over 50 arias composed by Mozart, Bellini, Puccini, Rossini, Verdi, Vivaldi and dozens of other composers - I sing in ten languages - and like Ray, I can distinguish between the music composed by robots vs. the music composed by humans - but probably soon, this distinction will no longer exist.
It will. AI has no attitude, stance, viewpoint, personal history, passion, soul. Competently generating some snippets of music in a given style and genre is not equivalent or surpassing a human composing a C Minor Mass. The Illiac Suite is from the 1950s. It was rudimentary AI with astonishing end result. We are now 70-ish years later. Has music died? Has it been overtaken by computers since then? A lot of the panic and utter nonsense surrounding current AI misses the fundamental, central aspect of it.
Please share w/ us your brilliant accomplishments and achievements in life - so we can all appreciate your brilliance - if you have no impressive achievements ... let that be a reminder to you that you're not very bright ... Then maybe you'll refrain from presenting yourself as an expert on a topic when your awareness is adolescent - at best.
@@WhirledPublishing Meaning?... how does this relate to a specific aspect of AI development? Apart from it being a logical fallacy (see ad hominem). Plus... how is this a response to a discussion on how AI cannot take over human creativity for quite some years?
The most inportant thing to remember about AI is that at any given moment, this is the *worst* AI will ever be. This was instilled in me at a talk I went to like 6 years ago, well before AI started to make its current dramatic progress. The speaker noted how important authenticity and creativity will become in the future. At some point, AI will be good enough to be indistinguishable even to the majority of experts. Thats why we need to value authenticity and live performance in and of itself.
As wierd as that is, good AI will have been effectively pruned and selected from a large pool of humans judges so that, despite its synthetic origins, it would nonetheless deliver a human-biased performance. Think of it as ready-made pizza vs freshly, homely pizza. Chances are that, in a casual context, you'd stop telling the difference between the two.
Violinist/composer/arranger/sound engineer here, I guessed 6/6 correctly watching this video for the first time. #1 was easy; A had the expected structured baroque sound, while B did "innovative" (and sometimes aimless) things that no baroque composer would have ever thought of. It was immediately noticeable at 1:00, and the solo violin didn't sound quite right afterwards (little things like the shifts, micro-timing, etc.). #2 was even easier; while the quality of A wasn't good, everything was in place. B sounded choppy from the first note (2:32, particularly in the trill), with unnatural timing issues and sloppiness in the solo line throughout, combined with aimlessness (2:44) and jarring key changes (2:48). #3 was harder but became evident when comparing the two. A had a timing/direction issue at 4:04, unusual double-stops at 4:07 (they didn't resonate right), and an impossible harmonic at 4:21 (we can do that on an A, but not on a Bb). Beautifully done though, particularly the passionate singing line at 4:14! Hearing B it was instantly, yes this is human; again, everything perfectly in place. #4 was very easy-there was no structure in A, with a mishmash of random noises tossed into the soundscape. B, even though chaotic, was obviously real instruments on a real stage, playing real notes in real places. #5 was hard, but I got it right. A was aimless and nondescript, with weird tuning issues and harmonic tones/noise tainting the sound. These are not issues you'd see with professional musicians, or with sample libraries, but rather with AI convolving sounds from scratch. B sounded synthesized, but it had musical direction, and consistent defined staging of the instruments. #6 my initial impression was that A was fake due to the strong noise and seemingly broken bass line (10:35), but the clear and natural fiddle playing at 10:50 gave me reason to question that impression. As soon as B started playing, I knew it was the fake one-aimless from the get-go, overlapping violin sounds smeared randomly on the sound stage with varying numbers of voices, weird meter issue at 11:07. So... I knew AI could write music (people have been tinkering with various forms of that for years), I knew it could "hallucinate" photos from scratch based on prompts (either text or imagery), and I knew it could generate/imitate speech, but I had no idea that it had gotten to the point of inventing music at the post-production audio stage with this level of quality! As someone who uses and understands how synthesizers and samplers/sample libraries work, I am impressed with the superior quality of everything AI generated here-it far surpasses anything even the best of those can do. Which makes me wonder what AI could do if we stopped asking it to do 100% of the work, but gave it the musical notes to play, put it into instrumental boxes (i.e. gave it structure, made each instance play a single instrument into a real, reverb engine instead of hallucinating a song from scratch as a complete audio recording from scratch). Could AI harnessed in this way replace sample libraries and take MIDI realizations of musical scores to the next level? I mean, when was the last time any modern composer ever got their sample library to sound as transparent as 1:11, where I was critiquing the lack of baroque style, not the quality of the sound? Even as choppy as 2:32 is, no sample library I know comes close to this level of realism overall. And the beautiful singing, well phrased, violin solo at 4:02? The world's best sample libraries can't even do 20% of this!
“Black Angels” by George Crumb is a phenomenal piece for string quartet- and probably one of the most difficult. Imagine playing dies irae with flageolets while at the same time whistling it in syncopation…
I got all of them - I went for the ones that I either felt an immediate connection or some emotion to (they also felt like they had a consistent direction or a theme), or that didn’t sound a bit warped between notes/parts.
Love your videos, Ray! Especially fun were the faces you made when you couldn't tell between AI and human composers! Your critique of why the sound was either AI/human was spot on. I totally agree about the Bridgerton mixing, as the rhythmic part did drown out the melody! The classical examples really sounded more baroque. And the human contemporary piece was, indeed, kind of cool. P.S. I am lucky to get to see you in Concert July 7, playing my fave Tchaikovsky Violin Concerto...so excited!
This would be like judging whether AI could write fiction with samples that are only a couple of paragraphs long. The value of classical music is not something you can assess by listening to a few seconds of it.
well u actually saved me a while ago so basically in my French test the writing was about a ceremony that i attend and i met my favorite musician and to describe it and i wasn't ready for that and the first person that popped up in my mind was ray chen and i wrote about u and everything IREALLY HOPE TO GET A GREAT MARK ON THAT TEST! 🙂❤❤❤
This was fun to watch. It was especially fun to see your reaction(s) when you did not get it right, haha. Thanks for all you do to engage (especially with young people)!!
contemporary composer here. 1- the first one was so obvious to me- the AI one had no feeling, it was wandering all over the place and sounded disjunct. 2- same exact thing, even more so. 3- same thing here, but this time i even sort of recognized the human composer's (Schubert's) sound. 4- this one i actually just recognized the piece. i am a huge george crumb "black angels" fan, so how could i not? the AI piece also just sounded a little too disjunct to me. 5- the AI piece was wandery and just generally very sus. it felt like it was trying to be emotional, but didn't know anything about emotion or how it ebbs and flows. i'm not sure how to describe it, but emotion in music has a very human shape, whereas AI generated "emotional" music seems to not follow a human contour.... i also sort of recognized the human piece.. 6- the first piece sounded like something i recognized- reminded me of Lindsay sterling. the second piece also reminded me of sterling, but it sounded like a bunch of different songs on top of eachother, or like someone made an awful mixing mistake.
I got every single one. I think Ray just didn't focus on the right attributes. I was focusing mostly on the melody. It seems to me like AI is not good at melody, and it just ends up sounding like the music isn't going anywhere, there's no story. You can hear this a lot in the film music and pop ones, but also the other ones.
Haha 😂 No, you’re not in trouble. My 9-5 job is to help technology industry (and ultimately people) deploy AI. Recently, I’ve also spend some of my free time thinking about how AI can be explored in really cool ways in practising violin, and in classical music in general. I’m not a professional musician, however, I grew up playing violin +10 years, so I do have a hunch what it is about. We often hear comments on how AI will replace people. I don’t believe in it. Not even though my job is to be an advocate for AI. I truly believe that ultimately it is all about how we as humans are able to harness the capabilities of AI to help us out and create value with AI for ourselves. In other words, to identify in which tasks AI is way better and faster than humans, and vice versa, where humans beat AI 6-0. Optimally, by deploying AI we humans have more opportunities to concentrate on the tasks that we love and cherish, and leave the boring, repetitive work for AI. Unfortunately, AI will not play the scales for us, but I think it can help in amazing ways with practicing violin. AI has no imagination and it doesn’t think but it can provide - with its own strengths - new horizons and boost our imagination to the next level. Would be happy to hear about your potential trials on exploring AI to boost your practicing, in case you already have some. Thanks for the great video 👍
I'd say 80-90% of jobs are repetitive and even now there are not enough jobs for everybody... And in pop music they already chose "perfection" over emotion by tuning every single vocal, all the voices sound the same... Where there is more money to gain, they will... AI isn't bad as well as money isn't but humanity with it's greed and stupidity can't handle it in a decent way 😅 I know I sound cringy but it's like a one way road to a dead end... How can economy grow to infintiy while taking away more and more jobs from an ever growing population? Maybe we should ask AI how to rework our system because the lack of greed and any emotion would lead to a more logic solution... 🤔 (Sorry for the rant 😅)
The problem is not so much where AI is at now the problem is how rapidly is developing and all the dangerous implications that brings, a good number of people seem to be playing fast and loose with AI not really thinking of the ramifications it can bring in the future. I also find this sentence you wrote quite puzzling: "Unfortunately, AI will not play the scales for us," . How is that unfortunate ?, why would you want AI to play scales for you?.
@@ericbernardi8116 Thanks for your comments 👍 I fully get what you mean. I also think that if we only try to glue AI into our current society and assume that everything else around will stay the same, the outcome will not be very good. I would hope that AI will be disruptive enough to force us to think also more profound, positive systemic changes needed in the society. Easier said than done though 😊
@@southside8551Thanks for your comments. I agree that power always comes with responsibility and it’s vital to consider the ethical aspects of AI. This type of work is taking place in various corners of the globe as we speak. My comment about AI playing scales for us was meant more as a joke; back then I did not enjoy playing scales too much, although naturally necessary 😊
@@heliharrikari It's not purely ethical , there are a myriad of unknowns when it comes to the development of AI and the possibility and dangers of becoming actually sentient. Glad I asked about the scales rather than just assume, it's difficult to infer context on comments sometimes.
I got them all right! You were listening as a player, I was thinking more compositionally. Last two were harder to call, but I got it. Anyway, great video!
It's not that AI in itself is scary or dangerous. It's how people intend to implement it in society. Currently, the governmental regulations surrounding AI aren't well formulated or defined. But since I've been studying and experimenting with AI, I can tell you this. AI is capable of much, much more, as the years go by.
Loved this - have a suggestion for you Ray! - |Make a video where you play a number of new compositions - and make us guess which is AI and which isn't! The winner gets a free lesson from you?
Yesterday was my first time to listen to life violin. I bought a CD after that. Wonderful night. Thanks, Ray. My girlfriend and I are very enjoying. 9:46
The Shubert piece has flawless thematic development beyond anything AI is yet able to achieve. As a composition, the IA piece is correct and satisfying on the short scale, but does not develop overall into a "sum is greater than the parts" cohesive musical statement. The true Turing test here is whether or not the composition thematically converges to strengthen as a musical statement, or more simplistically flows predictably note-to-note without actually going anywhere.
Went 5-1 with this. At first, I thought AI doesn’t know how to develop chord progressions very well and can go all over the place but then it caught me off guard on the Film Score segment. Jeez, this will be tougher 5 years later.
I got wrong contemporary and film score. I can still hear the difference quite often, but damn, these quizzes are getting tougher and tougher every year.
Indeed. And remember that we're comparing music by highly successful humans. If you take the music from an average composer, could you tell the difference to best AI even today? I predict that future human composers will have trouble getting good enough to win against AI because you cannot get paid until you're better than AI and the bar is constantly raising.
@@MikkoRantalainen Lol, I wanted AI to work so I can study and do art, not the opposite. Having to create something to do the things that we enjoy to do shows how ridiculous commercialisation of art has become.
@@pierfrancescopeperoni I don't think everything is lost but it will be probably hard to make living with composing music in the future. We still have professional chess players, too, even when the best software chess programs can easily beat top players even when executed on smartphone level hardware.
I'm learning to program AI from scratch, so I'm familiar with how it works. I am also starting to compose, so I am familiar with the "flow" of the music, which gave a lot of them away. A good thing to keep in mind is that machine learning agents can mimic, but they cannot understand. An alternative example of this is if you made someone memorize a very large set of coordinates and the rgb value of the Mona Lisa. If they went to a contest of how well people knew the Mona Lisa, they would be able to construct the most accurate copy of it, but they wouldn't be able to tell you whether the right hand was on top of the left, or the left on the right.
What a wonderful video! A bit of advice from a viewer who got 5/5: you are overthinking the problem! Just go for what gives you real musical pleasure. Crazy random stuff can't do that. Thanks for this very informative presentation.
I think I got 4-2 as well as an amateur listener. I guessed correctly for the Baroque, Classical, Romantic and even the contemporary, have to say there are still lots of areas for AI to improve. But actually I couldn’t tell the difference for film score and pop, maybe since lots of them they are easy to compose and AI can handle this well, no need to bother good human composer to do that 😂😂😂.
I need to say, I am very thankful for tonic. I am very afraid of playing in front of people. Everyone till now was so sweet on tonic that it didn't make me afraid to practice piano there. Thanks Ray ❤
I think this should have been implemented by converting the human made music and AI made music into midi tracks and then playing the midi tracks with identical virtual instruments. That would have prevented instrument and recording quality affecting the comparision. Interesting video nonetheless and I guess similar round one year into the future is only going to be harder.
This was far easier than feared! 😅 I can't really explain why, because I lack the theoretical foundations, but all of those were quite clearly assignable!
Round 3 was harder, but what gave it away for me were the weird artifacts of a sound being ambiguous between solo and orchestra on a single note. The film score was definitely the most ‘mechanical’ human example, but still recognizably intentionally human. The others were fairly obvious. In fact, in all of them, the discriminating factor is a clear sense of metric organization in the human examples. The AI can’t replicate what we sense as “meter” because it only hears discrete chunks of sound and cannot organize them into a cohesive whole. The human examples also make better use of repetition as an organizing principle, the Corelli being a prime example.
I wasn’t fooled by the Baroque or the Classical pieces. I would have said that both the Romantic pieces were human. The harmonic language, the double stops, the orchestration with the counterpoint of the horns, the melodic leaps between registers, and obviously the beauty of the sound and the realness of the expressivity and of the vibrato were absolutely spot on. I’m a classical musician with a background in musicology and I totally fell for it.
I got all but the film score right. However, I was glad I was wrong on that one because I much preferred the human one since the AI didn't feel like it was going anywhere.
I got all 5 correct, but to be fair, I vaguely knew 3 of those works. I might have been fooled by the contemporary AI work, but the Dies Irae theme in Black Angels is pretty iconic.
I started to scare from AI bcz a piece or concerto composed by a wellknown composer. The thing is it took time from him or her like months or years maybe. Now AI is composing a piece in seconds which sounds so good or real like the concerto made by a human. I hope it wont effect the musicians too much
Honestly is quite surprising to see how much AI has evolved is just a matter of 2 or 3 years for us won't be able to recognize the differences between AI and humans, for now though, I think there's a little difference, over all in tonal function and rythm, it seems to me that sometimes the AI struggle to comeback to the tonic chord and do some weird things that spoils the progression, as well as when it works with complex melodies, or the melody or the acompaniment, but something gets a little bit out of rythm.
The Film Score got me goo, the human one sounded off! I was able to tell most of them because either the timing or something sounded un-natural in a way. The Lindsay Sterling was easy because I heard the song when it came out, it's in my feed - but if I hadn't heard the song, the production sounds a little AI, AI can do some cool stuff, I think it will make us better songwriters and musicians, because as a songwriter, I can write lyrics and get some inspiration from an AI demos and see if AI can make the flow of the lyrics work, I also compose music on violin and AI isn't helpful for that, when I write on violin or for ensemble it is literally emotion and AI doesn't have emotions, so it's like a computer playing Mendelsohn concerto, it would be technically correct, it doesn't have the emotion that every great player add's to the music and each great player such as yourself, adds subtle nuances that make it great. But that's also why AI is good at pop music because it's just catchy simple melodies, beep boop beep boop Look At Me! Look Look Look at Me!
Very interesting video, thank you! I think what the AI versions have in common is the fact that 1) it usually doesn't "go anywhere" musically, in other words there seems to be no real "logic" to the music nor does it "make sense" (these are very subjective notions that are difficult to explain, but probably more obvious to professional musicians) and 2) they mostly sound fake, in other words they try to replicate something using elements they already have, but do not combine these elements in a very convincing manner (the contemporary excerpt is a good example), and sometimes do weird things that a "human" composer would never do, such as the violin solo in the classical/baroque, as you pointed out (around 3:00, it immediately struck me as well). Cheers!
I think looking at periodicity/form/phrase structures wind up giving it away very easily. The AI stuff tends to wander around aimlessly, where the real pieces use musical sentences, suspension chains, regular periods, repetition/call and response, and structured development of motives. That being said the audio quality of the AI recordings is astounding and sounds very realistic
I love this video that was great and I was trying to guess along with you and I think you got more right than I did but excellent. Keep up the good work.
I wonder: If the AI generated score would be interpreted by an orchestra - would we still hear a difference? For me it was the playing more than the score itself.
I think there would. I personally had to ignore the sound because when I noticed distortion and mp3 artifacts in the recordings, those made me think for a moment it's the fake. Then when thinking about the quality of the writing, like the 'noodling around' Ray mentioned, it became very obvious which ones were created by AI when the phrases would lead to nowhere.
@@seveera That's exactly the part I am talking about, because the "noodling around" might be fixed with musicians who can elevate those loose phrases or at least lead them somewhere. And I know that is not always possible, I had one piece by an amateur human composer that simply ... could not be fixed/unnoodled so to speak.
@@blauespony1013 but I think the musicians could only fix the noodling or phrases that lead to nowhere by ignoring the integrity of the composition, then I don't think that's a fair thing to do in a comparison of what composers are able to do vs AI. It would be like having a 3 year old scribble on note paper and then have musicians just play whatever. Then going like "oh wow the 3 year old composed that??" would be really unfair I think 😸
@@seveera Not what I meant. They should not change the score. But interpretation of a piece can change so much (just listen to a piece as midi score vs. the same piece played by actual musicians).
@@blauespony1013 Oh I totally get that, but the aimless noodling around to me has nothing to do with interpretation. That would be a melody written in the score, and an interpretation in my opinion should not change that, that's kinda what I meant by integrity of the composition. Like I get some accents or if looking at the baroque example a turn could be interpreted differently, but you still wouldn't change the progression of a melody that's actually written down in a score
Thanks! It seems that AI in music generation still remains inconsistent, with difficulties in the logic of material development (in terms of intonation, rhythm, etc.)
YES! I got right answers in all rounds. Machines are better at solving difficult or logical problems, but they will never be better in music than human. They can be at the future on the almost same level as human, but not better.
Would be easier to tell if the clips were longer-LLM's (which are not "AI" but just a more advanced version of a tourist phrase book) have a huge problem with the finer aspects of composition: direction, thematic unity, lyricism and emotional character, because it doesn't actually compose, what it does is just try to predict what comes after-it's like your first year composition student who can only start writing at bar 1 and finish at the end. Compare it to some truly great works like Brandenburg, Beethoven 7, Mozart C major sonata or Symphonies of Wind Instruments and you will hear a clear difference.
Very interesting video Ray 👍 Btw the only case I can guess is the Film score. Because I’m a fan of Bridgeton and listened so much cover of this theme so I could know right away.
With good headphones well clamped on your ears you can't miss the right answer and find AI: you'll hear solo instruments weirdly jumping from one side to the other of the stereo field. This is why I almost always edit AI music by reducing the stereo opening to 50% and then recreating the wideness by adding a long yet subtle reverb that recreates the "space".
I knew the film score, because the AI one didn't sound humanely captivating at all, AI sometimes miss the musical depiction of feelings, tension or actions.
what’s scary is AI learns from humans but Mozart or Beethoven did the same every musician learnt from his predecessors. So AI produce this by studying human music. So at the end of the day , If we delete everything classical music data AI used to produce this it won’t be able. But I’m sad of listening this cuz music is soulful and could become literally shallow if we listen AI music without the background, feelings and emotions, moods of an artist while he produce his music. You can ask an AI to produce sad music it would probably make it but that sucks💔💔💔💔
Interesting test! I got them all right except for the contemporary music. I don't understand that stuff at all and both sounded equally meaningless to me. I guess "meaning" was what I was listening for, something that sounded like it meant something to the composer. The pop music was the easiest to get right. The first tune started with a rock vibe, then suddenly switched to a Celtic folk vibe that complemented the rock vibe perfectly. I don't believe any AI program could understand the concept of a musical vibe, let alone play with it creatively like that.
6 out of 6, and I thought it was pretty easy. But a couple of the A.I. tracks were really interesting, so perhaps this same test a couple years from now will be a lot tougher.
I have very unreliable understanding of Baroque and Classical but the Romantic was a little clearer, in particular the second piece's run from the high to the low register felt more nuanced and like the soloist is in control. Definitely shows how "AI" can mimic the concrete physical but not the abstract emotional.
The AI interfering everywhere already taught me to ignore too flawless paintings. It is sad, considering some of them might actually be someone's real amazing art... 😐 But I just don't feel like praising pretty things with no heart in them. I might like the way something looks (or sounds) but it's actually the big amount of work and effort and ingenuity it required, that makes me go: Wow!
Don’t know contemporary well at all but I knew and have performed all three of the “real” pieces of baroque through romantic, so recognizing the pieces kind of tipped me off pretty quickly lol
The giveaway for me with the second track was that there was a trill cadence in the violin, (normally denotes the end of a section, like in Mozart violin concertos), but AI doesn't know this so it placed this solo violin trill material in the middle of a section. AI is probably reading this comment and will learn from it. 😮
Don't let the robots win! 😆Practice and share your music with others on Tonic: tonicmusic.app/join-in
I’m a big fan of you, I use tonic too, can you follow me my tenth birthdays tomorrow pls thx if u do
Ray, could you make a video where you teach us how to play czardas like a pro? I’m currently working on it
Yeah
I am in tonic already ❤❤❤
🎉
this is SCARY as a composer…because people who we will work for might not be able to see the difference between AI and human…this is a fear becoming more valid by the month
Sad but true 😢
ai is g
Orr you can work faster by using ai and just making corrections
@@VictorIbelles yes but... That's actually editing, not composing 😐
If youre scared of AI, then you must not be that good of a composer
I was 6-0, and it came down to a few things:
1. AI doesn’t understand cadences or ends of phrases. Especially in the film and pop rounds the AI felt a real lack of musical direction and definitive « end points. » The Baroque AI also had a lot of weird out of place chords.
2. Certain notes. In the romantic AI the violin hits a B naturaly against a Bb in the orchestra melody that was super jarring and felt like something that a real composer wouldn’t do.
3. Recording artefacts: In the contemporary human recording, there was a slight echo from the hall they recorded in that the AI recording didn’t have.
There were other things too (Violin tone quality, etc.) that tipped me off but the big thing is that while the AI can really closely emulate the sound of an orchestra, they don’t yet have the knowledge of how composers actually write music, structure chord progressions, etc. To make it sound truly human.
I would say the major thing AI does not have is a multilayered idea within the music. Each peace of art has several layers which you can discover during your journey. But AI has only one layer. That’s it.
Yeah... The thing is, most of listeners don't understand any of that, either. I can only assume that this new tool will find its place in the ecosystem kind of the way synthesizers did, as another possibility to do things.
I, too, got 6/6 right, and mostly for similar reasons, even though (or maybe: because) I intentionally ignored the sound and only assessed the compositions. However, to nitpick, I both agree and respectfully disagree with your second point. I agree because this was also the exact point where I realized this was the AI one. I disagree however because first of all, it's rather a c flat over b flat than a "b natural" over "b flat", and this suspended none over a a dominant chord is actually omnipresent in most romantic composers' music. The reason, in my opinion, that it feels "out of place" here is that it does not occur in a "natural" or "organic" context: It's an element that would usually mark a point of high tension - but melodically, harmonically, and dynamically the tension is already decreasing/being released when this minor none suspension occurs: We have already been in the dominant for one entire bar (no harmonic progression), the melody reached its apex in the previous bar and reaches this suspension from above rather than below, and the dynamic goes back.
However, to circle back to your main point, I, too felt that the AI compositions clearly lacked what I would call "musical dramaturgy". It's as if, when tasked with writing something in the style of, say, Shakespeare, it would produce a series of sentences that, on their own, sounded completely Shakespearian, but, together, did not actually tell a Shakespearian story.
Yes, I have noticed similar things... It is even more evident when you feed it with an audio source that the AI can use as a starting point for generating the music. AI is very lacking especially in harmonic progressions: Its music always conveys something static as if it were assembling material that is aesthetically related but without a clear destination to head towards.
the only thing I got from your comment is that it's PhD level people like yourself who can even tell a difference anymore. I've been playing piano for 40 years, I can't tell. Ask my T Swift listening wife and she'd just say, it's easy listening classical good for reading!
What is scary to me is not that AI might surpass human music in quality or soul, but that it will be just passable enough and so easy to produce that it will be all that the majority of people hear.
This is the danger of modern music and consumerism. We've made human produced "art" so mechanical and soulless that people aren't trained to seek out authentic art
@@sivad1025 And it is like with performing music. Did anyone hear of anybody who booked a whole professional big band or choir-and-orchestra (incl. conductor) for, say, a wedding? ---
Mostly, they book a one-person-plus-midi-keyboard-band. On booking platforms, those people write that they are willing to pay 500 EUR maximum. ---
And of course you don't hear of any of those who only play already recorded music at their parties.
Then, if AI music of a certain quality will be so low cost and produced in no time, who will care about human composers, studio engineers, musicians, instrument makers etc? ---
Already, most people are happy with a low quality sound from recordings and electronic devices. A mobile phone is enough for them, they will never know how a really good, or even: excellent, hi-fi sounds like!
That's so shockingly dumb: someone walking down the streets while listening to music from a squeaking device while good bass speakers would be vital for that genre!
@birgits.3702 Totally agree!
I'm so glad you said that though. I hope to get married this year and I've been playing with the idea of investing in an orchestra accompaniment. My church occasionally has their orchestra accompany mass. For Mother's Day, they played Mozart's Coronation mass. We just need to be deliberate in supporting real music
The popular music is so bad quality that it does not matter if many listen to AI classic. Creative minds will alway stay with true human music from the past.
This is exactly the thing. Normal people do not hear these fine nuances, they literally are completely unaware of what we are discussing here. It will be cheap and easy.
I got all of them correct, the way I could tell is that AI doesn't seem to care for the beat. Sometimes it feels like it just randomly switches time signatures, or 2 different parts don't exactly match up.
This is so right.
Same. I got the same score and noticed this too.
Exactly
You mean the rhythmic pulse?
@@sarahyeoh0111 I think he means the meter (the time signature). That's what I reacted to. It will seem to be in 3/4 and then in 4/4 and then in something where you can't even really be sure what meter it is. (In the music from the periods where this was not commonly done.)
I think the difficulty in the film music round speaks more to the declining quality of modern television and film music, which tends to act simply as "sonic wallpaper".
It’s a shame since there’s some truly moving film music. Both of them just weren’t it.
Film music is declining by choice, not because of some crisis. Sadly film directors now reject the idea that film music should be significant enough to stand alone as an independent voice. They think this creates a double narrative, treating viewers as naive and disturbing them. Of course I don’t agree, and I refused to write music for films.
@@federicoaschieri I refused to marry a princess. lol
@@RichardWagner-hi4zn That's great. Indeed you should marry a girl that you love, independently of her social status 👍🏻
@@federicoaschieri Well, that's a recipe for disaster as well.
Early Classical-era DOES sound a bit more Baroque. It's technically "Galant" music - CPE Bach and early Haydn are good examples.
Late Beethoven would present a similar problem - too "out there" to represent Classical and also not quite like any Romantic-era music.
That's what I was thinking (must be Galant-style). They pulled from too early in the Classical era. Should have done something more mid-era.
Where things break down is... e.g. J.S.Bach. A large data model based on essentially age-old Markov chains does not step outside its fundamental constraints. Easy example again, ChatGPT never produced an original text that argued in new ways a certain point. It, per definitionem, can't. Move that into the sphere of music... the limitations are obvious. But clueless tabloids and then investors going for hyped BS about AI fundamentals they don't understand completely take over.
"Does it go to somewhere or not" - this was a good recipe, helping me to get all. For my ears, the important difference is a meaningful composition vs. a senseless composition - more than the "sound" of the audio.
To add here: the selection was REALLY GREAT! It allowed the composers to shine, we heard really hidden masterpieces in all genres. Great-great selection, congratulations to your team, Ray!
The AI lines meander and felt directionless, while the composers' had a more clear sense of resolution.
How many did you get right? I used the same judgement and got 1/2...where is it going?
I went 5-0, "guessed" correctly. I don't think I can say my assessment is better than Ray's, but it did came down to "feeling". The 3rd one, the 2nd romantic felt well structured, human touch indeed. And the last one, the 2nd song felt uplifting and a bit more distinguinshable.
Agreed, the AI generated music will play the right notes in their own independent bits, but it doesn't feel like there is a 'purpose'/'idea' to it, which makes sense... since there isn't, it's just a generative algorithm. Just like pretty much all AI generated stuff, it seriously needs an editor. Also I think pop music could be the hardest one by far. The tracks are generic enough already
Ah, but there are 6 rounds ^^
Yeah, Ray seemed to be more focused on the technical aspect of the playing, whereas I thought most of these were pretty obvious from the composition side. The AI tracks can never put together a cohesive melody.
@@Tweeteketje I think the commenter meant the 0 as the one they did not guess right. Definitely the human touch! I guess they pass as not AI for the time being.
@@Zareh_Abrahamian yes, so it should be 6-0 ☺️
Got them all. I was applying the same rules I use when either reading or listening to verbal content. AII has weird repetitions. Inconsistent "pronunciations" or throwbacks to something earlier in the content in a way that just doesn't fit - that's why a human editor is needed to polish the content.
I had to laugh about your reaction to the Bridgerton theme.. I'm sure I will get lots of negative comments here, but that's the wsy I feel about the whole series - it is artificial and tries too hard to be accurate to/reproduce something it isn't. It's very funny that is reflected in the scoring as well.
Hi Ray Chen! It’s FamousePianist from Tonic.
I know you probably won’t see this message, but I have to say I love your playing and I am so thankful for learning about Tonic. Tonic motivates me to practice and I really enjoy the opportunity to listen to you and other professional musicians practice. ❤ Thank you so much! 😊
I think it is quite easy to understand the difference. Especially in old classical music where music tends to be derived from speech and tends to have a rhethorical speaking quality such ML based systems cannot reproduce. With contemporary music it is harder, as there often you need to get the big picture to understand the design. The systems behind this are good at reproducing the generic aspects of a style, which also explains why it works so well for pop music ...
I hope someday you can go to México, you are a big inspiration for the people ! I love your interpretations
*But who knows ai is secretly checking this video for information where it lacks?* 😅
AI sucks up whatever is posted on the internet and there's no escape for humans.
oh no
uh oh...
we're screwed
It can check all it wants, it will never have understanding.
@@baldrbraa HA...We (never) expected this also...The fact that classical music is rule based opens a wide window..Wait and see!
I'm an opera singer - my performance repertoire is over 50 arias composed by Mozart, Bellini, Puccini, Rossini, Verdi, Vivaldi and dozens of other composers - I sing in ten languages - and like Ray, I can distinguish between the music composed by robots vs. the music composed by humans - but probably soon, this distinction will no longer exist.
Unfortunately doesn’t matter if we can distinguish. Can the audience distinguish?
@@halohack789 Pretty soon, the AI will be so brilliant that audiences will have no interest in the music composed by humans
It will. AI has no attitude, stance, viewpoint, personal history, passion, soul. Competently generating some snippets of music in a given style and genre is not equivalent or surpassing a human composing a C Minor Mass. The Illiac Suite is from the 1950s. It was rudimentary AI with astonishing end result. We are now 70-ish years later. Has music died? Has it been overtaken by computers since then? A lot of the panic and utter nonsense surrounding current AI misses the fundamental, central aspect of it.
Please share w/ us your brilliant accomplishments and achievements in life - so we can all appreciate your brilliance - if you have no impressive achievements ... let that be a reminder to you that you're not very bright ... Then maybe you'll refrain from presenting yourself as an expert on a topic when your awareness is adolescent - at best.
@@WhirledPublishing Meaning?... how does this relate to a specific aspect of AI development? Apart from it being a logical fallacy (see ad hominem). Plus... how is this a response to a discussion on how AI cannot take over human creativity for quite some years?
The most inportant thing to remember about AI is that at any given moment, this is the *worst* AI will ever be.
This was instilled in me at a talk I went to like 6 years ago, well before AI started to make its current dramatic progress. The speaker noted how important authenticity and creativity will become in the future. At some point, AI will be good enough to be indistinguishable even to the majority of experts. Thats why we need to value authenticity and live performance in and of itself.
I don't want AI generated music .... I want humans to play more music. Music is about connecting to each other i don't want to connect to a machine.
You d never even know
As wierd as that is, good AI will have been effectively pruned and selected from a large pool of humans judges so that, despite its synthetic origins, it would nonetheless deliver a human-biased performance. Think of it as ready-made pizza vs freshly, homely pizza. Chances are that, in a casual context, you'd stop telling the difference between the two.
Stfu and just listen to good music no matter the source. Just enjoy it and stop bitching about it.
Violinist/composer/arranger/sound engineer here, I guessed 6/6 correctly watching this video for the first time.
#1 was easy; A had the expected structured baroque sound, while B did "innovative" (and sometimes aimless) things that no baroque composer would have ever thought of. It was immediately noticeable at 1:00, and the solo violin didn't sound quite right afterwards (little things like the shifts, micro-timing, etc.).
#2 was even easier; while the quality of A wasn't good, everything was in place. B sounded choppy from the first note (2:32, particularly in the trill), with unnatural timing issues and sloppiness in the solo line throughout, combined with aimlessness (2:44) and jarring key changes (2:48).
#3 was harder but became evident when comparing the two. A had a timing/direction issue at 4:04, unusual double-stops at 4:07 (they didn't resonate right), and an impossible harmonic at 4:21 (we can do that on an A, but not on a Bb). Beautifully done though, particularly the passionate singing line at 4:14! Hearing B it was instantly, yes this is human; again, everything perfectly in place.
#4 was very easy-there was no structure in A, with a mishmash of random noises tossed into the soundscape. B, even though chaotic, was obviously real instruments on a real stage, playing real notes in real places.
#5 was hard, but I got it right. A was aimless and nondescript, with weird tuning issues and harmonic tones/noise tainting the sound. These are not issues you'd see with professional musicians, or with sample libraries, but rather with AI convolving sounds from scratch. B sounded synthesized, but it had musical direction, and consistent defined staging of the instruments.
#6 my initial impression was that A was fake due to the strong noise and seemingly broken bass line (10:35), but the clear and natural fiddle playing at 10:50 gave me reason to question that impression. As soon as B started playing, I knew it was the fake one-aimless from the get-go, overlapping violin sounds smeared randomly on the sound stage with varying numbers of voices, weird meter issue at 11:07.
So... I knew AI could write music (people have been tinkering with various forms of that for years), I knew it could "hallucinate" photos from scratch based on prompts (either text or imagery), and I knew it could generate/imitate speech, but I had no idea that it had gotten to the point of inventing music at the post-production audio stage with this level of quality! As someone who uses and understands how synthesizers and samplers/sample libraries work, I am impressed with the superior quality of everything AI generated here-it far surpasses anything even the best of those can do. Which makes me wonder what AI could do if we stopped asking it to do 100% of the work, but gave it the musical notes to play, put it into instrumental boxes (i.e. gave it structure, made each instance play a single instrument into a real, reverb engine instead of hallucinating a song from scratch as a complete audio recording from scratch).
Could AI harnessed in this way replace sample libraries and take MIDI realizations of musical scores to the next level? I mean, when was the last time any modern composer ever got their sample library to sound as transparent as 1:11, where I was critiquing the lack of baroque style, not the quality of the sound? Even as choppy as 2:32 is, no sample library I know comes close to this level of realism overall. And the beautiful singing, well phrased, violin solo at 4:02? The world's best sample libraries can't even do 20% of this!
“Black Angels” by George Crumb is a phenomenal piece for string quartet- and probably one of the most difficult. Imagine playing dies irae with flageolets while at the same time whistling it in syncopation…
That's really a great piece - and the God-music movement reminds me of Messiaen somehow.
@@aveyenxharmonically, it is indeed a Messiaen pastiche.
2:19 Sounds extremely baroque to me. I'm somewhat of a professional. I've been baroque for several years now
I got all of them - I went for the ones that I either felt an immediate connection or some emotion to (they also felt like they had a consistent direction or a theme), or that didn’t sound a bit warped between notes/parts.
2/3 for me the AI ones lack coherence, direction, like a vague impression of the styles
I got most of these right! But it’s ridiculous how good the AI is. Please let us know which AI you were using for this demonstration.
Love your videos, Ray! Especially fun were the faces you made when you couldn't tell between AI and human composers! Your critique of why the sound was either AI/human was spot on. I totally agree about the Bridgerton mixing, as the rhythmic part did drown out the melody! The classical examples really sounded more baroque. And the human contemporary piece was, indeed, kind of cool. P.S. I am lucky to get to see you in Concert July 7, playing my fave Tchaikovsky Violin Concerto...so excited!
This would be like judging whether AI could write fiction with samples that are only a couple of paragraphs long. The value of classical music is not something you can assess by listening to a few seconds of it.
My smile was so big when Roundtable Rival played! Such an amazing piece
well u actually saved me a while ago so basically in my French test the writing was about a ceremony that i attend and i met my favorite musician and to describe it and i wasn't ready for that and the first person that popped up in my mind was ray chen and i wrote about u and everything IREALLY HOPE TO GET A GREAT MARK ON THAT TEST! 🙂❤❤❤
J'espère que t'auras une bonne note 🤣
@@Ketsen merci beaucoup 🙏💕
This is so entertaining. Love the live analysis streams of consciousness.
I’m going to your concert in 6/29! Always thank you for making these videos and doing concerts❤
This was fun to watch. It was especially fun to see your reaction(s) when you did not get it right, haha. Thanks for all you do to engage (especially with young people)!!
contemporary composer here.
1- the first one was so obvious to me- the AI one had no feeling, it was wandering all over the place and sounded disjunct.
2- same exact thing, even more so.
3- same thing here, but this time i even sort of recognized the human composer's (Schubert's) sound.
4- this one i actually just recognized the piece. i am a huge george crumb "black angels" fan, so how could i not? the AI piece also just sounded a little too disjunct to me.
5- the AI piece was wandery and just generally very sus. it felt like it was trying to be emotional, but didn't know anything about emotion or how it ebbs and flows. i'm not sure how to describe it, but emotion in music has a very human shape, whereas AI generated "emotional" music seems to not follow a human contour.... i also sort of recognized the human piece..
6- the first piece sounded like something i recognized- reminded me of Lindsay sterling. the second piece also reminded me of sterling, but it sounded like a bunch of different songs on top of eachother, or like someone made an awful mixing mistake.
That was cool Ray! Nice editing too 😎 I was also fooled by the same ones you were. Cool format! Looking forward to seeing more ✌️
I got every single one. I think Ray just didn't focus on the right attributes.
I was focusing mostly on the melody. It seems to me like AI is not good at melody, and it just ends up sounding like the music isn't going anywhere, there's no story.
You can hear this a lot in the film music and pop ones, but also the other ones.
I love this video Tay. Well done, such fun. -Will
Haha 😂 No, you’re not in trouble.
My 9-5 job is to help technology industry (and ultimately people) deploy AI. Recently, I’ve also spend some of my free time thinking about how AI can be explored in really cool ways in practising violin, and in classical music in general. I’m not a professional musician, however, I grew up playing violin +10 years, so I do have a hunch what it is about.
We often hear comments on how AI will replace people. I don’t believe in it. Not even though my job is to be an advocate for AI. I truly believe that ultimately it is all about how we as humans are able to harness the capabilities of AI to help us out and create value with AI for ourselves. In other words, to identify in which tasks AI is way better and faster than humans, and vice versa, where humans beat AI 6-0. Optimally, by deploying AI we humans have more opportunities to concentrate on the tasks that we love and cherish, and leave the boring, repetitive work for AI.
Unfortunately, AI will not play the scales for us, but I think it can help in amazing ways with practicing violin. AI has no imagination and it doesn’t think but it can provide - with its own strengths - new horizons and boost our imagination to the next level. Would be happy to hear about your potential trials on exploring AI to boost your practicing, in case you already have some.
Thanks for the great video 👍
I'd say 80-90% of jobs are repetitive and even now there are not enough jobs for everybody... And in pop music they already chose "perfection" over emotion by tuning every single vocal, all the voices sound the same... Where there is more money to gain, they will... AI isn't bad as well as money isn't but humanity with it's greed and stupidity can't handle it in a decent way 😅
I know I sound cringy but it's like a one way road to a dead end... How can economy grow to infintiy while taking away more and more jobs from an ever growing population? Maybe we should ask AI how to rework our system because the lack of greed and any emotion would lead to a more logic solution... 🤔
(Sorry for the rant 😅)
The problem is not so much where AI is at now the problem is how rapidly is developing and all the dangerous implications that brings, a good number of people seem to be playing fast and loose with AI not really thinking of the ramifications it can bring in the future. I also find this sentence you wrote quite puzzling: "Unfortunately, AI will not play the scales for us," . How is that unfortunate ?, why would you want AI to play scales for you?.
@@ericbernardi8116 Thanks for your comments 👍 I fully get what you mean. I also think that if we only try to glue AI into our current society and assume that everything else around will stay the same, the outcome will not be very good. I would hope that AI will be disruptive enough to force us to think also more profound, positive systemic changes needed in the society. Easier said than done though 😊
@@southside8551Thanks for your comments. I agree that power always comes with responsibility and it’s vital to consider the ethical aspects of AI. This type of work is taking place in various corners of the globe as we speak. My comment about AI playing scales for us was meant more as a joke; back then I did not enjoy playing scales too much, although naturally necessary 😊
@@heliharrikari It's not purely ethical , there are a myriad of unknowns when it comes to the development of AI and the possibility and dangers of becoming actually sentient. Glad I asked about the scales rather than just assume, it's difficult to infer context on comments sometimes.
I got them all right! You were listening as a player, I was thinking more compositionally. Last two were harder to call, but I got it. Anyway, great video!
The romantic one I'm pretty sure ended with a slide up to a Bb harmonic, a harmonic that doesn't exist naturally...
It's not that AI in itself is scary or dangerous. It's how people intend to implement it in society. Currently, the governmental regulations surrounding AI aren't well formulated or defined. But since I've been studying and experimenting with AI, I can tell you this. AI is capable of much, much more, as the years go by.
This is so much fun. Having said that I wanna listen to each piece separately especially the Baroque ones, Haydn, and Crumb.
Loved this - have a suggestion for you Ray! - |Make a video where you play a number of new compositions - and make us guess which is AI and which isn't! The winner gets a free lesson from you?
Yesterday was my first time to listen to life violin. I bought a CD after that. Wonderful night. Thanks, Ray. My girlfriend and I are very enjoying. 9:46
The Shubert piece has flawless thematic development beyond anything AI is yet able to achieve. As a composition, the IA piece is correct and satisfying on the short scale, but does not develop overall into a "sum is greater than the parts" cohesive musical statement. The true Turing test here is whether or not the composition thematically converges to strengthen as a musical statement, or more simplistically flows predictably note-to-note without actually going anywhere.
Thoroughly enjoyed this video, Ray! You are the best! 🎉
Went 5-1 with this. At first, I thought AI doesn’t know how to develop chord progressions very well and can go all over the place but then it caught me off guard on the Film Score segment. Jeez, this will be tougher 5 years later.
I got wrong contemporary and film score. I can still hear the difference quite often, but damn, these quizzes are getting tougher and tougher every year.
Indeed. And remember that we're comparing music by highly successful humans. If you take the music from an average composer, could you tell the difference to best AI even today?
I predict that future human composers will have trouble getting good enough to win against AI because you cannot get paid until you're better than AI and the bar is constantly raising.
@@MikkoRantalainen Lol, I wanted AI to work so I can study and do art, not the opposite. Having to create something to do the things that we enjoy to do shows how ridiculous commercialisation of art has become.
@@pierfrancescopeperoni I don't think everything is lost but it will be probably hard to make living with composing music in the future. We still have professional chess players, too, even when the best software chess programs can easily beat top players even when executed on smartphone level hardware.
"Stuck in the kitchen of a hotel" is a great title for a song actually!
I think we need to revisit this in 12 months. And as for 2030, god help us all.
I'm learning to program AI from scratch, so I'm familiar with how it works. I am also starting to compose, so I am familiar with the "flow" of the music, which gave a lot of them away. A good thing to keep in mind is that machine learning agents can mimic, but they cannot understand. An alternative example of this is if you made someone memorize a very large set of coordinates and the rgb value of the Mona Lisa. If they went to a contest of how well people knew the Mona Lisa, they would be able to construct the most accurate copy of it, but they wouldn't be able to tell you whether the right hand was on top of the left, or the left on the right.
You are so funny, keep these videos coming please. Thank you.
I got the romantic one right. The movement of the notes and the placing of the cello accompaniment around felt balanced and responsive.
The romantic A.I is SO convincing, and it sounds like Chopin.
Okay, but that AI romantic era piece was kinda fire... any chance of sharing where to find the tracks??
Thanks for the exciting experiment!
What a wonderful video! A bit of advice from a viewer who got 5/5: you are overthinking the problem! Just go for what gives you real musical pleasure. Crazy random stuff can't do that. Thanks for this very informative presentation.
I'm an amateur composer, and I only missed the contemporary one lol, that sounded like pretty authentic contemporary music to me!
At 7:50 I realized that it is by a human composer because the melody of Dies Irae plays
Corelli's 12 Concerti Grossi are well known. As a violinist, you must know or at least heard them before.
I think I got 4-2 as well as an amateur listener. I guessed correctly for the Baroque, Classical, Romantic and even the contemporary, have to say there are still lots of areas for AI to improve. But actually I couldn’t tell the difference for film score and pop, maybe since lots of them they are easy to compose and AI can handle this well, no need to bother good human composer to do that 😂😂😂.
I need to say, I am very thankful for tonic. I am very afraid of playing in front of people. Everyone till now was so sweet on tonic that it didn't make me afraid to practice piano there. Thanks Ray ❤
I think this should have been implemented by converting the human made music and AI made music into midi tracks and then playing the midi tracks with identical virtual instruments. That would have prevented instrument and recording quality affecting the comparision.
Interesting video nonetheless and I guess similar round one year into the future is only going to be harder.
For round 3, the second one pulled on my heartstrings a lot more than the first one
Ray...I'm impressed by your ears...instant transcribe and playback...are there jazz gigs in your future?
This was far easier than feared! 😅 I can't really explain why, because I lack the theoretical foundations, but all of those were quite clearly assignable!
Round 3 was harder, but what gave it away for me were the weird artifacts of a sound being ambiguous between solo and orchestra on a single note. The film score was definitely the most ‘mechanical’ human example, but still recognizably intentionally human. The others were fairly obvious.
In fact, in all of them, the discriminating factor is a clear sense of metric organization in the human examples. The AI can’t replicate what we sense as “meter” because it only hears discrete chunks of sound and cannot organize them into a cohesive whole. The human examples also make better use of repetition as an organizing principle, the Corelli being a prime example.
I wasn’t fooled by the Baroque or the Classical pieces. I would have said that both the Romantic pieces were human. The harmonic language, the double stops, the orchestration with the counterpoint of the horns, the melodic leaps between registers, and obviously the beauty of the sound and the realness of the expressivity and of the vibrato were absolutely spot on. I’m a classical musician with a background in musicology and I totally fell for it.
I got all but the film score right. However, I was glad I was wrong on that one because I much preferred the human one since the AI didn't feel like it was going anywhere.
Romantic one - I straight away knew it was Schubert because of its sytle.
I find it easy to tell the difference but I’m blown away regardless.
There is a very distinct 'warbling' effect in the high frequencies on all the AI tracks
I got all 5 correct, but to be fair, I vaguely knew 3 of those works. I might have been fooled by the contemporary AI work, but the Dies Irae theme in Black Angels is pretty iconic.
George Crumb's "Black Angels" is a masterpiece. I recommend you familiarize yourself with it tout de suite.
6:15
bro, the second one was too beautiful to be AI
Yeah it's the only piece that gave me chills
I started to scare from AI bcz a piece or concerto composed by a wellknown composer. The thing is it took time from him or her like months or years maybe. Now AI is composing a piece in seconds which sounds so good or real like the concerto made by a human. I hope it wont effect the musicians too much
Honestly is quite surprising to see how much AI has evolved is just a matter of 2 or 3 years for us won't be able to recognize the differences between AI and humans, for now though, I think there's a little difference, over all in tonal function and rythm, it seems to me that sometimes the AI struggle to comeback to the tonic chord and do some weird things that spoils the progression, as well as when it works with complex melodies, or the melody or the acompaniment, but something gets a little bit out of rythm.
The Film Score got me goo, the human one sounded off! I was able to tell most of them because either the timing or something sounded un-natural in a way. The Lindsay Sterling was easy because I heard the song when it came out, it's in my feed - but if I hadn't heard the song, the production sounds a little AI, AI can do some cool stuff, I think it will make us better songwriters and musicians, because as a songwriter, I can write lyrics and get some inspiration from an AI demos and see if AI can make the flow of the lyrics work,
I also compose music on violin and AI isn't helpful for that, when I write on violin or for ensemble it is literally emotion and AI doesn't have emotions, so it's like a computer playing Mendelsohn concerto, it would be technically correct, it doesn't have the emotion that every great player add's to the music and each great player such as yourself, adds subtle nuances that make it great. But that's also why AI is good at pop music because it's just catchy simple melodies, beep boop beep boop Look At Me! Look Look Look at Me!
oh gosh , i need to practice on tonic more often
0:53 Wow that’s really good. I wish that was a real piece and look at rays face! 0:53
Very interesting video, thank you! I think what the AI versions have in common is the fact that 1) it usually doesn't "go anywhere" musically, in other words there seems to be no real "logic" to the music nor does it "make sense" (these are very subjective notions that are difficult to explain, but probably more obvious to professional musicians) and 2) they mostly sound fake, in other words they try to replicate something using elements they already have, but do not combine these elements in a very convincing manner (the contemporary excerpt is a good example), and sometimes do weird things that a "human" composer would never do, such as the violin solo in the classical/baroque, as you pointed out (around 3:00, it immediately struck me as well). Cheers!
Baroque
1. 0:31
2. 0:50
a. 1:36
Classical
1. 2:04
2. 2:32
a. 3:38
Romantic
1. 4:35
2. 5:16
a. 6:08
Contemporary
1. 7:17
2. 7:45
a. 8:20
Film Score
1. 8:34
2. 9:12
a. 10:03
The violin of the Romantic piece #1 has a problem with the stereofield 🧐
I think looking at periodicity/form/phrase structures wind up giving it away very easily. The AI stuff tends to wander around aimlessly, where the real pieces use musical sentences, suspension chains, regular periods, repetition/call and response, and structured development of motives.
That being said the audio quality of the AI recordings is astounding and sounds very realistic
I was half right. I agree with you on most of them.. the one I was wrong on was the contemporary one. I thought the first one sounded like Bartok.
Yay! Got them all right! The romantic one just felt familiar - the AI was strange, even for a less known composer - just not right.
It doesn't disappoint. It's an unexpected topic, and I'm intrigued.👍🏻👍🏻✨
I love this video that was great and I was trying to guess along with you and I think you got more right than I did but excellent. Keep up the good work.
I wonder: If the AI generated score would be interpreted by an orchestra - would we still hear a difference? For me it was the playing more than the score itself.
I think there would. I personally had to ignore the sound because when I noticed distortion and mp3 artifacts in the recordings, those made me think for a moment it's the fake. Then when thinking about the quality of the writing, like the 'noodling around' Ray mentioned, it became very obvious which ones were created by AI when the phrases would lead to nowhere.
@@seveera That's exactly the part I am talking about, because the "noodling around" might be fixed with musicians who can elevate those loose phrases or at least lead them somewhere.
And I know that is not always possible, I had one piece by an amateur human composer that simply ... could not be fixed/unnoodled so to speak.
@@blauespony1013 but I think the musicians could only fix the noodling or phrases that lead to nowhere by ignoring the integrity of the composition, then I don't think that's a fair thing to do in a comparison of what composers are able to do vs AI. It would be like having a 3 year old scribble on note paper and then have musicians just play whatever. Then going like "oh wow the 3 year old composed that??" would be really unfair I think 😸
@@seveera Not what I meant. They should not change the score. But interpretation of a piece can change so much (just listen to a piece as midi score vs. the same piece played by actual musicians).
@@blauespony1013 Oh I totally get that, but the aimless noodling around to me has nothing to do with interpretation. That would be a melody written in the score, and an interpretation in my opinion should not change that, that's kinda what I meant by integrity of the composition. Like I get some accents or if looking at the baroque example a turn could be interpreted differently, but you still wouldn't change the progression of a melody that's actually written down in a score
Thanks! It seems that AI in music generation still remains inconsistent, with difficulties in the logic of material development (in terms of intonation, rhythm, etc.)
YES! I got right answers in all rounds. Machines are better at solving difficult or logical problems, but they will never be better in music than human. They can be at the future on the almost same level as human, but not better.
Would be easier to tell if the clips were longer-LLM's (which are not "AI" but just a more advanced version of a tourist phrase book) have a huge problem with the finer aspects of composition: direction, thematic unity, lyricism and emotional character, because it doesn't actually compose, what it does is just try to predict what comes after-it's like your first year composition student who can only start writing at bar 1 and finish at the end. Compare it to some truly great works like Brandenburg, Beethoven 7, Mozart C major sonata or Symphonies of Wind Instruments and you will hear a clear difference.
Very interesting video Ray 👍 Btw the only case I can guess is the Film score. Because I’m a fan of Bridgeton and listened so much cover of this theme so I could know right away.
With good headphones well clamped on your ears you can't miss the right answer and find AI: you'll hear solo instruments weirdly jumping from one side to the other of the stereo field. This is why I almost always edit AI music by reducing the stereo opening to 50% and then recreating the wideness by adding a long yet subtle reverb that recreates the "space".
I knew the film score, because the AI one didn't sound humanely captivating at all, AI sometimes miss the musical depiction of feelings, tension or actions.
what’s scary is AI learns from humans but Mozart or Beethoven did the same every musician learnt from his predecessors. So AI produce this by studying human music. So at the end of the day , If we delete everything classical music data AI used to produce this it won’t be able. But I’m sad of listening this cuz music is soulful and could become literally shallow if we listen AI music without the background, feelings and emotions, moods of an artist while he produce his music. You can ask an AI to produce sad music it would probably make it but that sucks💔💔💔💔
Ray, I don’t know if you read the comments, but can you please do a how to play czardas like a pro(with octaves) Also, I’m loving the Tonic app❤❤
Interesting test! I got them all right except for the contemporary music. I don't understand that stuff at all and both sounded equally meaningless to me. I guess "meaning" was what I was listening for, something that sounded like it meant something to the composer. The pop music was the easiest to get right. The first tune started with a rock vibe, then suddenly switched to a Celtic folk vibe that complemented the rock vibe perfectly. I don't believe any AI program could understand the concept of a musical vibe, let alone play with it creatively like that.
6 out of 6, and I thought it was pretty easy. But a couple of the A.I. tracks were really interesting, so perhaps this same test a couple years from now will be a lot tougher.
I have very unreliable understanding of Baroque and Classical but the Romantic was a little clearer, in particular the second piece's run from the high to the low register felt more nuanced and like the soloist is in control. Definitely shows how "AI" can mimic the concrete physical but not the abstract emotional.
The AI interfering everywhere already taught me to ignore too flawless paintings.
It is sad, considering some of them might actually be someone's real amazing art... 😐
But I just don't feel like praising pretty things with no heart in them.
I might like the way something looks (or sounds) but it's actually the big amount of work and effort and ingenuity it required, that makes me go: Wow!
AI vidual art is very far from flawless!
I have been playing violin for 11 months now and im 12. I was in a beginner orchastra for school and now my teacher moved me up to intermediate.
That's really great to hear! Teach wouldn't have done it if they didn't think it was worth doing. It will just keep getting better if you keep going!
Don’t know contemporary well at all but I knew and have performed all three of the “real” pieces of baroque through romantic, so recognizing the pieces kind of tipped me off pretty quickly lol
The giveaway for me with the second track was that there was a trill cadence in the violin, (normally denotes the end of a section, like in Mozart violin concertos), but AI doesn't know this so it placed this solo violin trill material in the middle of a section. AI is probably reading this comment and will learn from it. 😮