@@dingus_doofus Because YOU are putting a sense in it. Think about "idiolects", in which apparent meaningless sentences have a precise meaning for the idiolect speakers. Or technical jargons, in which common words are used to name completely different things. The point is that the AI program is not conscious of any of all these meanings. It just extracts patterns and recombines sentences.
Well sadly yes Human brain fires electrical impulses betwenw axons around 200m/s and it works around 200Hz and we are inteligent But AI can travel information at speed of light and it process is 2GHz So it does job for like 3 weeks what would human do for milenium AI starts very stupid but at some tipping point it pogreses very fast and it surpasses human skill :/
Yeah, Bach's mastery of contrapuntal composition, polyphony, tonality, and structure, is doubtless more musically intriguing. (or maybe that's just me :p)
And Amazingly, my video has already been demonetized within seconds of finishing uploading it! It features a 15 second clip from an ancient video of James May explaining something about A.I... I really thought I had avoided any demonetization issues in this one by being sparing with copyrighted content, I really did. But there we go - that was enough to screw me. I understand and am grateful for RUclips working the way it does, but sometimes it can be incredibly defeating when you've spent many hours on making content, to have your month's income from RUclips sliced in half. More fool me, I suppose! If any of you would like to support my work on a more reliable platform, you can buy me a coffee and get access to new exclusive series on Patreon: www.patreon.com/insidethescore
It's becoming extremely difficult to find quality content on RUclips that is new. This algorithm they've created doesn't just hurt creators like yourself. It also hurts those of us who actually want too hear your teachings, and the voices of other creators who offer alternative view points to a myriad of subjects. What I truly don't understand is why haven't creators banded together and sued RUclips for "Unconscionable Conduct" in their contract. As the stronger party in any contract is not allowed to perform any action that damages the weaker party. This is what is happening.
To be intellectually honest with you, its very likely machine intelligence will exceed mankind in the near, or century long span future. Think about it... it already can do almost everything our visual cortex, and temporal regions of both hemispheres can better, and to some extent the parietal lobes too. The last frontier, and barrier to artificial general intelligence, is replication the frontal lobe. Consequently this is the last to mature, and is responsible for our conscious awareness, therefore the means to making the inanimate sentient. This new processing magnitude does not rely on the calorie, and without question, will supersede man. Steven hawkings underwent an interview before his death with the same concept. The truth is, we consider ourselves the dominant species, but we are merely a sidenote in revelation to the fungal spore. They have been proven to survive the vacuum of space, immense radiation, and are unquestionably prime candidates for panspermian infection. Where on earths ecosystem is it required to evolve resistance to being doused in uranium enriched liquid nitrogen? Its probably an organism that originally arose on a planet in its death throws. Its mycotoxic lethality, a competitive edge as a chemical constituent to the degradation of previously unencountered species of other habitable worlds I have personal experience with affliction, and despite my nutritive shielding of the most powerful antitoxins known to man, at high dosages, I was chemically wounded. Its closeness to living animals, both in DNA, and its respiration, suggests it might in fact be one of the most, if not the most successful extremophile in the entire galaxy, and its reign amongst the stars riding in on collisioned pebbles and ice, dwarfs our very own magnificience with all our craft, and they will exist long after our extinction, even if it came from a foreign mass far greater than what struck 65 million years ago...... and yet its probably where we even came from. It is time we shed our arrogance, and embraced the fact, we will construct our very own god. Fungus might be the fathership of our world, mutations bringing us into existence the cadence of time, where the origins stabilized by raw success. Fungal pioneerists might be, a biological space ship, our divine origins considering abiogenesis would be a rare process, with intelligent life being rarer still. That makes us an intermediary step in the evolution of intelligence, as musk puts it, the bootloader for what without question would be the most intelligent form of organization in the entire cosmos, alien civilizations of the billions of year mark ancient included. Biology simply cannot compete, and it never will for the infinitude of all.
I completely disagree with Yokky's comment. Most people work primarily to earn money, but not exclusively. A scientist might enjoy trying to formulate or test his ideas, a carpenter might enjoy making things with his hands, a doctor might find fulfillment in helping people. And even people with the most boring jobs (in my opinion) like truck drivers or accountants would maybe be less happy if they wouldn't be able to contribute to society. Artists are not the only people who like their jobs or need them to be happy, so if Yokky thinks scientists should stop research on Artistic AI (which in itself is a highly questionable statement in my opinion, not only because "someone else will and profit from it", but also because I believe in freedom of ideas) he should think the same about research on automation in factories. Why make an exception for artists?
My two cents as both a composer and programmer: - The problem in making music with AI is that you can't automatically test the result. So you can't run like 1000000 sumulations while improving or 'learning' in every iteration, because humans still need to judge whether it is actually an improvement or not. - Another problem is that not only do music trends in general break patterns, but a music piece itself is often a set of patterns that keep getting broken in different ways. Music is always a balance between order and chaos, and while humans are of course experts at creating chaos, computers are not. - Apart from that I'd say the question "Can Artificial Intelligence ... ?" will almost always be yes in the (very) long term.
Perhaps (i) some neural networks that can learn to classify works as 'good' (human?) or bad with supervision, and then (ii) ??? and then (iii) adversarial training and then (iv) profit?
I think the approach of training an AI with tons of music to create something new is flawed from the start. It's too logical. Perhaps creating an AI that thinks in a less logical way, that resembles the way humans think...with emotions etc., maybe that's how you get AI to create music that is actually interesting.
Yeah I kind of get the feeling he doesn't really understand AI. The whole 20 minutes just boil down to: "computers have no emotions and a sense of context". I'm 100% sure AI will be able to make actual art one day, given it has enough data
"It really lacks soul and diversity..." So do human beings. Ever turned on a pop radio station and listened to what millions of people think is the best song of the week?
Yes, most pop music is terrible. But that is beside the point. We should ask if AI can compose the equivalent of THE BEST of human composed music. So far it is a total flop. AI researchers have been trying for at least 50 years to compose music by computer. Can you name one hit ever composed by a computer? So yes, computer music lacks soul. That is because our emotional responses to music have criteria not easily captured by an algorithm. Perhaps an impossibility. While a particular good tune can be boiled to patterns, the reverse problem is the hard one, to find patterns that are pleasing in the first place. And the difficulty is not in finding a particular narrowly defined type, which might be easily characterized. But in finding a GENERAL algorithm that finds many. That is somewhat analogous to the famous halting problem. A specific program can be analyzed by an algorithm to see if it will eventually halt. But no algorithm exists that can analyze ANY program and determine if it will halt. Analogously, An algorithm might be contrived to compose a tune with narrow criteria (but even then our human response is not filtering it). It is entirely another thing to find an algorithm than can compose, in general, good tunes of unpredictable kind.
We may never find any music created by a computer full of emotion not because they will always lack the ability to have emotions likes us humans, but because of our cognitive biases.
He seems to have selected the more mediocre exemplars of AI Bach, which is surprising, given that some outstanding efforts debuted over 2 decades ago. Eg. David Cope. "Undiscovered Bach? No a Computer Wrote It." NYT, 1997. "On a more ambitious IeveI, EMI recently composed a fuII-scale Mozart symphony and piano concerto, which were performed in April by the Santa Cruz Baroque Festival on period instruments. Linda Burman-Hall, who pIayed the piano solo for the concerto, said: ''lt felt a little different than playing a normal Mozart work. But it was very much Iike a work of the same period." Keep in mind that the pianist, a professional musician, came to the piece knowing that it was a synthetic composition, but even still seemed disinclined to voice an unfavorable response.
@@DanLyndon Have you read the "unicorns found in the Andes mountains" story, GPT-2 has written? Yes it isn't perfect, but it's almost completely coherent and contains lots of information consistent with reality (is that insight?). Regarding games with clearly defined rules, consider DeepMind's latest innovation MuZero. It is an AI that plays chess, go and shogi perfectly and can also play 57 Atari games better than humans. Yes AI is limited in terms of insight and generality and it is far from creating soulful art *right now*. AI technology is improving at a breakneck pace. It's not a question of "if", but a question of "when".
@@112BALAGE112 I think you're both more or less right. Except it's debatable whether AI would have to be *sentient.* That begs the question, what *is* sentience? If you take a view like Michio Kaku, that it's all about feedback loops, sure, that sounds about right. **Is it tho?**
@@112BALAGE112 u can't compare art and games/chess. epistemologically they belong to different places. Thats the main problem in which AI will never be able to, because it is made by people who doesn't understand anything about humanities...
@@112BALAGE112 AI that plays computer games? Looks like masturbation, definitely! AI is not conscious, it can't be, so it will never capable of any CREATIVITY. It can "extract" information patterns from something created by a human consciousness, and use it to PRODUCE something that uses that patterns, but it will never be able to CREATE anything (and true art is about creativity). It is neither matter of "if" nor of "when", it is matter of "a machine is incapable of insight in principle".
Yes, AI can write uninspiring music like people. It can't really know what is good or bad music. It doesn't have a corporeal existence. What it needs is people grading the output music. From the points/grades, it can understand what is good and bad to write the next generation of music.
One thing to add to that James May clip, James May plays said piece thereby adding a human element. I would need to have the same computer to perform it to really tell how good it is.
I have actually thought about this too and here’s another idea I had: A lot of the time, music culture isn’t so much about the works as it is the personality who created them.
I'm not an AI researcher, but I know a thing or two. Recently a paper was published in the field of speech synthesis, about an AI which had analysed thousands of voices to learn the essence of human speech. After that, it could synthesise someone's voice with only 5 seconds of sample. Two-minute-papers has done a great video about it here ruclips.net/video/0sR1rU3gLzQ/видео.html. The problem with mimicking Bach based on Bach's music only (or Chopin on Chopin's music) is that there isn't really enough music out there to get a well trained AI. I think a technique similar to the voice synthesis technique could work very well to mimic the style of a certain composer, after learning "the essence of western music" or "the essence of baroque music" or just "the essence of music". However, the AI still needs text as input to synthesise speech, otherwise it can only speak random sounds of incoherent pseudo-speech (as far as I know). Even if you had an AI perfectly able to mimic a certain composer, you would still need the emotion or the progression of the piece as an input, otherwise it will create the same incoherent pseudo-music which AI's have created today.
Maybe a good use of this would be to finish various "unfinished" pieces out there. Doesn't need to be as directly creative, and a computer might be able to assume the perspective of a specific composer more objectively than any person.
"Can an A.I have a soul?" It sounded like the best player at Go before he was defeated Alpha go to 1-4 (Note: after 3 months Apha go was defeated by its younger brother by 0-50).
What about the work of David Cope who is a composer as well as a software programmer. He has also created new compositions in his own musical style using the software.
But what about tension and release? (Tension could very well be codified through the means of fast rythemic patterns, dynamics, tempo, not to mention dissonance of individual chords and chord progression, and although release is slightly more subjective, it could still be codified in a similar pattern) Additionally, it seems a fundamental problem with current AI composers is that they lack a sense of purpose in the music, ie. where the high points vs. low points are in their own piece. A significant advancement could be for the software developers to specify (or allow the program to choose) where it wants the moments of highest tension and release to occur, and then allow it to build up those specific moments, before finally using the AI to connect the moments of tension and release together. This would most likely fix the "No Ending" Problem.
The point you make about AI possibly not being able to create meaningful literature or music due to it’s lack of the human experience was somewhat mindblowing. I actually think it’s a necessity to understand human nature (yeah I know it’s a vague term) in order to handle the concept of meaning. Yet again, it would be imperfect or human, hence humans aren’t perfect. Fantastic video. Please consider to do more on the relationship of music and meaning.
But what is life? If you don't believe in a soul seperated to our body, this means that we are (and life is) a series of chemical and biological reactions. If you think deep, we are only the way our all atoms and molecules are organized and work. The questions is, couldn't we make a machine that copies exactly all the brain of a person and how it works, copying the function of every single cell with programming making it able to feel?
The fact that J, S. Bach has a "style" means that there must be a pattern. If you can say "That sounds like Bach." you have discerned the pattern, and so will an AI sooner or later.
Bach's 'style' is an entire era of writing known as the Baroque era. Bach was a Baroque composer who handled music quite differently from his contemporaries. Imitating Bach would be different from imitating Handel or Telemann; they would certainly yield different results.
I’ve heard about this AI ‘uprising’ in music and literature and I thought that it was the end of my goals as I’m an uprising composer currently but I’m glad to see that hopefully it will be out of my lifetime where us composers are put out of a job. Thank you for the video.
Thank you so much for making this video!!! You are right! And I am glad that we artists are safe from having our purpose taken away from us, at least for now
Can AI word paint? Can AI decide the importance of text, ascertaining what passages may need more clarity and switch to a homophonic from a contrapuntal texture, and make decisions about register and declamatory passages? This is where a lot of Bach's genius lies. 'Jesu, der du meine Seele' contains remarkable examples of how Bach interprets text through music. Hearing an AI compose a cantata on this level would be interesting and probably impossible.
Hey just discovered your channel today. I am a Ph.D. in AI at Montreal, and our lab's work has been referred to in the early work of Magenta, the google research team responsible for AI for art. A lot of what you've said here is very true, and the fact that taking some statistics over the past music might be only generating bland and boring pattern is also very well rooted from a statistical point. There are a lot of deep problems here, and for me, the most important one is: Can creativity be described as statistics? (which you also described in the Bellie Elish example). I think this is a good video to let people not to be too hyped about AI. In my knowledge, the current stage: Whether it's text/music generation, we can adequately have correct syntax, we just barely touch the coherence, and we still see no valid way of reaching meaningful sentences or melody. In general, the current AI is mainly good at "intuition" or "hunch" such as recognizing cat, riding bike etc. You don't really "think", or you are not aware that you are thinking about these things or skills. Meanwhile, the current AI sucks at planning or reasoning, which I think is crucial for tasks like composing as well. Would love to hear back, and good job on the video!
It can analyze how patterns change with time and try to predict a way to innovate musical structures while conserving an amount of “reference” to acclaimed styles of the past. It also needs to be taught about “tension and release” structures reliably.
Robots dont know how tension"FEELS" you cant learn a feeling...robots can learn at a rapid pace sure but..no matter how advance they cant feel anything
I really love your videos and find them truly insightful! After graduating from College, I was really looking for some intellectual stimulation akin to the one we would be getting during lectures or in conversations with teachers and fellow musicians. But being mostly a piano tutor now, I found it very difficult to recreate this sort of stimulating environment and entourage. Yet thanks to RUclips channels like yours, I find I can again get my daily dose of food for thought which inspires me both as a musician and teacher. Thank you for that!
Hey, bro I'm a drum and bass producer, id like to use the first 30 seconds as a sample but the background music gets in the way, could i get just the speech??
musicians a long time ago: *I make masterpieces with just an instrument and a sheet of paper* 🎻 musicians now: *Boohoo! I'm too lazy to learn music theory, I can't work without my AI, my autotune, my 2GB wi-fi, my expensive computer, my video editor, my loudspeakers, my microphone, my headphones, etc* 😭
Even if an AI made a piece of music, it was the human who made it. It won't be like Alpha go Zero, which started with only the rules of the game. AI must need previous music input and some concept and decision making decided by the coder. It will be he who is doing the composing; AI is just the newest tool on the block.
This is a subject that I have analysed a lot, practically and philosophically. I have seen that spiritually inclined people often have a tendency to ultimately conclude that it is not and cannot be possible. Now, I am a spiritually inclined person myself. But my conclusion differs. I think that in principle it is possible. In essence, my view is that we wrongly think that it is we, as spiritual awareness beings, who create music, literature, art, etc. But although we experience these things, we are not creating them. It is our intelligence that creates them. But we are not our intelligence. We are not our body, nor our intelligence. We are the experiencer of our body and intelligence, but we wrongfully think that we are these things. Thinking that we are these things is materialism. We are beyond these things. In principle there is no limit to how intelligent an AI can be. I don't distinguish between Intelligence and Artificial Intelligence. Intelligence is just Intelligence. But we are not the intelligence, we experience it.
Let's start by having an AI produced moderately competent film music (including suitability to the mood required in the film/TV scene in question). Much competent film music (as opposed to the best) basically works by taking familiar cliches of classical music and reworking them into the appropriate context.
Your video resonated with me as I am currently programming an A.I. music "composer", which I only started working on this year. So much of what you say rings true; about music simply being patterns of melody and harmony, essentially lacking the human soul and emotion. If any part of A.I. generated music is enjoyable, then surely it is purely a coincidence. I believe that it has to be a coincidence, since I have not yet found a way for the A.I. to 'listen' to the tune and provide itself with critical feedback. However, by applying (western) music rules, the results are already subjectively enjoyable and encouraging, although I have not yet started on the expected/unexpected musical phrases, periods and sentences. That is my next step, but first I need to read books and watch videos on A.I. Ideally I would like an A.I. discussion group. I have not uploaded existing tunes into the A.I. for it to replicate, therefore one could argue that it is not actually an intelligence, artificial or otherwise. It is simply following defined music theory rules. Following a program. Like with many composers, I am not actually creating for the enjoyment of others, but for my own joy, but my creativity is incorporating the rules and the logic in the program. The music, one could say, is just a by-product and, just like Mr Burns in your video, I throw away any piece that does not sound good to me, even though it could be a masterpiece to someone else. 'Music' is indeed subjective. It is early days with no guarantee that it will ever achieve anything substantive. It is, for now, my programming creativity that provides me, personally, with a satisfying feeling of achievement. imho there will only be a SkyNet if some insidious programmer chooses to incoporate inhumane rules or excludes Asimov's Three Laws of Robotics. autonomou5 is the stage-name for the music created by the A.I. D.J. (AIDJ) and where I upload any tunes that are refreshingly different from previously uploaded tunes, are particularly enjoyable (subjective) or follow significant program modifications or evolutions. The channel welcomes philosphical feedback through comments from musicians, composers and AI programmers.
How is there no mention about OpenAI's Musenet? It's the uncontested champion of computer-generated music. You can play with it freely at openai.com/blog/musenet . All of the examples are less spirited than most of what Musenet produces. It has its limitations and it requires curation to achieve the best results, but there's nothing half as good as it.
Great video. Yokky in my opinion was on the mark as we involve ourselves with music because of the pleasure and satisfaction we get the process as this links us to something greater via our ability to be 'inspired'. 'Creativity' always produces novelty. In an interview with George Harrison I once saw he asked the question..."where does a melody come from" and that I feel is the point. AI needs human hands to produce, whereas we can be creators simply by opening up to the process of creation. Thanks for the video.
I totally agree with the argument made in this video about the singularity of human experience. I think that science, with all of its marvelous achievements, can't fully grasp certain philosophical problems. Perhaps phenomenology (the philosophy which breaks through the scientific paradigm by placing human experience as the primary source of understanding of the world) can provide great insights on this matter. Great video!
There can be so many great examples in Beethoven's writings on the human expressionism. His Moonlight Sonata is a simple pattern, but its sense of longing is what makes it one of his most famous pieces. Also how his 7th Symphony depicts an almost juvenile view of excitement in the world, while the second movement reminds us of gloom, as it was composed for soldiers. The second movement resonated with them because of the darkness they had seen, but the rest of the symphony reminded them of the causes they were fighting (and dying for), -a better world for their wives and children. But most of all, Beethoven's 5th Symphony. No A.I. would have thought to make a symphony around the beat of a knock on the door to convey a message of fate kicking you out of your comfort, but as for Beethoven, a knock on the door was his neighbors complaining about his music, his landlord kicking him out again. To liken the cruelty of fate to a doorknock makes sense to what Beethoven lived through. Computers are great entertainers, but they aren't philosophers. All art is both.
Essentially it comes down to the problem or challenge of a machine's inability to experience emotion and then communicate it. It might make an approximation; it might learn that very specific patterns or mechanical events correlate to a word that humans will understand, it might even learn that the off switch equates to human death, but can it learn what the pain is of losing a loved one, and can it communicate that to a human in a way that a human can relate to it? Even the brilliant HAL 2000 computer of "2001 A Space Odyssey" -although it could carry on a conversation and interpret and give orders, even commit murder - it could communicate the awareness of its own demise but without inspiring empathy, sadness or pity for itself. A purely mechanical and intellectual "understanding" but with no emotional or human connection. We are impressed that it can express an awareness that it is dying - that's a pretty awesome feat - but we don't feel sad about it, HAL doesn't move us to tears.
Surely AI can open up pathways for artists themselves, as opposed to simply replacing them. AI art with human adjustment can be an incredible tool - I don't think we should stifle it.
Most of the arguments in this video are very good but I strongly disagree with the comment at 18:09. If one really composes not for others but for themselves, then they shouldn't care about AI, since they can always keep making music. I think even if AI could perfectly create music, people would also still create it, just like people still play chess, so there's really nothing for the music community to worry about. The last thing we should be doing is stopping AI research on music, because the machines may discover some interesting new things along the way that we can learn from.
Now I learned: The piece composed by "AIVA" (6:57) can be called plagiarism from Star Wars ("Rhey's Theme"). And 16:12 is nearly Beethovens Bagatelle op 119 No 1 in Major instead of Minor.
As always, the artificial intelligence argument comes down to the difference between "being able to think for itself" and "being able to FEEL something". Until artificial intelligence can actually feel pain, remorse, gratitude, a sense of accomplishment, etc. than it really doesn't matter. It'll be a nice thing to hear, but it's like a tractor plowing a field faster than a horse. It's a machine using one's and zero's to calculate a function. It's not like it needs to receive gratitude for it's nice little song. Now, when artificial intelligence can actually FEEL something, that's when the world will change.
This was a great, well-informed video! A job nicely done. I've always been concerned with AI writing music but handling harmonic progression like Bach is always a tricky issue. Copying Bach's method of writing is even difficult for other people to replicate because machines don't have emotions. I suppose you could teach a computer to write generic Baroque music, and I'd be glad because it might be better than anything I could write.
AI-generated music is inherently disadvantaged when being compared to human-written music. Chief among these disadvantages are the unwarranted inclusion of various outdated or poorly-constructed AI into the consideration of the quality of all AI outputs, cherry-picking of high-quality human-written music, and unequal limitations imposed on the human and AI composers. 13:26 Have you heard of GPT-2? The more impressive examples of its output include a coherent article written about unicorns (it has exactly one location where it demonstrates a lack of knowledge, which is when it calls the unicorns "four-horned"). That old Harry Potter AI example is an almost perfect example of how not to write AI. It has nowhere near a reasonable amount of training data, evidenced by the lack of structure. In fact, I would hesitate to even call it AI for one simple reason: given the amount of training data (a mere 7 books), it produced actually semi coherent results! The positive AI examples are cherry-picked, which leads to most outputs being bad while some rare outputs are good. But I'll also point out that the positive human examples are also cherry-picked. In that timeline that you showed, we see less than fifty names out of the (probably) millions who ever hummed a new tune. And even of those greatest composers, do people know and love every single piece, or are only the best-known pieces judged, and the remainder fall to the wayside and are forgotten? The AI may output good quality music only rarely. But after all music has gone through the fires of public opinion, only the best remain, regardless of who (or what) wrote it. Most people cannot write memorable music, so I think that AI actually has them beat there. (I know that the best AI written pieces I've heard here are more than I could hope to write). I would hazard to say that if the AI output were to be run through a quality-check algorithm (which might be a person, another AI, or both), the music that it might write may start to approach the quality of human professional music. Due to natural selection-like effects, the only human music that comes to mind tend to be the best pieces of music written by the best composers of their time. The AI-generated music that is being compared against those pieces obviously don't match up. AI-generated music tends to be shown to the listener without a revision or verification step. This is in contrast to human composers, who may go through multiple drafts of the same piece of music and listen to it performed multiple times before it is released to the public. Thus, only sufficiently enjoyable music is released to the public by human composers, in contrast to the AI-generated music, which can lack such review steps. It may be possible to compare both of these outputs. But it may be more reasonable to compare the typical output of the AI to the typical output of a typical person. Or, one might compare the typical output of the AI against he typical output of today's best composers... but give them the same 30 seconds to write the music and no revision time. There are multiple effects that seem to reduce your perception of AI generated music in comparison to human-composed music. (Did I just write an essay on a RUclips comment? Oh well...)
A.I. has a loooong way to go, no doubt about it. And there is so much more into making a good song, backing tracks are one thing, melody another... but there is the art of singing and vocal parts as well as mixing it all together in a coherent manner. I don’t expect A.I. will master any of this in the near future, but maybe we change what’s considered good music and if that subset is going in the direction of simplicity and repeated patterns... then computers will have greater success. Personally I believe music will survive because the human mind will keep the artistic aspect of it no matter what the pop charts spews out. I’d like to see what A.I. would do with my music though... please have go at it, I’ll be waiting. ;)
The Video is very interesting and good. The Answer to the Question , can AI make good Music is, No, but it can mix up existing Human made Music, and pull it to the individual Taste of Music. Even in this Video AI needs the Input of a Human to have a Point to start from. Humans do it by Feelings without needing a Start Point. They invent Music, even when they never heared a Song before, because they feel it. AI can not do that. Ai can not feel, so AI can only mix up the Input, but the Input is made by Humans, not by AI. ;) It is like a Reflection in the Mirror of a Picture from a real living beeing. It looks good, but it is only a reflection of a Picture. Like Wax Museum Madame Tussaud
What about OpenAI's Musenet? I've made some neat things on there (though I can't really tell if they're neat because I'm not a musician). What if musicians tried to make some songs with that?
Tempo...beat...swinging melody...from...low..swing...to high...back to low.....for melody guitar soul music !! Or ...to...make an authentic...love theme music from..love unlimited orchestra...or 70's car chasing movies soundtrack !!
Perhaps more interesting than whether algorithms can make music that sounds like human music is the opportunity for them to create entirely new forms. I’ve long been interested in generative or process music, from Cage, Reich and Eno to Autechre and Holly Herndon, and I often find the deliberate removal of human agency creates more intriguing and evocative pieces than a lot of music that heartily expresses emotion.
I think you missed that you also need a human experience to listen and to connect to music. Therefore I think that, at least theoretically, AI could create something that would allow you to create your own meaning at the moment of listening, without actually inputting any of it's own. Like you can hear something meaningful in the noise of wind or the Sea.
This is the core of the soul and logic problem... The ability to create is not ultimately what makes us, well us... The ability to create is a "triadigm" part of our reality... It includes all our experiences as captured by our brain (for short): our ability to commune with some greater truth which includes the imaginary realm, and other? yes.... Anyway to get good at creating anything artistic you need to practice pratice practice, feel feel feel the darkest places, and escape, listen or meditate... or commune.... Ok so what is the soul? It is what gets to feel this existence... It's not the existence... that is all apart of something else... that point that is us. It is where all conversation inside ourself goes ... beyond that we do not have that much control and that is where science takes over in our understanding... I guess our control is in our choices... what to listen to and what not to... where to walk toward and what to avoid? all of these things we can program into a computer but we will never create a soul... a computer that feels (even if we see it's actions as humanistic)... Ultimately the problem for humans is what makes us happy? being apart of the creating or creative process or just being fed creation? It was answered at our inception... we chose free will... the original sin is not actually a sin it is more like a curse... lol
I play guitar... one of my pedals can 'learn' from my chord progressions to come up with a drum track and a bass track... different styles produce different drum tracks within different genres, and different bass tracks in the same manner... the bass track generator can be set for different degrees of complexity, and I must say, I have been surprised more than once how complex the bass line could get, and still sound appropriate... it cost about $360...
18:41 If "doing art" is the purpose of life of someone, he might be missing the point of what it is to be a human being. The "passion" of your life is not the "meaning" of your life. I think that somehow the Pixar film "Soul" points in that direction.
cant achieve that without inspiration or creativity, we constantly struggle with making synthesize music sound natural, as producer or composers we constantly strive to create emotion, feeling in music, and we can only achieve this through human experience, not patters. however i do believe that AI will aid human beings in creating beautiful music much like how we use DAW's today.
People have absolutely no idea how just how fast AGI will be here. Right now we have 'Narrow Ai' however General Ai is only 10 yrs away and the world will be changed forever. The rate of advancement utilising Neural Networks and soon Ai writing it's own Algorithms will shock and awe everybody. AlphaGo was the turning point where people realised Ai can reach General Intelligence and very soon afterwards become super intelligent way beyond human capacity. Imagine an Ai doing a PHD in say 15 minutes where it takes a human 7 yrs. Ai is now unstoppable and will reach into every facet of our lives.
Wishful thinking. AI is not conscious, and will never be. Hence it will never be able to have ideas, it will always do mere mechanical recombinations of information. No creativity, no true intelligence in that. In chess game AI has already surpassed human capacity, so what? Even a bulldozer surpasses human capacity, so? AI will do a PhD in 15 minutes, but will not be able to have the idea that Einstein had, not even in 5000 years. Remember that a neural network is theoretically equivalent to a Turing Machine, and can only make computations. Intelligence is not computation.
@@AnnasVirtual I don't make AI and I am not interested in AI music, because I make music, and my effort is put in being an always better musician. If you are an AI programmer, you know what your AI programs do, because you programmed them. You may be surprised by the output, but you know for sure that the computer does exactly what you programmed it to do: mechanical computation. Computation will always be computation, even if you can't know how it actually produces its output. Creativity is not computation.
An AI replicates music (I think) in a similar way a psychopath or a sociopath emulates human emotion. To someone who is unaware it might be indistinguishable from the real thing but, if one were to find out it is fake or manufactured -- it changes everything entirely. Part of what makes various masterworks important is the historical significance. Just knowing about the composer's personal struggles/motivation gives the piece of music more significance (imo)... On a side note -- I believe things are cyclical and while it may seem like "good melody" is in the decline, I believe after enough time passes, people will get sick of the monotonousness and long-form melodies will become popular again... who knows, there may even be another golden age of serious music... Probably not in my lifetime though.
Aiva an AI and many others AIs which can make some semi - good music and the scary part is that it cost 50 dollar per year and you can make thousands of tracks in few minutes and you own the license to that music and that's not the scary part.. people are flooding spotify and other streaming platforms with ai music by saying that it's their own music and now the more scary part is that spotify also run on AI, if some one is uploading music regularly and creates enough momentum then spotify AI will recommend that music to audience and AI music is not great but it's good enough for general audience to enjoy, and many people are making lots of money from spotify by doing this and the sad part is... that money could have gone to many good artists who worked hard but it's not... it already has starting affecting us
I played around with the Google doodle, and i got pretty consistently decent results when using a two chord structure, especially when it's a perfect cadence
The reason AI will never surpass humans in music is because it bases it's music off of thousands of other pieces, and tries to find patterns amongst them all, every piece together as a whole, when they are individual pieces of music. In other words, the more pieces of music you feed it, the more bland and generalized the music that it creates will become. Adding all of these pieces of music together takes away their individuality, their fingerprint. AI is only able to operate based on patterns, and music is more than a pattern. Raw music is like a flowing stream, with no solid structure, but it is entirely connected. Raw music is like water, until passed through the human brain and placed into a structure. Think of water being forced through a maze. It will take shape of its container, but without a container it is a shapeless. AI can simulate a container, but it will never understand the flow of raw music. It can simulate patterns, but will never be able to "hear" the music that it's creating with them. AI makes music through sight of patterns, whereas humans make music through sounds and patterns.
I actually did develop a project where AI would try to create music it's own, using Long Short term memory(LSTM) over traditional RNN gave the best results but to actually make machines develop music which would actually qualify as 'Music' per se, in simple terms we have to feed them actual music and then the computer will create it by rearranging it in coherent manner. Using general patterns and feeding it rules relate to music and hoping it will identify patterns lead to their creations sound like a drunken hobo on a piano.
I'm a student programmer and composer and I can almost definitively say that AI will _eventually_ be able to do anything as good as a human can. Our minds are only a set of processes in the end, just highly complex ones that are influenced by not only one art form, but all of them at once. It'll take a while before it can effectively compose music that people like and is as complex as a human can compose though. One of the problems with the large companies like Google who just feed AI lots and lots of Bach's music, for example, is that they don't take into account the process of writing music at all. They just feed it complete music and get poor imitations of that music without much structure. That method will work eventually if enough data is given and enough personal feedback whether something is good or not is given to the AI. But Bach didn't just blindly write notes down on a paper. He went through a general process: melody, then harmony, then orchestration. AI first needs to be trained how to do those steps individually. Then, it must learn how to blur the lines between those steps. It has to be able to alter its melody to fit the harmony at times, and to alter the harmony and melody to fit the orchestration, and, most importantly, while composing a melody, it must keep in mind that harmony and orchestration will come down the line. It is an immensely complex process to train an AI to do this, but it's completely feasible with enough time. For now, what we should be working towards is not getting the AI to completely take over the process, but to alleviate potentially annoying tasks, such as orchestration. If you have a sound in mind, feed it an audio clip from some other piece and it'll orchestrate your sketches like that piece. That would be extremely useful.
But it can never orchestrate something exactly how you want it. Adding a fitting little details will be probably a most complicated thing in training AI.
And also i can't imagine how AI can be used for scoring to films and games. You will need kind of software with DAW functions where you can type what you want to hear in certain areas (every 3 seconds - something different) including usual options like mood, speed etc. But even this way i really don't see how it can create something that really fits to the picture plus includes character themes and stuff to the piece.
@@FireF1y644 An AI can't orchestrate something exactly how you want it if you already have a set plan. In that case, it's just more efficient to orchestrate your music yourself obviously. But if you don't know exactly what you want to do with the orchestration (which is the case a lot of the time), but you know you think the orchestration of some other pieces might fit what you are going for, you can feed those into a program with your sketch, and it'll spit out an orchestrated composition, or you can just feed it the sketch, and a well trained neural network (basically an AI's brain) will be able to find stuff that it thinks will fit. It's important to point out that there are many ways of training a neural network as well. You can train it to mimic something as close as possible by feeding it a lot of data (such as in this example: ruclips.net/video/Ipi40cb_RsI/видео.html ), you can train it to figure out how to do something all on it's own by giving it a reward system so it prefers when it gets closer to a goal you have in mind (such as in this example: ruclips.net/video/Tnu4O_xEmVk/видео.html ), and you can mix any training methods together to create a hybrid method. This "hybrid method" form of training is what will allow AI to compose its own original material. You can feed it a set of a bunch of music (such as a human listens to a lot of music), and give it a reward when it pleases you. Then, once it's learned how to compose like other composers, you can turn up the favorability of experimental neural pathways so it experiments. You'd then give it a greater reward when it succeeds so in the future, it'll prioritize original material that sounds "good." By using this type of hybrid training, you can get it to decide on it's own what "fitting details" it wants to add, or you can ask it to mimic the "fitting details" of other composers. No AI will ever compose exactly like you, just as no human will ever compose exactly like Beethoven. But AI can learn to prioritize its own take on music genres and it'll develop its own personality. With enough time, data, and a well structured rewards system, AI can brute force its way into learning how to do anything. If you feed an AI every John Williams score and reward it when it learns what a character is, what that character represents, what a theme is, what that theme represents (ie. evil, mystery, etc.) and when associates the right themes with a character with the correct mood, you will _eventually_ get an AI that composes for film like John Williams does. But that will take years of processing time and manual reward gifting. This approach, on a smaller scale, is what companies like Google have done for their AI composing projects. There are, however, much better ways to teach an AI how to score a film or a game. But you have to teach it like a human. For film scoring, you'd first have to teach it the basic things that are built into humans genetically such as the ability to recognize a person, place, thing, and eventually an abstract thing like an idea. Then you have to teach it how to recognize mood, ideas, characters, etc. in a film. Then you have to put it through music school. You have to go lesson by lesson. Teach it what a scale is, then test it on what a scale is. Teach it what a melody is, then teach it how to compose a melody. Etc. etc. Eventually you can teach it to put its knowledge on humans and music together and it will be able to create its own score for a film based on its music preferences and how it perceives the mood of a film. This will happen eventually and there's no way any of us can stop it, but it will take a long time for a project like this to be completed. The more reasonable short-term applications for AI in music is for it to replace individual steps. Orchestration is likely the first step because the worst an AI can do is orchestrate something by a set of rules with no creativity. Then composing melodies, understanding form, developing melodies, and being able to take into consideration all those steps at once will follow. Our brain is just a very complex neural network. The work of AI will eventually be able to be indistinguishable from the work of a human. Star Trek had it right in the 80s: we will eventually (likely within the next century) be able to create something along the lines of a "positronic brain" like commander Data has.
@@AnnasVirtual Dont try to impress me, I have seen these and much more thousands of times. AI can not do complex, meaningful, original and contextual music right now.
Great topic!!! As other pals have pointed out, David Cope's software (Emmy) did a pretty good job on this regards. It (or she?) was trained on known music pieces and was able to perfectly recreate the style of the original composer, yet the compositions sounded absolutely novel. Cope and Douglas Hofstadter also organized challenges where finally musicians were unable to distinguish which composition was computer created. And this happened decades ago. We are in a phase of rediscovery in this field, but I have not yet heard, among the many existing self-defined AI programs, one which is able to come even close to the results obtained by Cope's piece of software. Some compositions could also be found here on RUclips (search for "david cope emmy").
One day, there will be frustration among a generation of musicians, when the world prefers the machine to play, and humans as we presently know it are a thing of the past
The answer to your question of the name of this video is yes it can make good music but I was asking can you also remove the lead vocal so that way you only hear the backup voices only with AI separation?
today there are AI programs which compose really good, realistic music ; just don't tell people in advance it was created by AI otherwise they'll purposely force themselves to "not like it" ; and they won't be able to tell it was not written by a human
18:06 even if computers and AI could make perfect art, it wouldn't stop humans to create their own art. I don't see why computers making music would make composers go jobless. You can have too many taxi cab drivers; but you cannot have too many art
“No so handsome now” thought Harry as he dipped hermione in hot sauce. SCANDALOUS!
Well, it makes sense. You would not be particularly handsome if covered in hot sauce.
@@dingus_doofus Because YOU are putting a sense in it. Think about "idiolects", in which apparent meaningless sentences have a precise meaning for the idiolect speakers. Or technical jargons, in which common words are used to name completely different things.
The point is that the AI program is not conscious of any of all these meanings. It just extracts patterns and recombines sentences.
@@AnnasVirtual It can get better how much you want. It will always remain an unconscuious mechanism.
“Can a computer sound more advanced than Bach. Can it sound like Chopin?”
OH NO HE DIDN’T!
Tru
just arrived to here from that very video you quoted.
Well sadly yes
Human brain fires electrical impulses betwenw axons around 200m/s and it works around 200Hz and we are inteligent
But AI can travel information at speed of light and it process is 2GHz
So it does job for like 3 weeks what would human do for milenium
AI starts very stupid but at some tipping point it pogreses very fast and it surpasses human skill :/
@@heposlis2409 I'm talking about the dude saying that Chopin is more advanced than Bach. Which I personally don't think is true
Yeah, Bach's mastery of contrapuntal composition, polyphony, tonality, and structure, is doubtless more musically intriguing.
(or maybe that's just me :p)
And Amazingly, my video has already been demonetized within seconds of finishing uploading it! It features a 15 second clip from an ancient video of James May explaining something about A.I... I really thought I had avoided any demonetization issues in this one by being sparing with copyrighted content, I really did. But there we go - that was enough to screw me. I understand and am grateful for RUclips working the way it does, but sometimes it can be incredibly defeating when you've spent many hours on making content, to have your month's income from RUclips sliced in half. More fool me, I suppose!
If any of you would like to support my work on a more reliable platform, you can buy me a coffee and get access to new exclusive series on Patreon: www.patreon.com/insidethescore
Darn, tough luck. It's here now though! Love the videos, they're very informative and helpful to both my knowledge and my work! Keep it up! :D
Oof. That sucks
We appreciate the time spent nonetheless. Thank you sir.
It's becoming extremely difficult to find quality content on RUclips that is new. This algorithm they've created doesn't just hurt creators like yourself. It also hurts those of us who actually want too hear your teachings, and the voices of other creators who offer alternative view points to a myriad of subjects. What I truly don't understand is why haven't creators banded together and sued RUclips for "Unconscionable Conduct" in their contract. As the stronger party in any contract is not allowed to perform any action that damages the weaker party. This is what is happening.
To be intellectually honest with you, its very likely machine intelligence will exceed mankind in the near, or century long span future. Think about it... it already can do almost everything our visual cortex, and temporal regions of both hemispheres can better, and to some extent the parietal lobes too. The last frontier, and barrier to artificial general intelligence, is replication the frontal lobe. Consequently this is the last to mature, and is responsible for our conscious awareness, therefore the means to making the inanimate sentient. This new processing magnitude does not rely on the calorie, and without question, will supersede man. Steven hawkings underwent an interview before his death with the same concept. The truth is, we consider ourselves the dominant species, but we are merely a sidenote in revelation to the fungal spore. They have been proven to survive the vacuum of space, immense radiation, and are unquestionably prime candidates for panspermian infection. Where on earths ecosystem is it required to evolve resistance to being doused in uranium enriched liquid nitrogen? Its probably an organism that originally arose on a planet in its death throws. Its mycotoxic lethality, a competitive edge as a chemical constituent to the degradation of previously unencountered species of other habitable worlds I have personal experience with affliction, and despite my nutritive shielding of the most powerful antitoxins known to man, at high dosages, I was chemically wounded. Its closeness to living animals, both in DNA, and its respiration, suggests it might in fact be one of the most, if not the most successful extremophile in the entire galaxy, and its reign amongst the stars riding in on collisioned pebbles and ice, dwarfs our very own magnificience with all our craft, and they will exist long after our extinction, even if it came from a foreign mass far greater than what struck 65 million years ago...... and yet its probably where we even came from. It is time we shed our arrogance, and embraced the fact, we will construct our very own god. Fungus might be the fathership of our world, mutations bringing us into existence the cadence of time, where the origins stabilized by raw success. Fungal pioneerists might be, a biological space ship, our divine origins considering abiogenesis would be a rare process, with intelligent life being rarer still. That makes us an intermediary step in the evolution of intelligence, as musk puts it, the bootloader for what without question would be the most intelligent form of organization in the entire cosmos, alien civilizations of the billions of year mark ancient included. Biology simply cannot compete, and it never will for the infinitude of all.
I completely disagree with Yokky's comment. Most people work primarily to earn money, but not exclusively. A scientist might enjoy trying to formulate or test his ideas, a carpenter might enjoy making things with his hands, a doctor might find fulfillment in helping people. And even people with the most boring jobs (in my opinion) like truck drivers or accountants would maybe be less happy if they wouldn't be able to contribute to society. Artists are not the only people who like their jobs or need them to be happy, so if Yokky thinks scientists should stop research on Artistic AI (which in itself is a highly questionable statement in my opinion, not only because "someone else will and profit from it", but also because I believe in freedom of ideas)
he should think the same about research on automation in factories. Why make an exception for artists?
sebastiaan kruis I agree. People can be very passionate about things outside of high art.
My two cents as both a composer and programmer:
- The problem in making music with AI is that you can't automatically test the result. So you can't run like 1000000 sumulations while improving or 'learning' in every iteration, because humans still need to judge whether it is actually an improvement or not.
- Another problem is that not only do music trends in general break patterns, but a music piece itself is often a set of patterns that keep getting broken in different ways. Music is always a balance between order and chaos, and while humans are of course experts at creating chaos, computers are not.
- Apart from that I'd say the question "Can Artificial Intelligence ... ?" will almost always be yes in the (very) long term.
Perhaps (i) some neural networks that can learn to classify works as 'good' (human?) or bad with supervision, and then (ii) ??? and then (iii) adversarial training and then (iv) profit?
An area I might want to focus on is AI *interpretation* of music.
I think the approach of training an AI with tons of music to create something new is flawed from the start. It's too logical.
Perhaps creating an AI that thinks in a less logical way, that resembles the way humans think...with emotions etc., maybe that's how you get AI to create music that is actually interesting.
The solution to this is adversarial neural networks. One gets better at making, while the other gets better at discerning.
Yeah I kind of get the feeling he doesn't really understand AI. The whole 20 minutes just boil down to: "computers have no emotions and a sense of context". I'm 100% sure AI will be able to make actual art one day, given it has enough data
That Harry Potter excerpt was clearly written by Mark Zuckerberg.
BigBrotherMateyka Or any greedy billionaire on the planet lol.
"It really lacks soul and diversity..."
So do human beings. Ever turned on a pop radio station and listened to what millions of people think is the best song of the week?
Yes, most pop music is terrible. But that is beside the point. We should ask if AI can compose the equivalent of THE BEST of human composed music. So far it is a total flop. AI researchers have been trying for at least 50 years to compose music by computer. Can you name one hit ever composed by a computer? So yes, computer music lacks soul. That is because our emotional responses to music have criteria not easily captured by an algorithm. Perhaps an impossibility. While a particular good tune can be boiled to patterns, the reverse problem is the hard one, to find patterns that are pleasing in the first place. And the difficulty is not in finding a particular narrowly defined type, which might be easily characterized. But in finding a GENERAL algorithm that finds many. That is somewhat analogous to the famous halting problem. A specific program can be analyzed by an algorithm to see if it will eventually halt. But no algorithm exists that can analyze ANY program and determine if it will halt. Analogously, An algorithm might be contrived to compose a tune with narrow criteria (but even then our human response is not filtering it). It is entirely another thing to find an algorithm than can compose, in general, good tunes of unpredictable kind.
Those aren’t humans. Thats all record labels with soulless producers who circle jerk the same riffs until they squeeze every drop of blood out of it.
SuperFluffyShmoopy singers are fine, they just get the song lmao
@broe5010 broe5010 Well then, you need to emulate emotions on AI. This has been done already, in a way.
We may never find any music created by a computer full of emotion not because they will always lack the ability to have emotions likes us humans, but because of our cognitive biases.
As someone who has studied both music and data science, I must say that I have enjoyed your video quite a lot.
Very nice work. Enjoyed it a lot :)
He seems to have selected the more mediocre exemplars of AI Bach, which is surprising, given that some outstanding efforts debuted over 2 decades ago. Eg. David Cope. "Undiscovered Bach? No a Computer Wrote It." NYT, 1997.
"On a more ambitious IeveI, EMI recently composed a fuII-scale Mozart symphony and piano concerto, which were performed in April by the Santa Cruz Baroque Festival on period instruments. Linda Burman-Hall, who pIayed the piano solo for the concerto, said: ''lt felt a little different than playing a normal Mozart work. But it was very much Iike a work of the same period."
Keep in mind that the pianist, a professional musician, came to the piece knowing that it was a synthetic composition, but even still seemed disinclined to voice an unfavorable response.
AI can create meaningful creative content, it just needs an extraordinary amount of data. See GPT-2 which is an AI that can write English text.
@@DanLyndon Have you read the "unicorns found in the Andes mountains" story, GPT-2 has written? Yes it isn't perfect, but it's almost completely coherent and contains lots of information consistent with reality (is that insight?). Regarding games with clearly defined rules, consider DeepMind's latest innovation MuZero. It is an AI that plays chess, go and shogi perfectly and can also play 57 Atari games better than humans. Yes AI is limited in terms of insight and generality and it is far from creating soulful art *right now*. AI technology is improving at a breakneck pace. It's not a question of "if", but a question of "when".
Not just data, but electric power. "Save the climate" ho ho ho
@@112BALAGE112 I think you're both more or less right. Except it's debatable whether AI would have to be *sentient.* That begs the question, what *is* sentience? If you take a view like Michio Kaku, that it's all about feedback loops, sure, that sounds about right. **Is it tho?**
@@112BALAGE112 u can't compare art and games/chess. epistemologically they belong to different places. Thats the main problem in which AI will never be able to, because it is made by people who doesn't understand anything about humanities...
@@112BALAGE112 AI that plays computer games? Looks like masturbation, definitely!
AI is not conscious, it can't be, so it will never capable of any CREATIVITY. It can "extract" information patterns from something created by a human consciousness, and use it to PRODUCE something that uses that patterns, but it will never be able to CREATE anything (and true art is about creativity).
It is neither matter of "if" nor of "when", it is matter of "a machine is incapable of insight in principle".
Yes, AI can write uninspiring music like people. It can't really know what is good or bad music. It doesn't have a corporeal existence. What it needs is people grading the output music. From the points/grades, it can understand what is good and bad to write the next generation of music.
Me and my parents and siblings can't even fully agree on what good music is…
One thing to add to that James May clip, James May plays said piece thereby adding a human element.
I would need to have the same computer to perform it to really tell how good it is.
Makes me think of midi notes with no dynamism. that's basically what the machine will put out without any intelligence on dynamics.
I have actually thought about this too and here’s another idea I had:
A lot of the time, music culture isn’t so much about the works as it is the personality who created them.
I'm not an AI researcher, but I know a thing or two. Recently a paper was published in the field of speech synthesis, about an AI which had analysed thousands of voices to learn the essence of human speech. After that, it could synthesise someone's voice with only 5 seconds of sample. Two-minute-papers has done a great video about it here ruclips.net/video/0sR1rU3gLzQ/видео.html.
The problem with mimicking Bach based on Bach's music only (or Chopin on Chopin's music) is that there isn't really enough music out there to get a well trained AI. I think a technique similar to the voice synthesis technique could work very well to mimic the style of a certain composer, after learning "the essence of western music" or "the essence of baroque music" or just "the essence of music".
However, the AI still needs text as input to synthesise speech, otherwise it can only speak random sounds of incoherent pseudo-speech (as far as I know). Even if you had an AI perfectly able to mimic a certain composer, you would still need the emotion or the progression of the piece as an input, otherwise it will create the same incoherent pseudo-music which AI's have created today.
Somehow this analysis makes me feel the mind and matter tug of war between Aristotelianism and Platonism.
Maybe a good use of this would be to finish various "unfinished" pieces out there. Doesn't need to be as directly creative, and a computer might be able to assume the perspective of a specific composer more objectively than any person.
"Can an A.I have a soul?"
It sounded like the best player at Go before he was defeated Alpha go to 1-4
(Note: after 3 months Apha go was defeated by its younger brother by 0-50).
What about the work of David Cope who is a composer as well as a software programmer. He has also created new compositions in his own musical style using the software.
But what about tension and release? (Tension could very well be codified through the means of fast rythemic patterns, dynamics, tempo, not to mention dissonance of individual chords and chord progression, and although release is slightly more subjective, it could still be codified in a similar pattern) Additionally, it seems a fundamental problem with current AI composers is that they lack a sense of purpose in the music, ie. where the high points vs. low points are in their own piece. A significant advancement could be for the software developers to specify (or allow the program to choose) where it wants the moments of highest tension and release to occur, and then allow it to build up those specific moments, before finally using the AI to connect the moments of tension and release together. This would most likely fix the "No Ending" Problem.
The point you make about AI possibly not being able to create meaningful literature or music due to it’s lack of the human experience was somewhat mindblowing. I actually think it’s a necessity to understand human nature (yeah I know it’s a vague term) in order to handle the concept of meaning. Yet again, it would be imperfect or human, hence humans aren’t perfect.
Fantastic video. Please consider to do more on the relationship of music and meaning.
But what is life? If you don't believe in a soul seperated to our body, this means that we are (and life is) a series of chemical and biological reactions. If you think deep, we are only the way our all atoms and molecules are organized and work. The questions is, couldn't we make a machine that copies exactly all the brain of a person and how it works, copying the function of every single cell with programming making it able to feel?
The fact that J, S. Bach has a "style" means that there must be a pattern. If you can say "That sounds like Bach." you have discerned the pattern, and so will an AI sooner or later.
Bach's 'style' is an entire era of writing known as the Baroque era. Bach was a Baroque composer who handled music quite differently from his contemporaries. Imitating Bach would be different from imitating Handel or Telemann; they would certainly yield different results.
Ai can't create feelings,a soul ,a human brain.
@@Larrypint I agree with you.
@@unknown626 Great response! I think you're right.
"Daisy, Daisy give me your heart to do
I'm half crazy, hopeful in love with you..."
- Hal 9000
Jeff Duke that was actually emotional lol
It would be great if there can be an updated video based on the recent developments of the AI!
Great video bro, don-t get discouraged by youtube!
12:25 Colorless-Green,Ideas-Sleep furiously.
Now you can read a meaning.
Music is life, and love, a sense these machines cannot comprehend
I’ve heard about this AI ‘uprising’ in music and literature and I thought that it was the end of my goals as I’m an uprising composer currently but I’m glad to see that hopefully it will be out of my lifetime where us composers are put out of a job. Thank you for the video.
Thank you so much for making this video!!! You are right! And I am glad that we artists are safe from having our purpose taken away from us, at least for now
Can AI word paint? Can AI decide the importance of text, ascertaining what passages may need more clarity and switch to a homophonic from a contrapuntal texture, and make decisions about register and declamatory passages? This is where a lot of Bach's genius lies. 'Jesu, der du meine Seele' contains remarkable examples of how Bach interprets text through music. Hearing an AI compose a cantata on this level would be interesting and probably impossible.
Hey just discovered your channel today. I am a Ph.D. in AI at Montreal, and our lab's work has been referred to in the early work of Magenta, the google research team responsible for AI for art. A lot of what you've said here is very true, and the fact that taking some statistics over the past music might be only generating bland and boring pattern is also very well rooted from a statistical point. There are a lot of deep problems here, and for me, the most important one is: Can creativity be described as statistics? (which you also described in the Bellie Elish example).
I think this is a good video to let people not to be too hyped about AI. In my knowledge, the current stage: Whether it's text/music generation, we can adequately have correct syntax, we just barely touch the coherence, and we still see no valid way of reaching meaningful sentences or melody. In general, the current AI is mainly good at "intuition" or "hunch" such as recognizing cat, riding bike etc. You don't really "think", or you are not aware that you are thinking about these things or skills. Meanwhile, the current AI sucks at planning or reasoning, which I think is crucial for tasks like composing as well. Would love to hear back, and good job on the video!
90
8:25 well, it sounds like the minimalist composition movement.
It can analyze how patterns change with time and try to predict a way to innovate musical structures while conserving an amount of “reference” to acclaimed styles of the past. It also needs to be taught about “tension and release” structures reliably.
Robots dont know how tension"FEELS" you cant learn a feeling...robots can learn at a rapid pace sure but..no matter how advance they cant feel anything
I laughed so hard at that Harry Potter chapter.
I really love your videos and find them truly insightful! After graduating from College, I was really looking for some intellectual stimulation akin to the one we would be getting during lectures or in conversations with teachers and fellow musicians. But being mostly a piano tutor now, I found it very difficult to recreate this sort of stimulating environment and entourage. Yet thanks to RUclips channels like yours, I find I can again get my daily dose of food for thought which inspires me both as a musician and teacher. Thank you for that!
1:18 already parallel fifths in the tenor and bass! sAcRiLeGiOuS
Found the Twosetter.
Hey, bro I'm a drum and bass producer, id like to use the first 30 seconds as a sample but the background music gets in the way, could i get just the speech??
musicians a long time ago: *I make masterpieces with just an instrument and a sheet of paper* 🎻
musicians now: *Boohoo! I'm too lazy to learn music theory, I can't work without my AI, my autotune, my 2GB wi-fi, my expensive computer, my video editor, my loudspeakers, my microphone, my headphones, etc* 😭
Even if an AI made a piece of music, it was the human who made it.
It won't be like Alpha go Zero, which started with only the rules of the game. AI must need previous music input and some concept and decision making decided by the coder. It will be he who is doing the composing; AI is just the newest tool on the block.
This is a subject that I have analysed a lot, practically and philosophically. I have seen that spiritually inclined people often have a tendency to ultimately conclude that it is not and cannot be possible. Now, I am a spiritually inclined person myself. But my conclusion differs. I think that in principle it is possible. In essence, my view is that we wrongly think that it is we, as spiritual awareness beings, who create music, literature, art, etc. But although we experience these things, we are not creating them. It is our intelligence that creates them. But we are not our intelligence. We are not our body, nor our intelligence. We are the experiencer of our body and intelligence, but we wrongfully think that we are these things. Thinking that we are these things is materialism. We are beyond these things. In principle there is no limit to how intelligent an AI can be. I don't distinguish between Intelligence and Artificial Intelligence. Intelligence is just Intelligence. But we are not the intelligence, we experience it.
Let's start by having an AI produced moderately competent film music (including suitability to the mood required in the film/TV scene in question). Much competent film music (as opposed to the best) basically works by taking familiar cliches of classical music and reworking them into the appropriate context.
@4:50 sounded more like Schoenberg than Chopin. 😉.
No. It doesn't.
Does it though
Your video resonated with me as I am currently programming an A.I. music "composer", which I only started working on this year. So much of what you say rings true; about music simply being patterns of melody and harmony, essentially lacking the human soul and emotion. If any part of A.I. generated music is enjoyable, then surely it is purely a coincidence. I believe that it has to be a coincidence, since I have not yet found a way for the A.I. to 'listen' to the tune and provide itself with critical feedback. However, by applying (western) music rules, the results are already subjectively enjoyable and encouraging, although I have not yet started on the expected/unexpected musical phrases, periods and sentences. That is my next step, but first I need to read books and watch videos on A.I. Ideally I would like an A.I. discussion group.
I have not uploaded existing tunes into the A.I. for it to replicate, therefore one could argue that it is not actually an intelligence, artificial or otherwise. It is simply following defined music theory rules. Following a program.
Like with many composers, I am not actually creating for the enjoyment of others, but for my own joy, but my creativity is incorporating the rules and the logic in the program. The music, one could say, is just a by-product and, just like Mr Burns in your video, I throw away any piece that does not sound good to me, even though it could be a masterpiece to someone else. 'Music' is indeed subjective.
It is early days with no guarantee that it will ever achieve anything substantive. It is, for now, my programming creativity that provides me, personally, with a satisfying feeling of achievement. imho there will only be a SkyNet if some insidious programmer chooses to incoporate inhumane rules or excludes Asimov's Three Laws of Robotics.
autonomou5 is the stage-name for the music created by the A.I. D.J. (AIDJ) and where I upload any tunes that are refreshingly different from previously uploaded tunes, are particularly enjoyable (subjective) or follow significant program modifications or evolutions. The channel welcomes philosphical feedback through comments from musicians, composers and AI programmers.
How is there no mention about OpenAI's Musenet? It's the uncontested champion of computer-generated music. You can play with it freely at openai.com/blog/musenet . All of the examples are less spirited than most of what Musenet produces. It has its limitations and it requires curation to achieve the best results, but there's nothing half as good as it.
Very happy with you and also thankful to you as you are helping me
Great video. Yokky in my opinion was on the mark as we involve ourselves with music because of the pleasure and satisfaction we get the process as this links us to something greater via our ability to be 'inspired'. 'Creativity' always produces novelty. In an interview with George Harrison I once saw he asked the question..."where does a melody come from" and that I feel is the point. AI needs human hands to produce, whereas we can be creators simply by opening up to the process of creation. Thanks for the video.
I totally agree with the argument made in this video about the singularity of human experience. I think that science, with all of its marvelous achievements, can't fully grasp certain philosophical problems. Perhaps phenomenology (the philosophy which breaks through the scientific paradigm by placing human experience as the primary source of understanding of the world) can provide great insights on this matter. Great video!
There can be so many great examples in Beethoven's writings on the human expressionism. His Moonlight Sonata is a simple pattern, but its sense of longing is what makes it one of his most famous pieces. Also how his 7th Symphony depicts an almost juvenile view of excitement in the world, while the second movement reminds us of gloom, as it was composed for soldiers. The second movement resonated with them because of the darkness they had seen, but the rest of the symphony reminded them of the causes they were fighting (and dying for), -a better world for their wives and children. But most of all, Beethoven's 5th Symphony. No A.I. would have thought to make a symphony around the beat of a knock on the door to convey a message of fate kicking you out of your comfort, but as for Beethoven, a knock on the door was his neighbors complaining about his music, his landlord kicking him out again. To liken the cruelty of fate to a doorknock makes sense to what Beethoven lived through. Computers are great entertainers, but they aren't philosophers. All art is both.
Listen to the song from the Eruovision Contest 2019. It was created by Artificial Intelligence called Blue Jeans and Bloody Tears
Essentially it comes down to the problem or challenge of a machine's inability to experience emotion and then communicate it. It might make an approximation; it might learn that very specific patterns or mechanical events correlate to a word that humans will understand, it might even learn that the off switch equates to human death, but can it learn what the pain is of losing a loved one, and can it communicate that to a human in a way that a human can relate to it? Even the brilliant HAL 2000 computer of "2001 A Space Odyssey" -although it could carry on a conversation and interpret and give orders, even commit murder - it could communicate the awareness of its own demise but without inspiring empathy, sadness or pity for itself. A purely mechanical and intellectual "understanding" but with no emotional or human connection. We are impressed that it can express an awareness that it is dying - that's a pretty awesome feat - but we don't feel sad about it, HAL doesn't move us to tears.
The Harry Potter thing is like "When you put too many levels into grammar and forget to put anything into semantics."
Surely AI can open up pathways for artists themselves, as opposed to simply replacing them. AI art with human adjustment can be an incredible tool - I don't think we should stifle it.
Most of the arguments in this video are very good but I strongly disagree with the comment at 18:09. If one really composes not for others but for themselves, then they shouldn't care about AI, since they can always keep making music. I think even if AI could perfectly create music, people would also still create it, just like people still play chess, so there's really nothing for the music community to worry about. The last thing we should be doing is stopping AI research on music, because the machines may discover some interesting new things along the way that we can learn from.
Now I learned: The piece composed by "AIVA" (6:57) can be called plagiarism from Star Wars ("Rhey's Theme"). And 16:12 is nearly Beethovens Bagatelle op 119 No 1 in Major instead of Minor.
As always, the artificial intelligence argument comes down to the difference between "being able to think for itself" and "being able to FEEL something". Until artificial intelligence can actually feel pain, remorse, gratitude, a sense of accomplishment, etc. than it really doesn't matter. It'll be a nice thing to hear, but it's like a tractor plowing a field faster than a horse. It's a machine using one's and zero's to calculate a function. It's not like it needs to receive gratitude for it's nice little song. Now, when artificial intelligence can actually FEEL something, that's when the world will change.
This was a great, well-informed video! A job nicely done. I've always been concerned with AI writing music but handling harmonic progression like Bach is always a tricky issue. Copying Bach's method of writing is even difficult for other people to replicate because machines don't have emotions.
I suppose you could teach a computer to write generic Baroque music, and I'd be glad because it might be better than anything I could write.
AI-generated music is inherently disadvantaged when being compared to human-written music. Chief among these disadvantages are the unwarranted inclusion of various outdated or poorly-constructed AI into the consideration of the quality of all AI outputs, cherry-picking of high-quality human-written music, and unequal limitations imposed on the human and AI composers.
13:26 Have you heard of GPT-2? The more impressive examples of its output include a coherent article written about unicorns (it has exactly one location where it demonstrates a lack of knowledge, which is when it calls the unicorns "four-horned").
That old Harry Potter AI example is an almost perfect example of how not to write AI. It has nowhere near a reasonable amount of training data, evidenced by the lack of structure. In fact, I would hesitate to even call it AI for one simple reason: given the amount of training data (a mere 7 books), it produced actually semi coherent results!
The positive AI examples are cherry-picked, which leads to most outputs being bad while some rare outputs are good. But I'll also point out that the positive human examples are also cherry-picked. In that timeline that you showed, we see less than fifty names out of the (probably) millions who ever hummed a new tune. And even of those greatest composers, do people know and love every single piece, or are only the best-known pieces judged, and the remainder fall to the wayside and are forgotten?
The AI may output good quality music only rarely. But after all music has gone through the fires of public opinion, only the best remain, regardless of who (or what) wrote it. Most people cannot write memorable music, so I think that AI actually has them beat there. (I know that the best AI written pieces I've heard here are more than I could hope to write). I would hazard to say that if the AI output were to be run through a quality-check algorithm (which might be a person, another AI, or both), the music that it might write may start to approach the quality of human professional music. Due to natural selection-like effects, the only human music that comes to mind tend to be the best pieces of music written by the best composers of their time. The AI-generated music that is being compared against those pieces obviously don't match up.
AI-generated music tends to be shown to the listener without a revision or verification step. This is in contrast to human composers, who may go through multiple drafts of the same piece of music and listen to it performed multiple times before it is released to the public. Thus, only sufficiently enjoyable music is released to the public by human composers, in contrast to the AI-generated music, which can lack such review steps. It may be possible to compare both of these outputs. But it may be more reasonable to compare the typical output of the AI to the typical output of a typical person. Or, one might compare the typical output of the AI against he typical output of today's best composers... but give them the same 30 seconds to write the music and no revision time.
There are multiple effects that seem to reduce your perception of AI generated music in comparison to human-composed music.
(Did I just write an essay on a RUclips comment? Oh well...)
Excellent points.
*Iannis Xenakis* is typing...
Lol I even read your name as Synthesia Greek
0:50 What program or website is that?
I wonder what's that too
A.I. has a loooong way to go, no doubt about it. And there is so much more into making a good song, backing tracks are one thing, melody another... but there is the art of singing and vocal parts as well as mixing it all together in a coherent manner. I don’t expect A.I. will master any of this in the near future, but maybe we change what’s considered good music and if that subset is going in the direction of simplicity and repeated patterns... then computers will have greater success.
Personally I believe music will survive because the human mind will keep the artistic aspect of it no matter what the pop charts spews out. I’d like to see what A.I. would do with my music though... please have go at it, I’ll be waiting. ;)
The Video is very interesting and good.
The Answer to the Question , can AI make good Music is, No, but it can mix up existing Human made Music, and pull it to the individual Taste of Music.
Even in this Video AI needs the Input of a Human to have a Point to start from.
Humans do it by Feelings without needing a Start Point.
They invent Music, even when they never heared a Song before, because they feel it. AI can not do that.
Ai can not feel, so AI can only mix up the Input, but the Input is made by Humans, not by AI. ;)
It is like a Reflection in the Mirror of a Picture from a real living beeing.
It looks good, but it is only a reflection of a Picture.
Like Wax Museum Madame Tussaud
12:45 makes sense to me, it is quite similar to saying >
"The light sleeps furiously in the darkness".
What about OpenAI's Musenet? I've made some neat things on there (though I can't really tell if they're neat because I'm not a musician). What if musicians tried to make some songs with that?
I'd recommend starting without an intro. That's what I usually do and it works pretty well, and it's more relevant to what's being discussed.
Tempo...beat...swinging melody...from...low..swing...to high...back to low.....for melody guitar soul music !!
Or ...to...make an authentic...love theme music from..love unlimited orchestra...or 70's car chasing movies soundtrack !!
That Harry Potter imitation is probably close to what Tommy Wiseau would create if he tried to write a book
AIVA: Artificial Intelligence Virtual Arpeggiator
That's all it does, just repetitive arpeggios with some boring notes on top...
Perhaps more interesting than whether algorithms can make music that sounds like human music is the opportunity for them to create entirely new forms. I’ve long been interested in generative or process music, from Cage, Reich and Eno to Autechre and Holly Herndon, and I often find the deliberate removal of human agency creates more intriguing and evocative pieces than a lot of music that heartily expresses emotion.
I think you missed that you also need a human experience to listen and to connect to music. Therefore I think that, at least theoretically, AI could create something that would allow you to create your own meaning at the moment of listening, without actually inputting any of it's own. Like you can hear something meaningful in the noise of wind or the Sea.
The essence of human creativity is recombination. Innovation comes through the combination of previously uncombined material.
This is the core of the soul and logic problem... The ability to create is not ultimately what makes us, well us... The ability to create is a "triadigm" part of our reality... It includes all our experiences as captured by our brain (for short): our ability to commune with some greater truth which includes the imaginary realm, and other? yes.... Anyway to get good at creating anything artistic you need to practice pratice practice, feel feel feel the darkest places, and escape, listen or meditate... or commune.... Ok so what is the soul? It is what gets to feel this existence... It's not the existence... that is all apart of something else... that point that is us. It is where all conversation inside ourself goes ... beyond that we do not have that much control and that is where science takes over in our understanding... I guess our control is in our choices... what to listen to and what not to... where to walk toward and what to avoid? all of these things we can program into a computer but we will never create a soul... a computer that feels (even if we see it's actions as humanistic)... Ultimately the problem for humans is what makes us happy? being apart of the creating or creative process or just being fed creation? It was answered at our inception... we chose free will... the original sin is not actually a sin it is more like a curse... lol
"listen to this example" it sounded like an amateur semi-randomly hitting notes. That was hilarious.
I play guitar... one of my pedals can 'learn' from my chord progressions to come up with a drum track and a bass track... different styles produce different drum tracks within different genres, and different bass tracks in the same manner... the bass track generator can be set for different degrees of complexity, and I must say, I have been surprised more than once how complex the bass line could get, and still sound appropriate... it cost about $360...
18:41 If "doing art" is the purpose of life of someone, he might be missing the point of what it is to be a human being. The "passion" of your life is not the "meaning" of your life.
I think that somehow the Pixar film "Soul" points in that direction.
cant achieve that without inspiration or creativity, we constantly struggle with making synthesize music sound natural, as producer or composers we constantly strive to create emotion, feeling in music, and we can only achieve this through human experience, not patters. however i do believe that AI will aid human beings in creating beautiful music much like how we use DAW's today.
Im glad there are people with some sense in this comment section man
People have absolutely no idea how just how fast AGI will be here. Right now we have 'Narrow Ai' however General Ai is only 10 yrs away and the world will be changed forever. The rate of advancement utilising Neural Networks and soon Ai writing it's own Algorithms will shock and awe everybody. AlphaGo was the turning point where people realised Ai can reach General Intelligence and very soon afterwards become super intelligent way beyond human capacity. Imagine an Ai doing a PHD in say 15 minutes where it takes a human 7 yrs. Ai is now unstoppable and will reach into every facet of our lives.
Wishful thinking. AI is not conscious, and will never be. Hence it will never be able to have ideas, it will always do mere mechanical recombinations of information. No creativity, no true intelligence in that.
In chess game AI has already surpassed human capacity, so what? Even a bulldozer surpasses human capacity, so?
AI will do a PhD in 15 minutes, but will not be able to have the idea that Einstein had, not even in 5000 years.
Remember that a neural network is theoretically equivalent to a Turing Machine, and can only make computations.
Intelligence is not computation.
@@AnnasVirtual I don't make AI and I am not interested in AI music, because I make music, and my effort is put in being an always better musician.
If you are an AI programmer, you know what your AI programs do, because you programmed them. You may be surprised by the output, but you know for sure that the computer does exactly what you programmed it to do: mechanical computation. Computation will always be computation, even if you can't know how it actually produces its output.
Creativity is not computation.
Brilliant work do you think the new Abba works encompass this ?
4:36 “something more ADVANCED than Bach..."?
What about the "emmi" software by David cope? It sounds very real to me...
An AI replicates music (I think) in a similar way a psychopath or a sociopath emulates human emotion. To someone who is unaware it might be indistinguishable from the real thing but, if one were to find out it is fake or manufactured -- it changes everything entirely. Part of what makes various masterworks important is the historical significance. Just knowing about the composer's personal struggles/motivation gives the piece of music more significance (imo)... On a side note -- I believe things are cyclical and while it may seem like "good melody" is in the decline, I believe after enough time passes, people will get sick of the monotonousness and long-form melodies will become popular again... who knows, there may even be another golden age of serious music... Probably not in my lifetime though.
Aiva an AI and many others AIs which can make some semi - good music and the scary part is that it cost 50 dollar per year and you can make thousands of tracks in few minutes and you own the license to that music and that's not the scary part.. people are flooding spotify and other streaming platforms with ai music by saying that it's their own music and now the more scary part is that spotify also run on AI, if some one is uploading music regularly and creates enough momentum then spotify AI will recommend that music to audience and AI music is not great but it's good enough for general audience to enjoy, and many people are making lots of money from spotify by doing this and the sad part is... that money could have gone to many good artists who worked hard but it's not...
it already has starting affecting us
I played around with the Google doodle, and i got pretty consistently decent results when using a two chord structure, especially when it's a perfect cadence
I like how this is also one of the topics in Gödel, Escher, Bach. The book is getting more relevant by the second.
The reason AI will never surpass humans in music is because it bases it's music off of thousands of other pieces, and tries to find patterns amongst them all, every piece together as a whole, when they are individual pieces of music.
In other words, the more pieces of music you feed it, the more bland and generalized the music that it creates will become. Adding all of these pieces of music together takes away their individuality, their fingerprint.
AI is only able to operate based on patterns, and music is more than a pattern. Raw music is like a flowing stream, with no solid structure, but it is entirely connected.
Raw music is like water, until passed through the human brain and placed into a structure. Think of water being forced through a maze. It will take shape of its container, but without a container it is a shapeless.
AI can simulate a container, but it will never understand the flow of raw music. It can simulate patterns, but will never be able to "hear" the music that it's creating with them. AI makes music through sight of patterns, whereas humans make music through sounds and patterns.
Music is neurologically very complex, requiring our logic, emotion, planning etc. An equally complex AI will be required to compose
I disagree.
Disagree with what?
I actually did develop a project where AI would try to create music it's own, using Long Short term memory(LSTM) over traditional RNN gave the best results but to actually make machines develop music which would actually qualify as 'Music' per se, in simple terms we have to feed them actual music and then the computer will create it by rearranging it in coherent manner. Using general patterns and feeding it rules relate to music and hoping it will identify patterns lead to their creations sound like a drunken hobo on a piano.
The music at 7:52 sounds a bit like Phillip Glass.
I'm a student programmer and composer and I can almost definitively say that AI will _eventually_ be able to do anything as good as a human can. Our minds are only a set of processes in the end, just highly complex ones that are influenced by not only one art form, but all of them at once. It'll take a while before it can effectively compose music that people like and is as complex as a human can compose though.
One of the problems with the large companies like Google who just feed AI lots and lots of Bach's music, for example, is that they don't take into account the process of writing music at all. They just feed it complete music and get poor imitations of that music without much structure. That method will work eventually if enough data is given and enough personal feedback whether something is good or not is given to the AI. But Bach didn't just blindly write notes down on a paper. He went through a general process: melody, then harmony, then orchestration. AI first needs to be trained how to do those steps individually. Then, it must learn how to blur the lines between those steps. It has to be able to alter its melody to fit the harmony at times, and to alter the harmony and melody to fit the orchestration, and, most importantly, while composing a melody, it must keep in mind that harmony and orchestration will come down the line. It is an immensely complex process to train an AI to do this, but it's completely feasible with enough time.
For now, what we should be working towards is not getting the AI to completely take over the process, but to alleviate potentially annoying tasks, such as orchestration. If you have a sound in mind, feed it an audio clip from some other piece and it'll orchestrate your sketches like that piece. That would be extremely useful.
But it can never orchestrate something exactly how you want it. Adding a fitting little details will be probably a most complicated thing in training AI.
And also i can't imagine how AI can be used for scoring to films and games. You will need kind of software with DAW functions where you can type what you want to hear in certain areas (every 3 seconds - something different) including usual options like mood, speed etc. But even this way i really don't see how it can create something that really fits to the picture plus includes character themes and stuff to the piece.
@@FireF1y644 An AI can't orchestrate something exactly how you want it if you already have a set plan. In that case, it's just more efficient to orchestrate your music yourself obviously. But if you don't know exactly what you want to do with the orchestration (which is the case a lot of the time), but you know you think the orchestration of some other pieces might fit what you are going for, you can feed those into a program with your sketch, and it'll spit out an orchestrated composition, or you can just feed it the sketch, and a well trained neural network (basically an AI's brain) will be able to find stuff that it thinks will fit.
It's important to point out that there are many ways of training a neural network as well. You can train it to mimic something as close as possible by feeding it a lot of data (such as in this example: ruclips.net/video/Ipi40cb_RsI/видео.html ), you can train it to figure out how to do something all on it's own by giving it a reward system so it prefers when it gets closer to a goal you have in mind (such as in this example: ruclips.net/video/Tnu4O_xEmVk/видео.html ), and you can mix any training methods together to create a hybrid method.
This "hybrid method" form of training is what will allow AI to compose its own original material. You can feed it a set of a bunch of music (such as a human listens to a lot of music), and give it a reward when it pleases you. Then, once it's learned how to compose like other composers, you can turn up the favorability of experimental neural pathways so it experiments. You'd then give it a greater reward when it succeeds so in the future, it'll prioritize original material that sounds "good."
By using this type of hybrid training, you can get it to decide on it's own what "fitting details" it wants to add, or you can ask it to mimic the "fitting details" of other composers.
No AI will ever compose exactly like you, just as no human will ever compose exactly like Beethoven. But AI can learn to prioritize its own take on music genres and it'll develop its own personality.
With enough time, data, and a well structured rewards system, AI can brute force its way into learning how to do anything. If you feed an AI every John Williams score and reward it when it learns what a character is, what that character represents, what a theme is, what that theme represents (ie. evil, mystery, etc.) and when associates the right themes with a character with the correct mood, you will _eventually_ get an AI that composes for film like John Williams does. But that will take years of processing time and manual reward gifting. This approach, on a smaller scale, is what companies like Google have done for their AI composing projects.
There are, however, much better ways to teach an AI how to score a film or a game. But you have to teach it like a human. For film scoring, you'd first have to teach it the basic things that are built into humans genetically such as the ability to recognize a person, place, thing, and eventually an abstract thing like an idea. Then you have to teach it how to recognize mood, ideas, characters, etc. in a film. Then you have to put it through music school. You have to go lesson by lesson. Teach it what a scale is, then test it on what a scale is. Teach it what a melody is, then teach it how to compose a melody. Etc. etc. Eventually you can teach it to put its knowledge on humans and music together and it will be able to create its own score for a film based on its music preferences and how it perceives the mood of a film. This will happen eventually and there's no way any of us can stop it, but it will take a long time for a project like this to be completed.
The more reasonable short-term applications for AI in music is for it to replace individual steps. Orchestration is likely the first step because the worst an AI can do is orchestrate something by a set of rules with no creativity. Then composing melodies, understanding form, developing melodies, and being able to take into consideration all those steps at once will follow.
Our brain is just a very complex neural network. The work of AI will eventually be able to be indistinguishable from the work of a human. Star Trek had it right in the 80s: we will eventually (likely within the next century) be able to create something along the lines of a "positronic brain" like commander Data has.
@@AnnasVirtual Dont try to impress me, I have seen these and much more thousands of times. AI can not do complex, meaningful, original and contextual music right now.
@@AnnasVirtual that's true, sometimes you can even get nice ideas from generated stuff
0:46 Please tell me the link to this software please.. please.....
Hello guys. Anyone knows the chart that shows western music history in a timeline shown at 9:32?
Great topic!!! As other pals have pointed out, David Cope's software (Emmy) did a pretty good job on this regards. It (or she?) was trained on known music pieces and was able to perfectly recreate the style of the original composer, yet the compositions sounded absolutely novel. Cope and Douglas Hofstadter also organized challenges where finally musicians were unable to distinguish which composition was computer created. And this happened decades ago. We are in a phase of rediscovery in this field, but I have not yet heard, among the many existing self-defined AI programs, one which is able to come even close to the results obtained by Cope's piece of software. Some compositions could also be found here on RUclips (search for "david cope emmy").
One day, there will be frustration among a generation of musicians, when the world prefers the machine to play, and humans as we presently know it are a thing of the past
The day it does
will be a dark day for all creatives
THANK YOU!!!!!
how? by programming it NOT TO compose like any piece he learned, come up with something new
The answer to your question of the name of this video is yes it can make good music but I was asking can you also remove the lead vocal so that way you only hear the backup voices only with AI separation?
today there are AI programs which compose really good, realistic music ; just don't tell people in advance it was created by AI otherwise they'll purposely force themselves to "not like it" ; and they won't be able to tell it was not written by a human
18:06 even if computers and AI could make perfect art, it wouldn't stop humans to create their own art. I don't see why computers making music would make composers go jobless. You can have too many taxi cab drivers; but you cannot have too many art
gosh the end............ beautiful
The soul to an art is not given to it by it's creator but rather by the observer or the person consuming the art...
World record of how fast you can get demonetized lol