Why AI is Doomed to Fail the Musical Turing Test
AI will get vibed at the jam session, and there's nothing that it can about it.
🎹 Get Nebula using my link for 40% off an annual subscription: go.nebula.tv/adamneely
🧠 Watch Tibees video on the original Turing Paper: nebula.tv/videos/tibees-the-o...
🚀Get lifetime access to Nebula for $300: go.nebula.tv/lifetime?ref=ada...
0:00 Intro
3:54 Part I - Musical Turing Tests
10:56 Part II - Thinking Like a Human
20:17 Part III - "Not music"
Sources:
tinyurl.com/5n85k5bk
Valerio Velardo's channel on AI music.
/ valeriovelardothesound...
(⌐■_■)
⦿ Adam Neely T-shirts! ⦿
teespring.com/stores/adam-nee...
⦿ SUPPORT ME ON PATREON ⦿
/ adamneely
⦿ FOLLOW ME ON THE INTERNETS ⦿
/ adamneely
/ its_adamneely
⦿ Check out some of my music ⦿
sungazermusic.bandcamp.com
insideoutsidemusic.bandcamp.com
adamneelymusic.bandcamp.com
Peace,
Adam
I am a computer scientist, and the category error problem constantly annoys me. We find a problem that requires a lot of intelligence in humans, like playing chess or go at a grandmaster level, and declare that AI is therefore "intelligent". For some reason, it's only AI that we use this kind of language about. The best human weightlifter is easily outcompeted by a small forklift, but we don't call the forklift "strong". The best human sprinter is outcompeted by a locomotive, but we don't call the train "fit". Hell, computers have been beating humans at mental arithmetic for ages, and that's even a marker of human intelligence. To quote the great computer scientist Edsger Dijkstra, "The question of whether a computer can think is no more interesting than the question of whether a submarine can swim." One of the under-appreciated aspects of the Turing test is that it was an activity that humans should find easy, not something like playing chess which humans find difficult. It's these "easy for us" problems where AI tends to fail, partly because they are the problems that machines find very hard, and partly because you can't get money to solve unspectacular problems. I want a machine to do things that I find easy but tedious, like cleaning my bathroom.
Sure, I'd like a machine to do the tedious stuff like cleaning. However, I'd also want a machine to cure diseases and make nuclear fusion viable, for example. If a machine can solve hard problems and make life on Earth better, I'm all for it. Also, I don't really like that quote. To me it's very interesting to ponder the question whether a computer can think. It made me contemplate about what thinking actually is and in what way our thinking is different than how an LLM outputs text. What's so special about us that would make this question uninteresting? Not much, me thinks. In the end, we are also just machines, albeit very complex ones.
@@bobrandom5545Prof Roger Penrose has proven that when the human mind understands something, it’s not running algorithms. But running algorithms is all computers can do. Whether computers can think depends on how you define thinking, I guess, but I think there will always be modes of thought that we don’t know how to make computers do because we have no idea how we’re doing them.
@@richardhunt809 Where did he prove that? As far as I know the prevailing stance in neuroscience is that our brains do work algorithmically.
@@richardhunt809 He didn't prove anything. Don't get carried away. Keep in mind that his theory about microtubules is an extremely fringe theory, barely considered as more than pseudoscience by most of the scientific community.
Yeah I don’t mean the microtubules thing. That’s more to do with Stuart Hammeroff anyway. I’m talking about Penrose’s mathematical argument in his book Shadows of the Mind. It doesn’t get into neuroscience.
0:00 "What would it take for a machine to jam?" Very little, actually, my printer jams all the time.
lol for a second I thought he was gonna talk about the halting problem
Best Comment
*brrrt* FACE MELTER
You dirty dog
PC load letter? The fuck does that mean?
This is so interesting. And such an honor to be featured! I feel like you and Jack could nerd out for days on AI (and music).
Yo, keep being cool Pomplamoose
you guys are super talented, please keep doing what you do
God y’all are so sick, such an enjoyable group of people
The funny thing is that I discovered your music yesterday. Oddly timed!!! Great stuff by the way!
hi pompalmoose
Adam!! Drive-by praise during the sponsor read, I was not prepared!
Adam Neely jumpscare
Well deserved dude! Adam’s got great taste clearly haha
Jacob you won't read this comment but your essays really are the best
I think the Red Lobster thing isn't proof that AI is now good at producing music, but rather an indictment of how bad music in advertising is.
Also how willing companies are to not pay any money
this comment is partially missing the point. the argument isn't just that AI music will never reach a level where it's considered "good" (whatever that means) by most people's standards (not just those who listen to the top 100 pop songs on spotify). the main problem with the Red Lobster AI song is that the inputs into producing the final product is of no importance to 99% of people who consume music. (If the Red Lobster AI song was actually good, would this completely eliminate your criticism of AI music?) replace the Red Lobster song with the newest drake or taylor swift song, the point still stands. nobody cares about the actual action of music, just the final product, which will be what allows these Red Lobster AI songs and more to proliferate so easily and readily.
AI is the new CGI - if capitalism can lower standards enough, people won't notice or care about its use in some low quality contexts. The decline in superhero movies is noticeable as they've been able to get away with more and more "film some people in front of a green screen, generate everything else in post." They drip-fed it to popcorn audiences until expectations were lowered appropriately. But there are still realms in film where you can't get away with that - Monkey Man still requires some belief on the audience's part that the action involves real people fighting, The Lighthouse couldn't get away with Willem Dafoe in a green screen suit and filling in the scene around him. AI can replace already pretty bad or forgettable music - commercials, grocery store radio. Getting to the next level where it can replace some one hit wonder Top 40 might be possible, but the step beyond that to creating the connection Swifties feel with Taylor or a Nick Cave audience feels with him I don't see that as possible in any universe.
@@mvsr990 I like the superhero movie allegory, because predictably, people are getting sick of that as well
AI really will be perfect for corporate music. soulless, lifeless, is as obvious as obvious can be in telling you what to feel, but hey it'll do for what it's meant to do
Things that will stop AI at jam sessions: “the usual key”
Until there going to be trained
Adam Neely's Blues sessions often explore various keys typical in blues music, including but not limited to E, A, and Bb. These keys are commonly used in blues to accommodate guitarists' preferences and the range of typical blues instruments like guitars and harmonicas.
Or something like, "MAKE IT BANGIN'!" See, "bangin'" is slang term with a largely subjective meaning. What you find bangin', I may find to be a snooze. And visa versa. Due to the subjective nature of a lot of musical slang terms, the AI would only be able to adapt to the situations for which it was coded. E.g.: "LET US HAVE A BANGIN' JAZZ RIFF-OFF!" gives the AI 3 directives: 1) Play Jazz. 2) Make it energetic and exciting. 3) Use riffs as the basis of the jam session. If you coded it to handle all 3 of these, then cool. If not, it's going to try and handle it, BADLY.
Or calling standards like Twinbay or Tingl
Ah, gotcha! In that case, let's jam in the key of G major. It's a pretty versatile key, great for both upbeat and mellow vibes. What instrument do you have in mind?
I’m about to start randomly yelling out “BLUES IN E FLAT” At the jam from now on. I’m sure everyone will love that.
“Ah 1 and ah 2 and ah 3” *starts ripping Phrygian dominate bass solo* “Did you guys hear that flat 2nd i was playing”
gonna show up to the rap battle and shout "BLUES IN E FLAT"
It’s an oldie where I’m from
"Sir, this is a funeral service."
It’s a good way to check for robots.
Imagine having to prove you're not a robot by playing free bird on a fucking hurdy gurdy
No, that ability would define AI over human skill.
One might even call it a Touring Test
ha HA!
Yes, no AI would be insane enough😅
The AI next frontier, dad jokes
Damn, that's good =D
@@BloodHassassin Write a dad joke about AI music: Why did the AI musician go to school? Because it wanted to learn the "al-gorithm" to hit all the right "notes"!
It doesn't need to pass a musical turing test to anihilate 90% of media composer jobs.
What a shame. We wanted technology to automate the boring stuff for us, like factory jobs, and give us time to make art and music, and instead we’re all working and the machines are making art.
It's horrifying, it should be illegal. There's no way to imagine any real social benefit behind this door they're unlocking. You can't even frame it in the context of "if we don't the Chinese will." Let them - who cares? AI generated music is not of some immense strategic value. The techies and Silicon Valley have betrayed us all to enrich themselves.
@@nickcalabrese4829 as if it didn't massively affect those either
Yeah. But where I live there is an significant uptick in audience of live performances. I hope there will be a revival of analog life styles as soon as people get overfed by fast (music)food, fast everything. I am not holding my breath though... Most folks want cheap and fast.
@@nickcalabrese4829it is. You just don’t see that part
I really very much like the concept of "musicking", it reminds me of a quote from someone I very much respect a few years back, they were talking about NFTs and the commoditisation of the visual medium, but this hit me so much I don't think I'll ever forget it: "Art is a verb. It's a process. It's an act of communion. What hangs on the wall is a fucking collectible. What you and the artist communicate across centuries is the art." Obviously Small's book and concept well predate this quote, but it honestly changed my perspective on why art and music mean so much to me, and what's truly valuable and meaningful to me as someone who appreciates these things.
I like to define art as a form of communication, which annoys AI art bros to no end because, well... AI art doesn't meaningfully communicate anything because AI doesn't communicate. AI only imitates the act of communication but it's imitation has no parseable information inherent in it. If a Massive Language Model outputs a message telling me tells me it hates my guts I know I'm okay because the AI cannot hate, doesn't know what guts are or who I am and it was prompted to act like an asshole anyway. The output is, functionally, not communication. The art as communication thing also keeps the term "art" useful. If anything I can experience or perceive anywhere from anything, even the natural world is art then that makes the term art so vague to be not practical. Limiting art to communication, preferably communication with meaningful information (so as to not concern ourselves with the question if the stars are alive and are communicating with us through secret signals in how they twinkle in the night sky) keeps it broad enough to encompass most human expression but limited enough to still be useful. Perhaps in the future a true general AI can overcome these hurdles but we're a long ways away from any of that.
I gotta say: I'm glad no human being was forced to record those Red Lobster lines...
I always forget that the guy from pomplamoose is the one who invented patreon, and every time I'm reminded of it I'm blown away all over again
the same patreon that is closing accounts of Ukrainian bloggers, because "we're all about peace and love, so the fact that you guys are getting screwed over by russians is kind of ruining our hippy vibes, man".
And Patreon's competitor that got bought out by patreon, Subbable, was invented by Hank Green (the vlogbrothers/SciShow guy).
@@lifeteen2also the vidcon guy and the microcosmos guy and the crash course guy and th
Someone clearly hasn't been jamming out to 'I Glued My Balls to My Butthole Again' as much as we all are.
Ahhh yeahhhhh boyyyyy-eeee. 'What happened to my schlong' is fire too.
You son-of-a-bitch. You KNEW I'd look that up, and you knew damn well that I'd be singing it tomorrow.
Absolute banger
Stuff like that is made using AI, not by AI.
Thanks for sharing this masterpiece, now I can’t stop listening to it 😭🙏
Professor of computer science & amateur musician from the Netherlands here. First, thanks for all the thorough and well-researched videos, always a joy to watch. "AI cannot do X" arguments are, in my opinion, always tricky: AI has surprised all of us, even researchers in the field, with its incredible progress. In particular, I am not convinced about the interaction argument. Reinforcement learning is branch of AI that is specifically tailored to interacting agents learning behavior in a dynamic environment. Amazing progress by companies like Boston Dynamics has enabled robots run and do back flips. I see no reason why in several years this technology would not be able AIs to play in jam sessions. Sure, there are challenges, like there were in AI before. So the real question is how should we musicians, writers, scientists and all other relate ourselves to AI? This video makes a good contribution to that debate and the various aspects.
AI will never do X arguments always make me cringe.
Yeah, Adam is a great resource when it comes to musical knowledge but he’s out of his depth here, and just flat out wrong. If there was any money in doing so, I’m fairly certain that one of the big AI labs could train a model _right now_ that could pass the musical Turing test. Let alone many years into the future.
Very true
You're absolutely right! I've been involved in a cancer reseaerch lab which was training an AI to do the pathologists' job. Back in 2020, their "current" version after two years of coding was slightly better than a panel of 12 pathologists with 20 years experience in the field each in the first world (they were from different countries). That made the software better than 99% of pathologists world wide at cancer detection and determination what kind of cancer it was and how to treat it. Same in drug development, ALL molecules are first designed by software to find out if they'd work in theory and only those with the highest potential are created in real life for further up studies. We tend to say that 1 in 10K potential drugs are followed up in a lab, and from those, another 1 in 10K make it to the market. Software again does the bulk of the work and that with companies actively keeping their work secret and not helping each other much in fear of losing profits.
OpenAI just released a video of their new AI singing a song, harmonizing and rhyming with another instance of itself, you should check it out.
The soccer analogy is the best way I’ve heard to show the difference between the end product and the process/creators behind it. An AI could very feasibly be trained to generate HD video of a full sports game. But I’d be very interested to ask sports fans if they’d be invested in the outcome of a game that wasn’t physically played and where none of the players are real people. Would they watch excited to find out who will win, worry about their team making a mistake, believing that any outcome is possible? Would they wear the jersey of their favourite AI player? Art isn’t art without artists the same way that sport isn’t sport without athletes.
People get invested in stories that don't physically happen all the time, or we wouldn't have novels or summer blockbusters.
@@JonBrase I'm tempted to say that in the case of movies or novels, people (as you say) are drawn in by the underlying story, whether it be grounded in reality or not. They attach emotion and value to certain moving story arcs, which can be often-used tropes or new, unique plots. As for sports, people tend to be strongly drawn in by the underlying human aspects: athletes pushing their boundaries, teams 'clutching' against all odds, sportsmen delivering a stellar performance despite injury. Perhaps most important of all: the prior expectations weighed against the actual outcome of a match or contest. A match played by a completely dominating soccer team against a significantly inferior one will not be engaging at all, because the outcome will in most cases follow expectation. Hence such matches, when played by machines, will not have the same unpredictability (and are in fact much more deterministic because of the fixed capacities of the machinery).
@@boristerbeek319true, but its a different kind of spectacle, human chess players have been outmatched by AI long time ago, but people still watch humans play chess as it's more about the human abilities against one another, rather than as a whole, since as a whole even the best human players lose to the AI. But if AI can create an interesting story in that match, then maybe some people will. It's true that people get invested in things that aren't real all the time, we usually call that art.
@@boristerbeek319 so do we think music is closer to novels/movies where the end product can be engaging without knowledge of a creator? Or is music closer to sports where the human beings behind it provide most of what we love about it? I think this would inform us on if AI music could really capture attention as art in itself or if it’ll never surpass marketing/elevator music because we need the story of an artist behind the art.
@@TheManinBlack9054 This raises another extremely pertinent point: AI chess has become its own category, and millions of people love to watch it. It's not about the story, but about the technical precision and how far AI can push the bounds of perfection which makes it so interesting to watch.
Hi Adam, this is one of the best "AI will never be able to do " argument I've ever seen ! For context, my current job *is* about GenAI (for instance I know exactly the ElevenLabs settings you used for Xenakis), and I started doing machine learning 20 years ago. So I've heard my share of silly "AI will never be able to do " arguments, that proved to be wrong in the end. I'm also an amateur musician who knows what jamming is, and the very subtle body language communication it can involve. I was very skeptical at the beginning of the video, but this is maybe the first time I hear such a compelling argument against AI taking the place of humans. Two thoughts to slightly moderate your argument : * I do think that machines will be at some point able to pass your musical Turing test, but it might require a big paradigm shift in architecture, training procedure, etc. (a shift that will not likely happen soon for the reasons you describe around economic incentives). * In a sense I feel that that Turing test is a bit unfair in the sense that it involves doing a thing well "with humans", and not "between AIs". It is like saying "Oh but humans will never pass the spider test which consists in being able to get offsprings with spiders". I'm sure that AIs will be able to jam "between them" because they will find digital ways to replace the subtle body language and so on. (It reminds me the "Hanabi" card game test : it is a cooperative game that AI can do very well when they are playing together, but do badly if they try to play with humans.) Keep up the good work !
Good points! Yes I felt Adam framed the argument strangely by giving AI a test it’s not even trying to solve (as far as I know). Generative AI music is just that: it’s all about the output, not the performance. Even IF we got to a point where we had robot bands…well, it’s utterly pointless, right? Performative AI is extremely pointless, and solves no problem - other than being a novelty for 5 minutes. Generative AI however DOES solve problems in that it can create things like ad / stock music very quickly. I’ve used Udio to “discover” new music and I’m finding it very fun. Of course it’s way less creative than building a track in Ableton but then Udio can be used to create only vocals (for example) so I could use it as another tool in the toolbox.
I don't think it'll take a paradigm shift in the way the technology is built. I'm pretty sure a transformer model with enough context would work. But I speculate that we're more likely to accidentally build something that can do it before any of the major players find it worth their time to intentionally do it.
@@codahighland Transformers are not realtime and none of them currently work with a context of other music + human spatiotemporal posture. So by that measure, there will indeed need to be architectural changes.
Sorry, but I don't get your point. If you say that you _do_ think that machines will be able to pass the musical Turing test, how do you agree with him at all?
@@ldlework Transformers are as realtime as you want them to be; it's not an architectural limitation, just a user interface one. There are definitely some transformer models that have been used for music already. And adding additional data inputs isn't an architectural change. There's nothing fundamentally new that would necessarily need to be invented; it's just a matter of putting together the pieces with the necessary intent and collecting enough data.
As a researcher on social interaction (including social and distributed cognition), I love that you did a deep dive into cognition beyond the computational model that is so prevalent in everyone's imagination at this point. I hope folks learned something from it. (I'm also a little jealous that you explained so clearly in a few minutes what took me years of PhD study to understand. Bravo!)
interesting, what field are you in? thanks
Around 25 years ago, I attended an informal live musical Turing test (the Output kind) hosted by Douglas Hosftadter at the University of Rochester, co-sponsored by the Computer Science department and the Eastman School of Music. Hofstadter was showcasing what was then a state-of-the-art generative music composition AI. I'm pretty sure it was a version of EMI by David Cope, but my memory of the event has faded with time and details are scarce on the internet. I've found articles about Hofstadter's work with music AI around this time and articles about David Cope, but nothing about any live experiments or presentations like the one I remember attending. The key element was that the AI was trained on a corpus of a real composers work; a large set of similar compositions (I think Bach's chorales). The idea being that there were enough that even the musically literate in the audience wouldn't know each one by ear. A skilled live pianist played us half a dozen pieces, some from the original set and some from the software, and the room voted which we thought were originals and which were computer generated. The audience was a split between people there because of the music connection and people there because of the computer science connection., and did merely an OK job identifying the real compositions. At least one of the computer generated pieces was obvious, but several were good enough to split the room, and at least one was only differentiated because some of the room knew enough of the chorales to know it wasn't one of them. Earlier in his career, Hofstadter made a list of ten things he thought computers and artificial intelligence could never do, and one of the things he said at this presentation was that the work he was doing had led him to cross "create beautiful music" off of that list. And this was 25 years ago. I think that one day, computers will be able to jam with you. I agree with most of the video about why it will be hard, and why there isn't money in developing it. I don't think it will be easy, or it will happen soon. But I think that eventually, after one or two or three more generational paradigm shifts, the ability to do so will fall out of some other profound advancement practically for free, like how so many basically economically worthless but really fun roleplaying and other conversational abilities have been achieved for free with the profound paradigm shift of the GPT-based chatbot advancement.
That's why i would advocate for thr strict regulation of a.i. The arts should be exclusively reserved for mankind. A basic right like the right to clean water, food and shelter. I see no reason why deserve to be outcompeted by the power-hungry, economy-destroying toys of the tech giants.
Transformers are really good at one thing and that's pretending to be something. I'm convinced that if ample training data of jam sessions, with specific instruments removed and added to the mix (using simple music editing programs) is provided to an AI, we could have a real time musical jam AI, similar to how we have real time voice changing AI now.
We absolutely could. Adam is a great resource when it comes to musical knowledge, but he was out of his depth in this video. He said he’s pretty sure the music Turing test will never be passed. In reality, if there was any money in doing so one of the large tech companies could probably train a model to pass the test right now.
Cybertronic Spree?
I think we actually will see software that can "jam". There isn't necessarily a "ghost in the machine", however. What we'll be getting is a reflection of the spirit of the people it's playing along with. If we fail to recognize that we're dealing with a sophisticated mirror, then it'll seem to have a spirit of its own. We're still probably going to be flooded by algorithmic music though, and "artists" that can do a great job pretending to play for "live performances". This is just a continuation of formulaic pop. My guess is that we'll get some genuinely novel creativity on the periphery which draws on AI, in the same way that digital audio has introduced entirely new genres. What I'm really curious about when it comes to generative AI art is whether the masses will actually get tired of it. If they've got a constant supply of generated pop-on-demand, I could imagine that there is a threshold where it's "too good", and they lose interest. Perhaps the "AI music revolution" will result in a retro movement, back to in-person performances where we can easily see that real humans are playing together.
You've made me recall how MTV was riding the "Unplugged" wave for a bit in the early '90s. It was a minor badge of honour for artists to be capable of playing their sets sans electric guitars/studio beats, as a contrast to the distortion and synthesizer heavy 80s. Maybe the next wave won't even be online at all!
Yes, for sure, the cheaper something is to create, the less value it has. When we can mass-produce any musical experience we want just by asking for it, it turns the experience into something disposable. Music (mainstream pop music) has been on that trajectory for a long time now. Every big hit is a flash in the pan, and people forget it as quickly as they embraced it.
I spent something like a week trying to get good tracks from Udio, and at first it was exciting to find something good. But with time it became pretty exhausting to listen bad take after bad take in search of something I deemed acceptable so when I finally had a finished track I was so tired of it, just listening to it made me feel nauseous.
After a while, AI generated music will all sound very generic, typical and sound similar to each other. What AI does is mix and match existing music that is out there, but it is unable to reinvent or improvise/create.
amazing comment. made me think of a quote from brian eno, "Whatever you now find weird, ugly, uncomfortable and nasty about a new medium will surely become its signature" i feel like the same way generations now are getting back into vinyl because of the vinyl distortion and scratches will be how people in an age of AI music will look back on more "primitive" and "imperfect" creations of music.
I don't understand why pop-science AI media has focused singularly on art. It is, especially in the context of music, one of the few situations in life where the human performing it is as, or more, important than the art being performed. "Another box to tick" is a good way of phrasing how many companies think about this problem. It is a briefly entertaining imitation of a human experience.
Probably your algo, if i speak with my cousin about ai and art, she often mentions drawing/design/visual arts
AI pop science is focused of new, impactful capabilities of AI, diffusion models have just now arrived at a point where they can reliably produce media, like text images and music. once ai systems can do more impressive and impactful things ai image generation will be boring old news.
Because it gets the clicks, is my guess. There's so much more important and usefull stuff AI is doing like helping develop new medications or researching material science but no one is gonna be outraged by that
Kurt Vonnegut once said something like, “Some of the important things in life are food, sex, but also a philosophical idea called becoming. This is why I don’t like computers in the home. You need to become. And Bill Gates is saying you don’t have to. Your computer can do it for you.” Part of his idea of how a person becomes was the creative process. He spoke often about how people need to write stories and poems for each other without caring whether or not they “make it”-it’s just a human thing to do, same as jamming with friends or playing sports or having a conversation. You need a creative outlet. I am totally convinced that machines cannot make music or art or literature. They can make things that look and sound like it, but they can’t make it because to make it you have to be a human and you have to mean something, and machines can’t mean.
While AI art might be ruining parts of art, it is still a fantastic tool to further the ability of artists, much like a sample library can limit the exposure of instrumental players, but has made way for so much creativity for artists in music like rap. We see that professional artists often have apprentices prep canvases with a simple or even somewhat detailed background that may not be the subject of a painting. It's quite amazing that an artist can prompt an AI to do that dirty work for them, it gives more time to spend on the parts that they care about. In fact, I view AI in the same way for almost all applications. People might say that engineers graduating on chatgpt aren't good to hire, but I heavily utilize chatgpt so that when I am approaching a problem that involves information I am not brushed up on, I can ask chatgpt to streamline the info relevant to the problem. Before I would have to find the topic, then backtrace through textbooks until I find the necessary concepts I needed to brush up on. It's just a tool after all. Every tool merely has the purpose of making a job easier, whether that be going from a rock to a sharp rock as our ancestors did, or from google searches to AI prompts.
"red lobster used ai for an ad" is very unconvincing when advertisement music can hardly be considered 'music' most of hte time
Great video, Adam. Love how you pull together a few different scholarly ways in to the question
What would it take for a machine to jam? As an IT tech, frighteningly little...
Without human input it would be impossible. The reason "AI" alone exists is due to human input.
@@lovescarguitarthey were making a printer joke
I think that i could work like a Transformer maybe... Supose that you have an AI that given a Input plays something... Train it with a lot of jams, and then just give the Model some pseudo random variables to start the song, a little clip makes the deal or some ramdom generated thing, and then Just kept the model listening for some seconds, then let him play and train the model i a way that it plays things that are close to the space where the input lands, in the same way a transformer predicts the next word, let it predict the next notes... and there you go. You can add a randomness value in order to make it more predictable or crazy, just like you do with Transformers
Yeah I think this possibility is a lot closer than Adam thinks. I think the only reason generative AI can't "jam" is simply because no one has tried to develop such a thing yet. All the pieces are there. I've heard some pretty good AI generated sax solos over Giant Steps. Adding the ability to visually and auditorily observe and respond to cues is just a matter of integration.
Ehehehehe, that printer joke is gold.
I think I’d fail it too tbh
God damn Bots posting in the comments section.
😂 x2
Then its a horrible test. no?
Sure but you know you are not a machine so its ok 😂
@@GarryLarryBarry high quality comedy right there 😂
great video! thank you for talking about losing the 1, is very comforting to not be alone on this
Lots to say. Adam, your videos always push my brain to it's limits and that's why your my favorite. I'm glad you recommended a bunch of other videos and creators as you could never make enough videos yourself to quench my mind thirst.
I think the line about making music a problem to solve is what really irks me. This AI push feels like the final stage of reducing all art to a product, to something to sell or show off for status rather than appreciate in any meaningful sense. I hope it all burns out, but I'm uncomfortable with the environment around it.
Why can't we have both? Like maybe ~10% of my musical time is playing the guitar, checking info, videos or playthroughs from the artists I like (or even watching a video like this one about the meta of music). But the other ~90% of the time, I just listen to music. That's it. And I don't see why I should care that machine learning was involved. I don't believe in souls and ghosts and magic. I think we are biological machines bound the laws of physics. So all those arguments about ai-art being soulless mean absolutely nothing to me. souls don't exist to begin with...
exactly this when these models first began rolling out, i was hopeful about their potential as a tool. something that could lay down a rough foundation for a song or an image or a story but that hope is long gone. they are not tools, or at least are not being used as tools - they are being used to mass produce superficially pretty at best but ultimately garbage 'art' that is pure commodity. it is so clearly and purely the 5% wanting a second go after their NFT project imploded and died can art produced by a large language model have meaning? maybe, but there's not even a genuine effort to do that right now, and certainly all of this 'art' produced for the sake of selling it alone will never mean anything
We're going to burn out and become uncomfortable with the environment, as vulture capital and pirate equity set billions of dollars and the literal atmosphere on fire, to power these large language models that will never produce a real ROI.
when you see AI evangelists discuss their vision of the future, it's obvious they view all media, not just music, as passing, pleasurable aesthetics to briefly consume before discarding. their utopian future is one where all media is slop designed to appeal to the lizard brain with no further thought put into it. nothing unique, nothing interesting, no consideration of connection to and of the artist through the music, no auteurism, etc. their goal is reduce the production of media to a method that appeals to the most base art consumer and nothing more.
@@hastesoldatour souls are our experience as humans. Which is entirely unique to each one of us. A machine doesn't have that so anything it makes will be hollow.
Adam's Nebula pitch convinced me to join when i remembered that i'm already a nebula member
Sponsor read so good it made you want to join nebula twice
That's pretty damning as a review. "So forgettable I forgot I had a membership" lol
Love this! Thanks for another amazing video Adam! So thankful to have someone representing this perspective in a landscape dominated by the types of people you've quoted here. We're very grateful!
Seriously great video Adam, I'm a grad student in computer science and musician/composer doing my thesis on generative music right now. I think an important thing to bring up too is the difference between generative systems being built today and those built by the original pioneers of 'generative music' like Brian Eno and David Cope from the 1960s-1980s. I'd argue that a lot of those older processes and systems thought more 'human-like,' in that Eno, Cope and others tried to make systems that extended their own compositional thought process (Music for Airports from Eno, It's Gonna Rain from Steve Reich, EMI, etc...). But today, that compositional curiosity is gone, and companies like Suno, Meta, Google, and others are looking at music like the next 'check box' for AI and trying to achieve greater technical prowess.
A few weeks ago I tried out the Udio AI. The more specific you are, the more impressive it is. Kind of fun. I asked it to make a funky jazz fusion track with heavy use of B3 and interesting polymeter. And it made a pretty cool song! But I forgot to select "instrumental" so it had some James Brown type voice shouting excitedly about organs and time signatures.
So what you're saying is, the AI can't actually pass the Musical Directive Test, but it can spit out bad imitations of James Brown tunes? 🤔
4! By 4! Yeahh yeahh
I don't know. I had an opposite expirience a few days ago. Like I asked a very specific thing - a violin solo. Generated 3 times, got six results. Only one had any strings in it at all and far from solo. It seems like it actually works better if you give it broader description on what to do. But then the problem comes that you have far less control over the content. Personally, I'd prefer these AI to output midi/sheet music because it seems like the most useful it can be to actual musicians and more precise music writting if it is used to suggest ideas. It is a common thing to borrow something from others in music (a long list of classical pieces that influenced Star Wars sound track comes to mind) and AI crudly speaking is first and foremost statistics, so AI could be a great tool to suggest you what to borrow. but unfortunally the people who make these AI don't care about music and musicians like Adam said. For them music is only data and business.
Really, I found the opposite to be true. When I try to get specific it gets weird. I have the same issue with image generators. It's like they really want to do their own thing, not what I want. I like weird though so it's all good
@@Grigoriy1996 A violin solo isn't specific enough tho.
Damn it, people cannot tell the difference between real and autotuned vocals. Of course people wont be able to tell if they are listening to AI music.
Because "many people" cannot (or do not care to) distinguish between outputs does not mean that there are some people who can and will. Humans will coevolve with AI, in an arms-race sort of way, and some people will develop heightened recognition of AI patterns. Already there are some who are fooled into thinking that ChatGPT passes the Turing test--not only because of the category error that Adam discusses, but also because their interactions with AI might be unoriginal enough that ChatGPT might actually pass an directive test in which they are one of the agents. Google "the most human human." I'm sure back in the 18th century there were people who could not distinguish Mozart from Salieri. They may have been the people most offended by Beethoven, once his music started being played.
People can at least trained musicians can
Well done - great video - keep it up man!
Loved this video, and of course your jam at the end
I am just completely and utterly terrified by the idea of removing creative people from the process of making art, because then what you are essentially getting is raw output straight from the boardrooms. For things like commercial music, the bad ideas and tastes of people with the money wont even have the buffer of being filtered through some creative individual. Truly, a fresh hell awaits us all.
This might unfortunately happen if big companies capture the space. We need to protect open source at all cost. And combat the push for license requirement on training data.
Boardrooms can be automated, too
truly creative people will not be removed. AI algorithms are imitative, still. They don't yet have the power to invent new styles. they will only replace imitative artists, for now at least.
So it's time to revive the good ol' driveway concerts for the neighborhood. Many of those. With a few tacos you've got something I would pay to attend....
@@goat8477 you cannot go backwards in time, on the macroscopic level.
After trying the likes of Suno and Udio, I'm in a weird mix of "impressed" and "not impressed". By that I mean that I'm impressed at the fact that it's even able to output poorly inspired derivative music, like a human would (only with a much higher energy consumption). But I wouldn't call that "music is solved", though. And that comes from someone who is super interested in (selectively) applying mathematical modeling, AI & algorithmics to music as alternative composition methods. (Loosely related note: try to ask Suno to output something in 7/4. It will output 4/4 but with lyrics _that talk about 7/4_ . It's hilarious)
Are you suuure it takes much more energy? Humans are fairly inefficient in regards to energy, especially depending on ones diet. If you only count the actual creation time (like maybe 5 minutes) its fairly limited, but if you count all the training and education that person has recieved split over all poorly inspired dirivative music he will produce im not sure if the human is more efficient from an energy perspective.
@@mlsasd6494I was about to comment the same thing. Suno almost certainly uses an order of magnitude or more LESS energy than a human would typically take to write the same music.
@@therainman7777 not when you factor in the energy required to train the model in the first place. the long and short of it is that gpu's burn oil. for reference, I recently watched an interview with the zuck where he was discussing the likelihood of new models requiring at least a gigawatt to train, which is comparable to the output of a small nuclear power plant. I’m not sure how much energy went into training suno but I’m sure it’s substantive
@@SYNKSENTURY But then you might enter into the discussion on how much energy was required in order to "train" us to be able to make music. The thing about AI is that we're trying to do what took nature a couple billion years in just a few decades. Of course it's gonna take a shit ton of work and energy to emulate all that process. But once it's trained, it's done, just like us (but even we need to train ourselves over the course of our lifetime)
@@brunoberganholidias5790 hm, i don't think your comparisons are fair. i'm not factoring in the energy consumption history of the computer revolution as a whole, but only how much energy is required to train a single model. so comparing against billions of years of evolution doesn't make sense. we should be comparing how much energy it takes to "train" an individual vs how much it takes to train a model. i did some math to support my intuition, i hope someone will correct me if it's flawed. for simplicity's sake lets just assume static calorie consumption over the first 23 years of life, i.e., the time it takes to grow and earn a degree. that's 2200 calories per day *365 days * 23 years = 18,469,000 total calories. so it takes approximately 18 million calories to grow a brain to maturity and pack it with some specialized knowledge. but we've left out the energy involved in growing and learning that comes from other people. obviously to really model this seriously we'd need a lot of factors, but again for simplicity let's assume a self sufficient nuclear family and a single full time master teacher (rather than calculating, for example, actual number of teachers and "teacher energy" divided by hours spent teaching). so now we have (approx) 18 mil for mom, 18 mil for dad, and 18 mil for the teacher. this works out to 55,407,000 total calories. now we need to convert that to watt hours. we simply multiply the calories by 0.001162 (the ratio of watt hours per 1 calorie) and we get: 64,382 watt-hours. meanwhile LLM's routinely consume over 1K MEGAwatt-hours during training, so we're not even in the same order of magnitude. the expectation is always that highspeed digital computation is not power efficient compared with analog or electro-chemical computation-this is why analog neural chips are being considered for embedded applications, even though they produce so-called mortal models-and the numbers here are consistent with that.
This was an excellent video, Adam! I found myself incredibly engaged and considering how this broadly applies to so much more than just music, although you eloquently sum it up well here for your use case! Mad props, dude 🙂
Found this on the same day I heard GPT4o sing and make up dittys on the fly.
Art (especially music, where the experience is live) is just as much for the person making it as for the people who view it. It’s unhealthy to think about art as purely for the viewing experience when for most if not all artists making art IS the part that’s important. The best artists don’t even care how it’s received.
The best artist don't even release their work. They work for the joy of creating.
I remember browsing KZhead cat videos, and I landed on one about a kitten stumbling, pretty normal, right? The short had annoying music playing, again normal. The 'cameraman' picked the kitten up and I saw something weird going on, the kitten was obviously dubbed over with a stock meow but when I looked at its mouth, a tooth disappeared when it meowed again! I realised it was AI generated. This one short made me realise what's wrong with AI generated content in general. Why is this bad? I watch cat videos not just to watch cat videos, I want to care about the cat because it exists or has existed somewhere, I follow cat channels like claireluvcat for example to know how some of the cats I've come to know are doing right now. AI generated cat videos don't have any of that. Just after swiping back to my home screen I felt a sense of betrayal and anger. Am I overreacting about this? I don't think so. I was emotionally scammed. It's the same for any art.
@@TheManinBlack9054 that's silly! Why would you think this is the case?
@TheManinBlack9054 I would propose that they would release music, and they wouldn't care how it's received. But they could care *that* it's received. They wouldn't care what anyone thinks, as long as people get to hear their unfiltered and authentic expression. It's like a writer or pholosopher who doesn't care if anyone agrees with him, but he absolutely wants people to fully understand what he has to say, and it's their choice whether they reject it. If they never wrote or released music, their thoughts wouldn't be heard or understood.
Blade Runner: Your spouse asks you why you need a Strat and a Tele, because they sound the same. Leon: *Shoots Blade Runner*
Hahhahaha. That was SUPERB! PKD should have called his novel 'Do androids dream of electric guitars?' ;)
"What's a synth?" "You know what a piano is?" "Of course." "Same thing."
“Is this a test of whether I’m a musician or a drummer?”
machines do not sleep, so they'll never pass the music touring test: sleeping on the van
Someone on tumblr said about AI-generated fanfics: "if you couldn't be bothered to write it, then why should I bother to read it?"
Exactly. If you're not enough of a fan to write the fiction, why should I care what you post.
I don't understand. Why should the effort involved in writing change whether I read?
@@DavidSartor0 fanfiction is the most quintessentially community-based space in the world, what do you mean?????? It's literally all about the shared experience of sharing our fantasies and love of the media we consume. If your work wasn't labor of love -- if you didn't love these characters enough to express your own thoughts about them, then why the hell are you writing fanfiction in the first place, and why should I care?? I'm trying to be in a fandom and connect with people who love this media like I do. If you don't, I don't want to read what you write . The standards aren't even high with fic. You can never have written anything before and people will read it and possibly love it. Just have it be your own and have it be sincere. There is no reason to use AI other than apathy and laziness. & I don't want to interact with someone who doesn't care the way I do. Like It's art. It's a social experience. You're not 'consuming content' you're 'engaging with a fandom'. I don't want to engage with a robot's idea of what a fan wanted to write. I wanted to engage with another fan.
@@bzzzzzzzzzz2075 Thank you for answering. I think I've read at least ten million words of fanfiction, and I don't experience that at all. I love talking to fanfiction authors about their work, but the fanfiction is just words to me, words making up sentences making up stories. I don't much feel like I'm engaging with anyone. Do you feel the same way about original fiction? I don't see how "there is no reason to use AI other than apathy and laziness". I'm a very very slow writer. If I were to try writing a long fanfiction, it might take me twenty years. If I were fifty years old, I'd know I might not have that much time before death or dementia. In that case, I'd probably use AI to write faster, even if it resulted in work that people don't like as much. Is this bad? "If you didn't love these characters enough to express your own thoughts about them, then why the hell are you writing fanfiction in the first place?" I've seen stories using only original characters, which were fanfiction because of the setting. Are those bad? What am I doing wrong? How do I engage with an author?
@@DavidSartor0 you're not doing anything wrong lol. I do struggle to believe that the writing is just words to you, though, if you're invested enough in it to specifically engage with the author about it. What do you speak to the author about if they wrote with AI? Their prompting specificity? No, you talk about their interpretations of the character, you complement the lines they wrote, theorize about what's happening next. If you don't care much about what's happening in the work itself, but you do care enough to talk to the author about it, is it really just words to you? Or is it a conversation starter? And does that conversation starter work as well if the work is generated by a predictive algorithm, not created by a human's hands? I'd also like to ask if you have ever had your interpretation of media has ever changed by knowing the authorial intent. Say, you watched a directors cut. Or you read a poem once, thought it meant one thing, then learned the author's tragic backstory and it hit twice as hard on the second read. This happened to me when I read a celebrity poetry book (Halsey) and i learned afterwards that one poem was likely about her miscarriage, and upon re-reading, I found it to be incredibly impactful where I didn't before. An algorithm like an AI can't do that. It doesn't have authorial intent. I can't complement the well-crafted dialogue, or the excellent characterization-- that was all up to a role of the dice. I can't have my interpretation changed because there is nothing to interpret. It's empty because a person didn't make any choices except which words to prompt with. So yes, the same goes for original work and characters. I was just speaking in the context of broader fandom spaces. One of my favorite fic authors has one specific oc in her work, and I love this oc with my entire heart. She needed a character to serve a plot function, then she painstakingly brought this character to life, and made her feel real and fit perfectly in the setting. If she had used an ai to write this character, I wouldnt be complementing it. I would be saying "wow that was lucky that worked so well." I feel like it's doubly egregious to use ai when you're writing a piece of original work(ignoring the ethics of using the world's best plagiarism machine) because those aren't just little guys in the show you like. Those are YOUR little guys that came from YOUR head. Why on earth would you put them in the thoughtless hands of an ai. Look, I'm a slow writer, too(although I got faster wth practice). At the start, it took weeks and weeks to get out maybe 2k low quality words. But I still have that Google doc. all of those words are mine. they're handcrafted and I love them even though they're NOT GOOD. If an ai made it, what connection would I even feel to 'my' work? It's a faulty product, not a passion project. It is bad to use ai to write faster. You're not writing it. If you wrote this comment to me in a timely fashion, you are not so disabled that you need ai to help you. Long fics take ages to write regardless of who's writing them. Remember: that's an entire book you're writing. The great gatsby is 50k words. Long fics get well, well over 100k. Authors regularly spend years and years writing books, 3-10 is the general range, and that's normal. If you 'write' 300k words with ai I will care infinitely less than if you write 1k words with your own keyboard. And frankly, YOU won't like what the ai spits out for you. It wasn't yours, and it wont ever be. You won't have control over it. It wont cater to your specific proclivities. The ai doesn't know what your favorite characters are meant to act like, or how you want them to act. It doesn't even know how they look. It doesn't know anything. You do. You're intelligent and you enjoy media enough to read fic about it. Don't you want to make something of your own? you don't ever have to publish your work to get fulfillment from it. You don't even have to finish it. Most of my work isn't even finished, and it won't be, but I go back and reread it because I LIKE what I wrote, and it matters to me, and I feel satisfied when I look back at that one line that took me 10 edits to get right. Art is wonderful and difficult and worth the effort. Go make something that matters to you. Anything. Good or not. I'm not even talking about writing in particular. Do something that lets you experience the joys of creation.
Personally, I think the main barrier for AIs in music like in acting (all those speach generating AIs) is that not everything can be prompted because not everything can be described in words. We communicate and process information not only verbally. There are a lot of things that we understand via empathy and feelings. E.g. of course, a conductor may give some remarks to players during rehearsals but when performing there are no words and it's not just gestures it is a certain emotional state of the conductor that players read and inturpret not so much consiously and rationally but rather 'empathizing' the conductor. The same goes for directors.
100%. We need more refined tools to interface with the generative AI. Inputs that go way deeper than just prompts.
@@hastesoldat I'm currently trying to push the AI developer to develop midi/sheet music output via "suggest a feature" forms. Because since AI is mainly statistics and music is full of borrowing anyway it could be a cool search engine for snippets of ideas to use in your own writing. Because I think we are very far from this kind of interface that would let us go beyond prompts. Funny thing, it may even turn out that in order to get from AI exactly what we want we may need to write the music actually. It's like with sample libraries. There are very good violin libraries, for example, and in certain contexts just a programmed violin could be enough. But to get a truly realistic performance with all actual details of violin playing fader and breath controllers aren't quite enough. There needs to be a better suited input device for that matter. And this device turns out to be a real violin and the best inputter is a professional violin player. Oops, we made a full circle.
AI's have become multi-modal. Non-textual data can be used as inputs.
@@haomingli6175 yes, but the issue is that we humans have a layer of communication that on one hand works for us fairly decently but on the other is by large unconscious and we ourselves don't exactly know how it works. It just does somehow. For instance, have you ever had a feeling that you know what you want, you may even have a few ideas how to get there but you have totally no idea how to describe it? I guess until AI learns to actually empathize, sort of feel what others feel I don't think the input method will be complete. And I doubt such a feature will come soon.
@@Grigoriy1996 well, just think about the gigantic models that these AIs are. GPT4 has 1.76 trillion parameters. you don't expect these to be interpretable or describable either.
Idk "It's raining bullshit tonight" is quite a banger that I can't stop listening to.
I had a full on existential crisis after hearing "I glued my balls to my butt hole again": catchy song, lyrics that demonstrate a knowledge of toilet humor and human anatomy, 2 live crew lyrics can be repurposed into a 60's motown style song, etc. The robots won't be scary like Terminator, they will distract us with hilarious songs before killing us.
@@60degreelobwedge82I might be wrong but I'm pretty sure the lyrics in most of those (which tbf to me are often the best part) are included in the prompt, i.e. not necessarily AI-generated.
Saw a friend post the arm stuck song the other day and was just shocked that people I know who are anti AI could not clock that it was a song made by a prompt.
@@garretta4911 AI is extremely good at replication. I'm a digital composer, super novice guitarist, and vehemently bothered by capitalistic greed taking away creative jobs that I love and even personally engage in for no pay, just the gratification, and I cannot tell apart most AI generated songs from human-made
they are with both udio and sono
About 15 years ago, for an encore piece, I transcribed one of the "Bach" fugues that David Cope programmed a computer to compose back in the 80s/90s. I selected a particularly beautiful one (D minor I think, but it was years ago and I've long since lost the sheet). I was deep into Gould at the time, and really picked out the pointillist versus legato spirit of his playing. (That's quite a lot of human interference, I will allow. So quite a "dirty" experiment.) I then told the audience that was a very new piece, had been computer generated - and had taken less than 3 seconds to compose. They hadn't expected that!
I can't say how delighted and grateful I am to see an Adam Neely music education video. I always learn SO much. I know you decided to largely step back from the education stuff to focus on playing and performing and I totally understand why, but these videos are so amazing. That said, while I completely agree with your scathing critiques of the techno-optimist/accelerationist paradigm, I'm inclined to invoke the words of Debora Iyall and Romeo Void when it comes to passing a musical directive/process test - i.e. never say never. I know I won't be around long enough (I'm over 60), but you might be...
Just came back from work and saw you uploaded a 30 minute video essay of AI music. I love the effort you put in every video and how cool it is just to find out what are you up to this time
With enough computational power, any directive test will become output test. A "jamming" AI wouldn't just process what noise it need to make to continue the jam, it will also make predictions on what your next move will be, and adapt this model on the fly as the jam session continues. Eventually it will know what you will play next, before even yourself. Then, it will formulate an output, and play it back in real time. This is definitely not possible on current hardware, but with advancements in technology, it will happen. It is similar to the chatbot Turing test in a way. Chatbots are only able to pass output Turing tests, but with their speed in processing and calculation, and through the setting of a text based chat room, any sense of a direct conversation actually becomes a purely output based test, this is why chatbots are passing Turing tests quite reliably. Another thing I'd like to add is, to pass a Turing test, you don't need the observer to make the wrong call every single time, you just need to approach 50% mark, as at that point every guess is a random 50/50 shot at being correct or not. This is much easier than most people think, and since us humans are imperfect beings ourselves, the AI doesn't need to be perfect either, it just need to be good enough.
thats where it will fail machine wont be able to predict the next chord and just play it. it will require active listening and constant adjusment to the change, not prediction. A musician can just jump to an obscure chord that doesnt resolve the previous and justify it with the following chords that resolves everything that happend. thing is will the A.I play the wrong note and make it good in a jam? because thats what happens. if A.I will go "the best route" it already failed. it needs to learn to recontextualize "improper" notes, chord, and play it to sound good. In short it needs to have taste, and evolving one that knows how to communicate in an abstract level.
@@gregrodrigueziii8075 If you were shown the capabilities of this tech now from lets say ten years ago. You wouldn't believe it unless the sample sounded glitchy or funny. Even then it would be hard to swallow. No matter the benchmark people will always raise it to feel they are in control. We fear it's power. Some are already trying to destroy it by poisoning its data, but it's too late. The box has been opened its only a matter of time before we are greeted by a new species.
@@gregrodrigueziii8075 ...except that's all down to the speed and power of the AI. Assuming that an AI can't ever predict or analyze in real time is a losing bet. We already have conversational AIs that can keep up with humans in real time (below the typical cognitive reaction time to keep humans from noticing the processing going on). The conversations get weird some times, but we're really only a few years along the path of generative AI, and moving REALLY fast. A suitably fast AI would be listening to the music being played, figuring out the branching chord possibilities, and watching for things like body language and stance. It would be comparing the music as it stands with a huge list of "music shapes" that fit similar patterns, and if a human player suddenly threw in a weird chord change, it would pick one of the next logical steps before hitting that next note. Don't confuse "we have sorta-okay AI that can't really do that" with "that can never be done with AI." Or "AI made a shitty commercial jingle because someone fed it the prompt that made it happen" with "AI can only make shitty commercial jingles, forever."
@@gregrodrigueziii8075 what if "best" is defined in terms of "human-like" for the AI? The AI will soon learn to throw in improper chords, if the data you feed to it is like that. If you only feed it well-behaved music then it will only know to produce well-behaved music, for sure. but the crucial thing is that you can decide what kind of data it learns from.
@@chadirby6216 I absolutely agree. I'm currently a college student studying Neuroscience and Computer science, so, while I'm not an expert, I do have some insight. You can program anything you want into an AI. And programming it to recognize abstract taste is not as complicated as it seems, because: 1) Much of human behavior is more stereotyped than we would all like to think it is. For example, as Adam constantly remarks, musical improvisation draws on a preexisting vocabulary already. 2) You can program into an AI some level of error so that it doesn't take the "best route." And it is a relatively straightforward task to optimize that level of error. 3) AI doesn't actually NEED to recognize that abstraction as of yet, because it can guess well enough to fake it. and 4) (and most importantly) In the past ten years, AI has gone from simply an algorithm, to systems of equations so complicated that they can solve problems that go beyond their training data. And that is in ONLY TEN YEARS. There is every indication that machine learning algorithms will continue to scale as time goes on. You COULD argue, that it would be hard to get data sets of musicians jamming on the scale that AI needs to formulate a copy. But to say it is impossible is simply wrong. If it is profitable enough, the inescapable march of capitalism will make it happen. I don't know if AI will replace musicians. Either people will push back on it because, as Adam said above, music is more than a product. It isn't just a can of soup to be mass created. OR as generations go on, it will be so pervasive that everyone will be desensitized to its source. Who knows? But I don't think its a matter of whether it is possible. Because it is. I think a more important question than, "Will AI do this or that?" is "How do we create a functioning society in a world where AI can do anything better than we can?"
This was insightful, and i enjoyed learning more about musical expression. But forever/never is a long, long time.
The jamming with the red lobster tune was a great scene. Amazing. Keep going, human.
As someone who studies AI and _does_ understand how it works... ...You are exactly right. There is no technical limitation that would stop a reinforcement learning model inside a human-shaped robot from being able to jam, and do it so convincingly that other musicians would still enjoy the process. The main reason it is doomed to never happen, is precisely as you said, there is no incentive to ever make something like that. When a company or grifter does not care about the process and only wants a finished product, they will choose generative AI ten times out of ten. The reason AI will never be able to _truly_ make art is because the people using AI don't want to make art. If they did, they would hire artists to do it.
Saying that there is no incentive to do something is already a strong motivation for someone somewhere to do it anyway. Lot's of people want to make art and use some A.I. as a tool where it helps their art-work. There are also lot's of creative artistic people working on A.I.-systems trying to make it do other potentially artistic things. The world is not perfectly separated in artists of one side and money-optimizing companies on the other.
@@richardbloemenkamp8532 Most of the people researching and developing AI are not artists, though. Sure you could argue there is a market for AI tools to streamline the actual artistic process, but on the thesis statement of the video, we are never going to get an AI that could convincingly act as a member of a jam session. Being able to and on-the-fly adjust the way they play, riff with their partners, give them feedback through movement, see their movement as feedback, all while passing the turing test, while *technically* possible, will never happen because the people with the money to fund this type of venture see art as an end-result, hence the focus on generative AI.
i wonder what you think about the middle path. about how artists use AI to lazily render parts their work (i hope nobody is rendering over imagegens as the foundation/base). or the other way around, how prompters learn drawing (or actually just photoshopping) to force more control into the AI.
@@weakspirit_ I'd argue that is still a form of art, but not the same as what they're trying to replicate. Artists who might use AI as a baseline and edit around it still don't care about the drawing process, which is why they aren't doing it. That is, notably, different from artists who use AI as inspiration and make their own art from scratch. I think a good analogy is the difference between a chef that hand-makes every part of a pizza, someone who buys pre-made dough and makes the rest of the pizza, someone who buys a frozen pizza and reheats it, and someone who orders Dominos. If someone says they "made" pizza, it would hold true less and less the farther down you go, because the process is being simplified and less work is done. In a similar way, you can still make art using AI as a tool, but less and less art is actually being done the more you rely on it. I'd have a problem if someone said they "made homemade pizza" if they just ordered Dominos or reheated a frozen pizza, so I wouldn't say someone "made their own song" if they generate a full song with AI or do some light editing on a generated song. Hopefully that makes sense.
@@bagodrago i see your point of having a spectrum from genuine human work to rather soulless slop. anyways, that was a lead into a different thing that's kinda intersecting this issue. i think the process/performative of making art is not made equal between types of works. what i'm saying is that sometimes nobody cares when considering *true* art. instead reaching for background checks of the prompter or AI inaccuracies left in the slop. i don't wanna be that guy, but the process sometimes is just... irrelevant. It's not always a social activity or a live performance, sometimes it's just... tedious practice, planning, iterations, and restarts. a mundane unimaginative journey unseen and not exactly engaging with anyone. removed paint layer variations. DAW project's single-branch undo history. straight up crumpled papers of unfinished paragraph variations in the trash bin. for these kinds of arts where the process is hidden and tossed aside, other properties take place when compared to slop. cohesiveness, accuracy (slop is fraught with inaccuracies), consistency, intent, novelty, background/context, the author themself, etc. all this to say that *true* art isn't just about having a process or interacting with fellow artists, but having a verifiable human behind the wheels with "human imprinting" in and surrounding the work. to conclude: replicating a whole ass artist is required to effectively solve an artwork turing test, music turing test, pottery turing test, etc. not just participating in the process dynamically.
If "musicking" was irrelevant, there would be no difference between a live performance, and a PA + iPod. That everyone implicitly appreciates the difference leads me to believe that music will be safe, just even less profitable than before without the corporate moolah :(
Wait until a generation grows up entirely on AI music, which can't be performed, and so going to see their favorite musician live will never enter their minds as something that could have been possible. They will not miss it. Their lame grandparents will never be able to convince them otherwise.
@@RoxfoxMusic that can't be performed? Makes no sense. As a hypothetical grandparent I would hear my child perform to AI music all the time though very off key.
@@chatboss000 The point is that people go to concerts because they like the recording of the music and want to see the people who made the music make the music right before their eyes. This can't happen with AI generated music. Nobody performed it in the first place. Let's ignore that Hatsune Miku exists for the duration of this thought exercise.
@@Roxfox I dont agree with the statement that people go to concerts because they "they like the recording of the music and want to see the people who made the music make the music right before their eyes". I think people go to concerts because its an extension of the experience listening to it into a social space. Experiencing it with other people, being lead by a parasocial relationship, having the process of the trip which is an event in and of itself. I mean, cover artists exist because people want to see their version, based on the person and the social experience. Same goes for classical pieces, none of these people "made the music" (and im many cases i might not even have heared their version before either) I'd go there because i know its an experience in and of itself being physically there and seeing people perform their passion and sharing this experience with other people. So people can also present their version of an AI generated song to the same effect. I believe the social aspect will see a boom (or at least no bust) with people interested in music appreciating the process more and enjoying physical events more. However the "bread winning" aspects of music will suffer. I dont think most advertisement music writers and many game music artists wont make it, simply because the end users gets no benefit from their existance. In these cases its IS simply about the product, not the process.
@@mlsasd6494 I hope you're right!
The idea that AI will ever fail at music is laughable when you consider the formulaic way pop hits are written.
You can get an AI question-answerer to answer differently if you tell it, if it makes a mistake, you'll kill its mother. It doesn't have one, but in threatening it, you're invoking a different sort of imaginary human to answer you. I think if you ask it for philanthropic ideas, you could make its answer worse by telling it a correct answer will win it a million dollars: you'd be invoking the wrong imaginary person to answer. So in musical interaction, it's not 'can it be human' but 'what part of humanness are you trying to invoke out of the large tapestry that the AI's trying to represent'. The reason you can't jam with one is, it doesn't want to jam, because it doesn't want anything: to find its wants you'd need to not look for human qualities and search for what its 'unconscious' likes to bring up.
10:50 dang man’s streaming Pat Metheny through his fingers, killer.
"The output-test is all that matters" - The customer That the AI Techbros don't care is an understatement. They solely care about how to generate more income for themselves. Peek lack of empathy, and such. F* AI... Keep slapping and having fun bro. You deserve that.
great video. thanks!
Some of the AI music I have listened to has already passed the "I can tell this is AI" test for me.
when the world needed him the most, he returned
@Stubbyblumpkin kinky
"But then there are people that I do not respect, like the people who run companies like Suno, Udio, and other AI companies, who have a very accelerationist mindset when it comes to this, it feels like music is just one more box to tick on the way to the singularity. Music is a problem that technology can *solve*." This resonates with a the philosophical thinking I've been doing recently. It's strange how it keeps on coming up in life.
This video is doomed to not age well.
I’d swear it was «in-ter-LO-cu-tor»
Chief Technical Officer / Musician here. AI doesn't have to think like a human or do human stuff to imitate human activities in a realistic way. It only needs a large enough dataset of sensory information of humans creating/interacting to train itself on. Your expected outcome will be much harder and it will require much more data. But it will not require new technology. It would not even need to simulate a body, it is not concerned with the why or the how, it is only concerned with the statistical probability that after one specific input comes an specific output. Enter a billion jams with all their visual and auditive cues into thw training data and the models will do the rest.
Yes, exactly.
This is a very reductionist approach that might not necessarily work with this kind of task. Improvising is not just a matter of behaving in the most likely way according to statistics, rather, it's the opposite. The sense of liveness in a musical exchange is a very qualitative experience and it might not even be quantifiable just by giving examples and inferring a function. If you approach it in the way you suggested, you might end up with a problem that requires so much data and so much computation that you would burn all the available energy in the solar system before making it work properly.
@@olbos_xyz the part you don't understand is the qualitative experience is performed by humans, and then digested into the training data. The process of creatinf art is not statistical, but again, AI models are not concerned with recreating the process humans follow. It really doesn't matter. What matters is to create an output humans can't distinguish from the real thing. Most of that computation was already performed by real people jamming. Now those jams need to be diggested by models. Most of the power needed in Adam's example was used ro recreate the human process. If you take that out of the equation the power needed will decrease dramatically. It will still be a huge endeavor, but not as dramatic as the original idea.
@@olbos_xyzTo improvise or not is still the same, information. Millions of improvisations will have their own type of patterns. Our computer (brain) has already done the heavy lifting by creating that information in large quantities, the AI just has to assimilate it.
Yeah exactly. Adam inadvertently laid out all the steps that AI researchers need to solve to create an AI that can jam. The hardware is probably not there yet, since current LLMs take minutes to generate a piece, but I didn't see any portion that he called out that isn't possible to train a model to do
Wintergatan, with the marble machine, occupies a strange outlier space in this whole conversation. A machine body with a human computer.
I mean.... if you think about it, every instrument is a machine. So I don't really see how Wintergatan would be any different from any other instrument besides just being more complex. Nothing about it would really denote artificiality in its musical nature more than a guitar or an oboe or a DAW
As the marble machine will be never finished, we will never bow to our machine overlords! 😂
I think you have an interesting perspective. there's something there
The marble machine is very cool, but it's effectively just a fancy midi playback machine. You still need to compose the thing that the machine will play.
That's just an instrument
You're such a deep inlet of refreshing common sense, thank you for that :))
Your introduction is a wonderful prompt for a Ai with access to your external sign of satisfaction to produce exactly what you ask for ,,,
Thank you for the shout-out Adam! I'm writing a book on AI Music for the general audience. Would love to interview you to get your take on the subject. We disagree on various points, but I definitely agree with you that several current generative music companies have a particularly problematic approach.
and have you met the AI Singer Songwriter that is trending on Spotify?
@@Cola_BB Nope. But I'm planning to interview various musicians / researchers / entrepreneurs in the space for the book.
@@ValerioVelardoTheSoundofAI I am the trending AI Singer Songwriter
Music is no longer a product. We are now the product.
Nailed it
you're right, and when my mom used to put me in the shopping cart so I'd stay still, it's technically product placement
@@stefevr Ok, I stand corrected. It is your 😀digital footprint/life on the internet and all that implies is the product.
There are so many parallels in every art field. For illustrators and painters, the advent of photography was seen as something to be loathed, something that went entirely against artistic expression. Now, it's considered its own form of art. Same with midi versus performance. In both cases, they're both tools that offer pros, as well as cons to the other. If you had to perform every musical piece you wanted to create as a composer, or needed a group to re-enact it for it to come out in any meaningful manner, through the limitations of the physical medium of instruments, there is a lot of music we could never have made. Add to that the advent of synthesis as a form of instrument and artistic expression, and you get even deeper into the brass tacks of why the advent of midi is something to be celebrated. It offers something completely different from what the performance of multiple musicians jamming together would be, as does photography to an artist/painter. I can definitely see the great many issues with AI, but my belief is that the fields AI will take over are the fields artists do not want to toil away for, generally speaking. That is to say, art made for others, for those others' benefits. You get to focus more on making art for you, to express who you are, rather than to express who somebody else is. Meanwhile, AI will largely be relegated to things that were largely considered soulless and consumer-oriented already. It's just so much less rewarding a thing to do than making art for the sake of art. At least, that's my take and my hope. The big issue here, of course, is the ability for artists whose works and soul have been copied and trained into an AI to subsist after it takes over the market for cheap labor.
Training on a bunch of existing data is literally exactly what musicians do. While describing LLMs you explained the process of learning to jam to a T. It's weird how that's a normal process for a human musician but it becomes offensive when a machine does it.
B i n g o
Adam, you playing along with the red lobster tune illustrates how much that tune NEEDED a human player in it in order to give it an authentic vibe. You improved it at least 100%.
Ai will replace commercial art. Corporations will use Ai to make art. But that being said, the beauty of art, the “human ” aspect will always have a space.
Weirdly, I feel like this is an area where art will improve...
Eh I feel like corpos will try but people will get absolutely sick of the decline in quality
@@bmac4 I mean, the threshold is very low. And AI generated images to go and blow up and frame cost nothing. But also it's you asking what you want it to create and getting to basically bother the machine with inane edits instead of the back and forth of a human process to acquire commercial art. No matter how low budge and copy paste, a brilliant human art team is going to have to spend more time and effort in just the back and forth of conferring and relating back edits and such and making the client's dreams come true so...that is a huge cost cut that can allow for (for a time) a boon of generosity of effort bc it cost so so so much less effort all around. Who knows.
Doesn't have to
There is nothing unique about human output. Individuals can be different from each other based on our heritage and upbringing and culture, etc. But in the end, we are still producing something with the same tools. That output is replicable. See the classical monkey's typing Shakespear analogy. Except our AI monkey's already have all English literature to draw from. The only unique part of a human making art is if we want that specific human to do it.
AI has no capacity of smelling or tasting or eating cheddar biscuits. As consumers, normally we are asked to believe that the entity behind an ad jingle is speaking from experience, therefore we should buy these things and eat them, too. With this Red Lobster ad there's no possibility whatsoever that there's any verity to the claims in the songs. It's not just the artistic process of making the music, it's because there's no human-sensory-life experience behind the generation of the music. Musicians aren't just musicians. We're alive and it's our life that goes into our art. 24/7. Art made by humans is more believable, not just because of the artistic process, but because of life itself. It comes down to suffering, really. There is no art without suffering. AI ain't got no pain, guys.
This is why I loved "Get Back" (the doc)
Important to note is that your definition of LLM is quite accurate. People overhyping AI really thinks Ai is thinking, but is not, is just trying to guess how to complete a context. This is quite different. Logic, for example, is not a prediction of a probable word. If you want to break any AI just throw any logical riddle in the query.
But people fail logical riddles all the time as well. I dont think thats as compeling as an arguement as you think.
Also, that's kind of what Turing's test is all about. If you can't tell the difference... Does it matter how it's being made?
People seem to think thinking is something aside from a weighted neural network converting inputs to outputs based on a desired reward outcome. Baffling and shortsighted.
@@theherk yep, its all just something they cant wrap their heads around. We are nothing special. Biological machines. Its all just a matter of complexity. And to think, silicon based AI will never reach that complexity is just short sighted and rather simple minded. The building blocks are all there now. Its only a matter of time. Even whats stated here will eventually just be 'cracked' by an AI. Jaming isnt even as complex or mythical as Adam makes it out to seem. We might not see it in our lifetime. I dont know, im not a prophet and can predict the time it will take. But it will happen eventually.
I exclusively use the Dvorak keyboard layout. My favorite thing ever is inviting people to use my keyboard and watching their frustration as the letters they think they are typing come out as complete gibberish.
Honestly, I once tried an AZERTY keyboard in France, and it was making me crazy. I don't know what is worse, because at least I wouldn't try to use a Dvorak keyboard in a normal way.
I gave a guy my phone to put it his snapchat and it took him over a minute to type it out 🤣. I offered to switch the keyboard half way through but he decided to power through it.
I was in Abu Dhabi and tried to pay for something by card. The touch screen numerical keypad they offered had the numbers all mixed up. I couldn't do it. I'm still not sure if that was a standard layout they use, or if it was randomised for extra security.
You failed the turing test
@@jaapsch2 i remember all my PINs and passcodes as shapes rather than numbers... i'd be completely screwed if i came across one of those.
I seriously hope you are doing this to stir up controversy, nobody in their right mind is still saying "AI will never..."
This is the first time I actually considered nebula. Great ad, Adam.
I'm more interested in why tech bros all seem to have this weird hatred/jealousy of creatives? They seem to revel in the idea of AI replacing human art/music despite having zero interest in art or music themselves.
I’ve noticed it. Sometimes it’s politically motivated because many artists are left-leaning, but there seem to be more psychological motivations at play. It’s like they see the idea of AI replacing people as a sort of revenge.
@@stoogelMost tech bros are left leaning, dude.
@@christopherjobin-official7440 There's a strong libertarian streak if you look at the Silicon Valley venture capitalists types and Musk
Because creatives are blooming with life and it’s wonders. Those code spitting dudes without any social skills aren’t. F* them! I literally consider AI tech bros as enemies. Like someone I would slap in the face easily (to say the least)
To increase the shareholder value. That’s it. The art doesn’t matter as long as it makes money. So what if it’s all derivative and sounds the same? The top 100 charts have a ton of tunes that sound the same. Pretty much the same chord changes, keys, etc. There are a few on each list that are different/unique, but the labels don’t care. They have 360 deals w almost all of them so it’s zero risk for them. If you can get AI to generate the 90% of the schlock, then they don’t have to pay writers, producers, artists, etc. and no royalties either.
Great video Adam. You always have interesting and worthwhile content! Keep jamming!!🤟🏻
The best part is where Turing pre-refuted all kinds of arguments against computer consciousness that you'll still see today
Noted. Come back to this video soon.
Most musicians fail the turing test. Musicians have slowly been removed from orchestras to one man bands. Once you remove the audience, the process is complete.
Sometimes even the tuning test
Depends on how your trained i suppose? Improv is a skill and I think that anyone can develop in and groove
this is so petty lmao
yeah I've been a musician for 20 years and have never jammed/improvised. It's all about the output test in my realm
this is so reductive
First that Laufey jazz video and now this. Neely is killin it!
Adam is a master of saying, "It doesn't matter, but here's why it does."
To anyone who hasn’t seen it, I highly recommend watching Jojo Mayer jamming with AI - especially the longer talk/performance, where he talks about how weird it is. Fascinating perspective from one of the best drummers alive.
I like that you pronounced Dvorak the keyboard layout like Dvořák the composer.
Oopsy, could have timed this better. Elevenlabs is getting into the music production stuff. It's fun going back a couple years to see all the things that "AI will never do".
Exactly
This is all super interesting and I wouldn't dare disagree with anything you say because I'm positive you know more about all of this than I ever will BUT I would like to alter up an alternate dichotomy. Music is not the same thing to everybody who enjoys it. There are many distinctions to be made but I think a HUGE split when it comes to the topic of AI is that of the music listener vs. the music maker. I love music. I listen to it all the time. I'm cool with "most everything" and when I say "most everything" I'm NOT saying "most everything except country and rap." I can jump from Paganini to GZA to Ahab to Carly Rae Jepsen in a single car ride. But when it comes to being a musician, well... I tried REALLY HARD in my high school and early college years to form some kind of band, or "project" with a friend or two, or even be my own one man show. At my peak I owned a guitar, 5 or 6 effects pedals, a bass, a keyboard, a drum machine, a couple of mics, and a 4-track recorder. I would literally pack ALL of this in my Buick along with 2 guitars from a friend and we'd drive 30 minutes to this other guy's house who had a huge basement with a piano and a drumset and we'd "attempt" music all weekend. During a separate period of relatively prolific output I made a lot of music with a friend who taught me a lot about what I guess would be considered a DAW now - back in 2001-2002 it was just "Cool Edit Pro 2.0." I had a lot of fun during these sessions. We made some OK music that I wish I'd kept better track of over the years. But none of it was ever "good." And I know now that it was never going to be. As I got older and got into the bar scene for a few years after college I actually met some *real* musicians. And that's when I realized that, for my friends and I at least, our need to have fun with it was never secondary to our need to create. But that doesn't mean I love or loved music any less, it just means I wasn't cut out to make it. I still love to punch around on my sheet music/MIDI program and arpeggiate certain chord progressions and things like that, but I'm merely having fun, not trying to create a cohesive piece of music. My point is that there are people who love music but who do not, cannot, or will not MAKE music, and from that perspective, the process of "musicking" or "jamming" doesn't hold the same value to us. There are people that love food and who are real experts on it but they have ZERO interest in cooking. The magic for me lies in the finished product. I don't want to say I could care less about the process, but that's not really where the spark lies, in my opinion. The fact that a bunch of individually mundane aspects came come together to create something transcendent is what gets my juices flowing when I listen to music. Don't get me wrong - knowing how every little blip and beep and click and buzz and dig and bop got put where they got put is very interesting but it has little to do with where my mind or my emotions are at when I really truly love a piece of music. The "directive" side of AI music generation might be an important component, but I'm not sure it should be hailed as the cornerstone of AI musical achievement. If I can make you the same steak dinner out of $25 from ingredients at Walmart in half the time it would take me to make that same dinner for $75 at the organic market, do you really care *how* it was made? If you're a cook, it probably does matter. If not, you probably couldn't care less. Neither is any more or less "correct" than the other. In so many ways music IS about the output. That's what endures. That is the part of "musicking" that is immortalized on physical or digital media. That is the beginning and end point of conversation - the finished piece is the culmination of whatever processes were used and as such will always be more highly revered than the process itself. A jam session might be fun to participate it, but when, out of your entire day, you have 30 min on the way to work and 30 min back to listen to music, how much of that time is spent listening to a jam session? When you introduce a friend to a new band or sound, how often do you play a jam session for them? When deciding what piece of your own work to release, would you EVER choose a jam session over something more polished and rehearsed? You say "AI can't jam" and as nicely as I can, I'm coming back with a, "so what?" If AI can nail that hit single, if they can make that song that you want to show a friend, if it can write a song that's worth being one of the 10 to 15 songs you're going to listen to that day, isn't that what matters? Or at the very least, isn't that what matters THE MOST? I know musicians will see this differently and that's fine - I just wanted to put this out there as someone who truly does love music but has long ago accepted that I lack the spark to be musician.
I agree with you to a point. Lots of people listen to the same recordings over and over. I believe they are replaying emotional content that is paired to the particular piece. Jumping through pieces of wildly different styles is a way of shaking up internal emotional responses a bit like jumping from the sauna into the snow and then back again. But this isn't the the making process of musiking. That is the territory of exchange, conversation, reciprocity in real time. Memory can be there but it stays in the background and can quickly be left behind as the jam goes off into new iterations.
I understand your points, and you are absolutely right under a capitalist economic system, the value is placed on the physical/digital manifestation of the art that can be used as a product. That’s what we learn intrinsically has value. The truth I’ve found, is just jamming with some folks is absolutely essential, and because we can’t turn that into a product, we view it as a “bad” thing. If you strip away your pre-conceived notions, the chance to play music with other musicians, communicating on a deeper, HUMAN level, is valuable on a conscious level, not a capitalist pursuit.
I asked why can't everything be just simplified to one one one one. This helped me clarify what time signatures are, a way to emphasise pulse. My friend was born on 11/11 and I wrote a birthday song on 2011. It was fun trying to reduce natural pulse with singing that many ones in a row.
I don't get this, man. The robots at Chuck E. Cheese have been passing the Turing test for music since the 80s.
More of that jam at 10:45 my man
Seriously, where can I find the full video or version of this?!
21:31 - What makes it good is that it's fun to jam with. NOT that the song itself is good. I got out my shoes and started tapping with you. It's legit.
Great video. Linking music to a physical activity, and not just a mental one, is profound… and the first time I’ve heard someone say that. Although it seems so obvious now… It makes me think about how little we still understand about the links between the physical and mental worlds. Mind and matter… we can’t measure mental phenomena (yet) with any objectivity so wtf even is it? We’ve no idea!
Adam your stuff is always excellent and one of my dreams is to get to jam with you someday and I promise I’m not secretly an AI
I don't know man, a bot that can jam with you, with the personality of a musician you ask for, sounds like a product that would go viral and sell incredibly well. I really liked the video because it shows that you make it with passion and trying to be as objective as possible from your point of view, but honestly it didn't convince me at all. I don't understand why so many people separate AI from humans as if they were opposite things that have nothing to do with each other. These AIs are the product of the history of humanity, of our behaviors, our curiosity, greed and our ambition among other things, it is an incredibly HUMAN technology, for better and for worse, it is not something foreign to our essence, rather a reflection of it. At the end of the day I agree that all of this is scary af if used the wrong way, which clearly it will happen.
I don't think this will age well. Currently, music AI is focused on output generation, but there's no reason why it has to be. That's just the thing most people are likely to want. After all, there are for more people who want instant good music than there are musicians. Want an AI that passes the directive test? Simple enough, we train a model to output steps in the process. Who would spend the money to train it? The same people that try to sell musicians software and gear. Don't have musician friends to jam with? Is your drummer too busy dating your guitarist's sister to show up for practice? Is your bassist on vacation in the Philippines? Do you have an idea that just can't wait for other people's convenience? Well well well, do I have a product for you! Say hello to your robot bandmates! They are available at all times and can even help you develop new sounds and songs.
2 minutes in, and the fact that you said there's a musical question and the machine has to answer, sounds exactly what ai was made to do, so I'm not convinced about it not happening yet.
Jacob Geller mention! love that guy's work