Categories:
Categories:
As artificial intelligence is creeping into so many applications these days, I thought I’d share what it “thinks” about jazz and improvisation. To do that, I had a conversation with a site called AIChat.
In its own words, the site allows users to interact with AI language models in real-time. In other words, you ask it a question, it “thinks” and then writes out its answer in front of you.
I use it for coding help when I am stuck on a programming problem, and also ask it questions I have about other things. While less thorough, it seems less biased that the human-edited Wikipedia.
It does feel like it is thinking as it types out its answer. It is not just regurgitating text from the web. In fact, testing some of its answers using a plagiarism site, it was completely clean. So, naturally, I had important questions for it. Why not ask it about jazz?
My first question was, “What is jazz improvisation?” It answered:
Jazz improvisation is the process of spontaneously creating music in the moment. In jazz, improvisation is an important part of the performance and is often a central element of the music. Jazz musicians improvise by using their knowledge of scales, chords, and harmonies, as well as their understanding of the structure and form of the piece they are performing, to spontaneously create melodies and solos on the spot.
Improvisation in jazz is often associated with soloing, or the practice of playing a melodic or rhythmic line over a chord progression or set of chords. Jazz soloists will often use their understanding of the harmonies of the piece they are playing to guide their improvisation and create a solo that fits seamlessly within the structure of the music.
Improvisation in jazz can also involve collective improvisation, where all members of the band contribute to the creation of the music in real-time. This can involve the use of call-and-response patterns, trading solos back and forth between band members, and other techniques for creating music spontaneously.
Overall, jazz improvisation is an important aspect of the genre and is a key element of what makes jazz music unique and exciting to listen to.
Okay, that’s a pretty good general definition. And again, this is not just some copy of Wikipedia.
Let’s ask it something a bit more technical. How about, “What scale should I play over f7 flat 9?”
The F7 flat 9 chord is a dominant 7th chord with a flatted 9th. One scale that you can use to solo over this chord is the Mixolydian mode. The Mixolydian mode is the fifth mode of the major scale, and it is built on the fifth degree of the major scale. In the key of C, the F Mixolydian mode would be: F G A Bb C D E.
You can also use the altered scale over the F7 flat 9 chord. The altered scale is a scale that includes all of the altered notes (b9, #9, b5, #5) from the dominant 7th scale. The altered scale for the key of C would be: C Db D# E F# G# A#.
Another option is to use the diminished whole-tone scale. This scale is built by alternating whole steps and half steps. The diminished whole-tone scale for the key of C would be: C Db D# F G Ab A#.
It’s important to keep in mind that these are just a few options, and there are many other scales that you can use over the F7 flat 9 chord. Ultimately, the scale you choose will depend on the musical context and the style of music you are playing. Experimenting with different scales and finding the one that works best in a given musical situation is an important part of the process of developing your musical voice as a jazz musician.
Let’s not completely dismiss this because the Mixolydian mode has no flat nine. After all, it did suggest altered scales.
What would an AI suggest that would make me a better improviser? It’s answer…
There are several things that you can do to become a better jazz improviser:
Practice regularly: Like any skill, improvisation requires practice. Set aside time each day to practice improvising over different chord progressions and melodies.
Learn and memorize scales and arpeggios: Knowing the scales and arpeggios that correspond to the chords in a tune will give you a solid foundation for improvising.
Listen to and study recordings of great jazz improvisers: Listening to recordings of great jazz improvisers will give you a sense of what’s possible and help you develop your own style.
Practice transcribing solos: Transcribing solos by ear is a great way to improve your ears, develop your vocabulary, and understand how different players approach improvising over different tunes.
Experiment with different approaches: Don’t be afraid to experiment with different approaches to improvising. Try playing over the changes in different ways, using different scales and arpeggios, and exploring different rhythmic and melodic ideas.
Play with other musicians: Playing with other musicians will help you develop your skills as an improviser. You’ll have the opportunity to listen to and learn from other players, and you’ll be able to practice improvising in a musical context.
Improvisation is a lifelong journey, and there is always more to learn and explore. By practicing regularly, studying the music, and experimenting with different approaches, you can continue to improve your skills as a jazz improviser.
I like that. I wondered if it considered itself capable of improvising. I asked, “Can you improvise jazz?”
As an artificial intelligence, I do not have the ability to perform music or improvise in the same way that a human musician can. While I can provide information and guidance on various musical topics, I do not have the physical ability to play an instrument or create music.
One more question. I wondered what it “thought” the difference was between man and machine. I asked, “Is an AI more capable than a human being?” This one took longer for the AI to answer. You can see the blinking curser before it starts to type its answer. Eventually, it came back with…
Artificial intelligence (AI) systems are capable of performing a wide range of tasks, and in some cases, they may be able to outperform humans in certain specific areas. However, it is important to keep in mind that AI systems are not capable of replicating all of the cognitive abilities and functions of human beings.
AI systems are designed to perform specific tasks and are typically trained on large datasets to perform those tasks effectively. They can be very good at performing certain types of tasks, such as recognizing patterns, analyzing data, and making decisions based on that analysis.
However, AI systems do not have the same level of general intelligence or cognitive flexibility as humans. They do not have the ability to learn and adapt in the same way that humans do, and they are not able to perform tasks that require creativity, empathy, or complex problem-solving in the same way that humans can.
It’s also important to keep in mind that AI systems do not have the same level of physical capabilities as humans. They do not have the ability to interact with the world in the same way that humans do, and they are not able to perform tasks that require physical dexterity or mobility in the same way that humans can.
Overall, while AI systems are capable of performing a wide range of tasks, they are not necessarily more capable than humans in all areas.
I have always been fascinated by computers. My high school rolled out its first computer course in my senior year, and I signed up for it as soon as I heard. I remember the first program I wrote: I found out the birthday of a girl on whom I had a crush, and proceeded to write a program that graphed out her personal biorhythms. She was suitably impressed!
We’ve come a long way since those ancient days of giant mainframes miles away connected by 1,200 Baud modems, but every step of the way, computers have been an important tool for us humans.
Have a chat with this AI. It’s free, and you just might learn something cool like, “Where is a good place to hear jazz in (fill in the city).”
Groovz Playground
backing tracks (FREE)
Trombonist, author, marketer, & tech guy
Share this post…
I recently watched a video performance by the amazing classical pianist Yuja Chang. I’ve seen her memorizing motion and heard her virtuosic playing before, but something hit me after seeing
I have created a AI chatbot called Jazz Master Chat that draws from 75 hours of interviews from my Jazz Master Summit event a couple of years ago. I interviewed
What is jazz improvisation? Let’s first define what I mean by jazz improvisation. Jazz improvisation is a spontaneous conversation, but instead of words, we use notes. Look at two possible
My recently turned 18 year old son is a passionate photographer He 8217 s got himself a little business
A couple weeks ago I sent Richie Beirach a YouTube clip from the movie Whiplash as a bit of
I originally meant to write this as a reply to a comment Richie Beirach wrote on my blog But
Tools for helping musicians at all levels learn about jazz and play to their full capability.
Web design and marketing by:
Michael Lake @JazzDigitalMarketing.com
This is just a fake book example for the type of website I can build for you. Just trying to use a little humor here!
12 thoughts on “A conversation about jazz with an Artificial Intelligence”
I asked ChatGPT very similar questions, Michael, and was generally impressed with the answers. I have found that asking it to tell me what chapters it would include in a book on X brings me to deeper questions and soon points me to material to learn and practice. Asking it for reference materials to support chapters or topics also shows me what materials it has used in its “learning”.
It would be interesting to train it or other similar tools on actual acoustic samples from Parker, Brown or whoever.
A nice sharing Michael
One of these days we should connect on zoom.
All my best
Mike garson
> You can also use the altered scale over the F7 flat 9 chord
…
> Another option is to use the diminished whole-tone scale. This scale is built by alternating whole steps and half steps.
“Altered Scale” and “Diminished Whole-tone” are different names for the same scale. I think it was trying to say “Half-Whole Diminished” based on the description, but it then spells the wrong scale (should be C Db D# E F# G A Bb). This is a perfect example of one of ChatGPT’s weaknesses/dangers: it makes incorrect information look convincing if you don’t know any better.
You are correct. As I tried to say in the post, AI is evolving but is still pretty impressive at this early stage. When I want to know the CSS for the transparency of a web design element, it is spot on with the code. When I want to know how best to play over altered jazz chords, it provides a starting point needing a human like you to correct and clarify.
Hi MIke: Thank you for this post. This was a very interesting read. I noticed several errors, etc..that I’d like to correct for some of your readers, and hopefully, the programmers of Mr. A.I.):
• Mr.A.I. — “…The Mixolydian mode is the fifth mode of the major scale, and it is built on the fifth degree of the major scale. In the key of C, the F Mixolydian mode would be: F G A Bb C D E….
1.) Regardless of the key, an F mixo. scale has an Eb, and not an E.
In the key of C, the diatonic mixo. scale is rooted on the 5th scale degree, which is G, and not F. (Mr. A.I. counted down a diatonic fifth which landed on the P.5 below DO; the root of the IV chord, and not V.
I suppose Mr. A.I. assumes that any note a 5th away from DO locates a diatonic mixo. scale. Mirror images can be tricky.)
• Mr.A.I. — “…The altered scale for the key of C would be: C Db D# E F# G# A#….”
2.) First, the chord tones are incorrectly spelled. The universal chord formula for spelling a 7(b5) chord on any one of 12 possible roots is : 1 3 b5 b7 = C E Gb Bb.
The #9 is spelled correctly, but, it’s more common to spell it as a “b3”, so, C Db Eb E …
Next, the F# is not the spelling of any kind of 5th above a C. That is the spelling of a #11 tension, which is not a chord tone.
The A# is not a b7 above the scale/chord root; it’s an augmented 6.
This kind of false function equivalence perpetuated by enharmonic spellings is very typical among analog humans, who’ve had just enough theoretical exposure for them to think that an abstract musical concrete is equal to a function. (Functions [the harmonic grammar] in music are determined by the limits placed on its usage, as either a stable or unstable sound, in the context of a tonal center.)
• Mr.A.I. — “…Another option is to use the diminished whole-tone scale. This scale is built by alternating whole steps and half steps. The diminished whole-tone scale for the key of C would be: C Db D# F G Ab A#….”
3.) This statement is the most convoluted — Notwithstanding the objectionable David Jamey Bakersold mnemonic, “diminished whole-tone scale”, The so-named scale is not constructed by alternating whole and half steps . The DWT scale is simply a renamed altered scale from #2 above.
The scale that results from alternating whole and half steps is called the Symmetric Diminished scale, which, when turned inside-out, is the half-whole scale, and it’s called the Symmetric Dominant scale. The C scale that is spelled is nothing more than a C Mixo. b9,#9,b13 scale with a missing E and its Bb mis-spelled as A#.
And finally, in the list of six things one can do to become a better improvisor, the most important consideration is missing: Learn tunes, learn tunes, and learn tunes. And when you have had your fill, then learn more tunes.
Without a deep and solid repertoire, the listed six will essentially be meaningless. The tunes will teach you how to play on the tunes.
Hi Mike,
ChatGPT has impressive capabilities but it embodies the old mantra from the early computer age, “Garbage in, garbage out” (GIGO). It is limited to the type and quality of information that has been fed to its databases. If that information is lacking in any way, ChatGPT will spew nonsense. It does not think: it processes. It does not imagine: it reproduces. There have been a lot of articles recently about how AI is making impressive strings of computer code and such. A lot of commentators also see AI as the end of research and writing. This is a very active thread in the academic community.
I thought I’d test ChatGPT and the results of my experiments are in an article on my blog:
https://thelasttrombone.com/2022/12/20/chatgpt-is-this-the-end-of-writing-and-research-i-dont-think-so/
I put a number of questions to ChatGPT. I asked it to improve a poor writing sample; I asked it to improve an excellent writing sample. It failed in both cases. I asked it questions about the history of the trombone and its answers were abysmal. It parroted the first line of the Wikipedia entry about the trombone – a guaranteed F on an essay since every trombone professor in the world knows that first sentence by memory, “The trombone is a musical instrument in the brass family.” When it pronounced that, “The trombone is played in a standing or sitting position, with the bell pointing upwards, and the player sliding the inner slide in and out to change the pitch,” ChatGPT exposed itself. AI’s knowledge about the trombone is abysmal, reflecting the abysmal state of trombone research by human beings. AI doesn’t conduct original research but there are those who are fooled by its flowery language and hope it will be a quick fix in lieu of doing the hard work of acquiring accurate information and knowledge oneself.
The New York Times recently asked several teachers and a noted author to read several writing samples and see if they could tell if they were written by a person or by ChatGPT. I didn’t have any trouble with this – I got 10/10 correct. Try it yourself:
https://www.nytimes.com/interactive/2022/12/26/upshot/chatgpt-child-essays.html
There’s no doubt that AI has tremendous capabilities. But it lacks a human soul. Using it for things it can do makes it exciting. Hoping that it will replace our thinking is naive. The temptation to “let ChatGPT do my work for me” leads to a lot of unintended consequences.
This isn’t a Luddite view; it’s an observation of limitations. AI will improve; no doubt about that. But it will always be limited by information it has been fed. If that information is faulty – AI cannot reason and understand that the information is faulty (rather like a lazy doctoral student who relies on secondary sources rather than doing the grunt work to uncover first sources)- then it will perpetuate lies and half-truths dressed up in the veneer of impressive sentence construction and good grammar. Caveat emptor.
Doug, I had a completely different experience. I asked it: “What is the history of the trombone?” and its answer was nothing at all like Wikipedia. In fact, I checked it against a plagiarism checker and it came back as 100% original.
I also asked it the exact same question (wording) in three different browsers and the answer was completely original each time, with different wording emphasizing different aspects of the trombone, the styles it’s capable of performing, some geographic references, etc.
I think you sell it short.
Hi Mike,
I don’t think I’m selling ChatGPT short. I’m just recognizing that it for what it is. For what it can and can not do. I’m recognizing its limitations and the consequences of those limitations. And the biggest limitation is that it cannot separate truth from falsehood. It can only assemble information it has been fed. If that information is faulty, it will assemble misinformation.
Today’s (January 6) New York Times has an interesting podcast on “The Ezra Klein Show.” Klein interviews Gary Marcus (who has been deeply involved in AI for a long time) who has a very insightful take on ChatGPT. Marcus and I are of like mind on this. Here’s a quotation from Klein’s opening of the podcast:
“Gary Marcus is an emeritus professor of psychology and neural science at N.Y.U. who has become one of the leading voices of A.I. skepticism. He’s not “anti-A.I.”; in fact, he’s founded multiple A.I. companies himself. But Marcus is deeply worried about the direction current A.I. research is headed, and even calls the release of ChatGPT A.I.’s “Jurassic Park moment.” “Because such systems contain literally no mechanisms for checking the truth of what they say,” Marcus writes, “they can easily be automated to generate misinformation at unprecedented scale.”
Including misinformation about the trombone, as I have shown in my article which I linked to in my previous post.
You can hear the podcast here:
https://www.nytimes.com/2023/01/06/opinion/ezra-klein-podcast-gary-marcus.html
or read the transcript here:
https://www.nytimes.com/2023/01/06/podcasts/transcript-ezra-klein-interviews-gary-marcus.html
AI can be very useful for some things. But it can be very problematic for others. If you’ve read Harry Frankfurt’s paper, “On Bullshit” – you can read it here:
http://www2.csudh.edu/ccauthen/576f12/frankfurt__harry_-_on_bullshit.pdf
– Frankfurt makes the distinction between something that is false and something that is phony.
Frankfurt writes:
“The essence of bullshit is not that it is false but that it is phony. In order to appreciate this distinction, one must recognize that a fake or a phony need not be in any respect, apart from authenticity itself, inferior to the real thing. What is not genuine may not also be defective in some other way. It may be, after all, an exact copy. What is wrong with a counterfeit is not what it is like, but how it was made.”
During his podcast with Gary Marcus, Ezra Klein makes this observation about Frankfurt’s thesis:
“And his point is that what’s different between bullshit and a lie is that a lie knows what the truth is and has had to move in the other direction. He has this great line where he says that people telling the truth and people telling lies are playing the same game but on different teams. But bullshit just has no relationship, really, to the truth..”
ChatGPT doesn’t know whether something is false or phony. Since it can’t reason, and it can only assemble content based on information it has been fed, it has the capacity to be a misinformation machine of epic proportions. If you ask ChatGPT to write the history of the trombone and it says the inner slide moves in and out, that sounds absolutely plausible. Unless you actually have seen or played a trombone. If you don’t know what it’s written is phony, then you parrot phony information.
How do we know if ChatGPT has referenced Trevor Herbert’s “The Trombone” (Yale University Press, 2006) which is arguably the best and most reliable source for information about the trombone’s history, or Hans Kunitz’s “Poasune” (Breitkopf & Härtel, 1959) which is unarguably one of the two worst books ever written about the history of the trombone? ChatGPT doesn’t filter truth from fiction. Many human beings don’t filter truth from fiction. But at least human beings have a CHANCE to activate the truth/fiction filter. The history of the trombone is grossly understood because people are still quoting Kunitz. Until AI is fed every shred of information and research from the beginning of recorded time, and until it can develop the capacity to reason and understand and do independent research on its own in order to adjudicate between truth, false, and phony, it is going to be very, very limited as a helpful tool for society. In certain respects. If you ask it to improve a bad sample of writing and correct spelling, punctuation, and syntax, AI can do that in an impressive, if not always accurate way. But ask it questions about something it doesn’t know about, it can spew bullshit. I just asked ChatGPT this question:
“Write about the long solo for bass trombone in the first movement of Beethoven Symphony No. 5.”
ChatGPT wrote this:
“The first movement of Ludwig van Beethoven’s Symphony No. 5 is known for its famous four-note motive, which is repeated throughout the movement and serves as the basis for the melody. One of the most memorable parts of the movement is the long solo for bass trombone, which occurs near the end of the exposition section.
In this solo, the bass trombone plays a prominent melody that stands out against the accompaniment of the rest of the orchestra. The melody is played in a very high register for the bass trombone, which requires the player to have excellent control and technique in order to execute it smoothly. The solo also features a number of wide leaps and fast scales, which add to its technical difficulty.
Overall, the bass trombone solo in the first movement of Beethoven’s Symphony No. 5 is a challenging and impressive feat of musicianship that showcases the virtuosity of the trombone player. It is a memorable and iconic moment in the symphony that has helped to make it one of the most famous works in classical music.”
This, my friends, is bullshit. ChatGPT took a false prompt and invented something. It could have said, “There is no solo for trombone in the first movement of Beethoven Symphony No. 5. In fact, the trombones don’t play at all in that movement.”
But it didn’t. It made up bullshit. Not just something that is false. It wrote something that is phony.
If we trust ChatGPT about the trombone in Beethoven Symphony No. 5, then my pal, Gerry Pagano and I have really missed something when we were counting rests for all those years!
When someone uses ChatGPT as a search engine, you’ll get a few paragraphs that answer your question. To this, Marcus says:
“But the trouble is the paragraph might be wrong. So it might, for example, have medical information that’s dangerous. And there might be lawsuits around this kind of thing. So unless we come up with some kinds of social policies and some technical solutions, I think we wind up very fast in a world where we just don’t know what to trust anymore. I think that’s already been a problem for society over the last, let’s say, decade. And I think it’s just going to get worse and worse.”
I don’t know if there will be lawsuits when a student gets an F on an assignment because they write that there’s a huge bass trombone solo in the first movement of Beethoven Symphony No. 5 (Grade: F, and possible placement on academic probation because the student will not be able to support his outrageous assertion with any source), but without the ability to reason and truly understand, a user needs to know when something is right or wrong, truth or fiction, in order to correct ChatGpt’s essay.
If you KNOW the paragraph is wrong, then you can correct it. When I asked ChatGPT to write a history of the trombone, it was loaded with “wrong.” I could correct it because I know what is “right” with the history of the trombone. But the freshman in the general music class is going to get an F on their paper since they don’t know the paragraph is wrong. They trust ChatGPT. And if we trust ChatGPT without knowing that what it is saying is true or false – remember Ronald Reagan’s famous line when dealing with nuclear disarmament with Russia – “Trust, but verify” – then it is worse than useless. It is dangerous.
It all depends on how one uses this AI tool. I’m not simply selling it short. I’m just recognizing it for what it is. And isn’t.
There is a trend in recent decades of treating improvisation as some kind of sport, a mix of chess and hurdles. There’s an academic part of knowing the most advanced moves in any situation such as F7b9, and the pure sports bit of playing it faster and or higher. AI can certainly do that better and faster.
I think it will be longer before it can replace the art of connecting with an audience and driving an emotional response.
And all these years, I thought AI referred to Allen Iverson! Huh, you learn something new everyday. (And in my case, quickly forget it)
Much like artificial sweeteners, artificial turf, I prefer the real brand of intelligence. Problem with that, it’s just really hard to find. I guess I’ll leave this big topic discussions for others, you know, the brainy types. Until someone proves otherwise, I think we have been covering the planet with something less than intelligence for some time, so stupid really knows how to replicate. Maybe we should focus on how to find intelligence in humans, and let the machines keep to themselves. We probably irritate them anyway…..
And for those who don’t know, Allen Iverson was a fantastic basketball player for the Philadelphia 76’ers a few years back, and was a big fan of practice! (We talkin’ bout practice? Not game! Practice?)
Well I’m totally blown away. That a conversation with AI can lead to such an academic discussion is amazing. And might I add, way over my head. I love it.
Happy New Year, Michael. Thank you for continuing to keep me in the loop. I hope that we all have a joyous, healthy, creative and peaceful year.
Cheers,
Kent from Canada