PG Music Home Forums General Discussion Off-Topic Frank Sinatra with a nirvana cover

Why do I feel this is sooo disrepectful to Frank?

So it appears this may be AI? If so, I'm impressed.

"Why do I feel this is sooo disrepectful to Frank?"
It's safe to assume Cobain fans would feel the same....

I like both artists and I thought that rendition was funny.

I'm not sure where I sit. I suppose there are more questions than answers.

Would Frank approve? If yes, then what's the problem? If no, then that's another story altogether.

Was AI involved, and if so, well, does AI deserve a 'pat on the back' for its remarkable ability to imitate Ol' Blue Eyes??

The bottom line:
As we move deeper, beware of what you see (and hear )

Everything now can be a deep fake.

We can get videos of people saying something they never said.

On the other hand, if someone is trying to blackmail you for something you shouldn't have been doing, there is your out—DEEP FAKE!!!

The leaps and bounds technology has taken in just my lifetime are amazing. Things advanced at a slower pace in historic times.

I don't want to get into the discussion about Cobain or Sinatra would approve, I'm just impressed by the technology.


Notes ♫

Neither of those guys were deities.
It's just a song.
"Disrespectful?" Get over it.
If you don't like it, don't listen.
Some problems really do have simple solutions.

Can someone fill in the blanks for me please. This sounds like a Sinatra song with a Nelson Riddle big band arrangement but I don’t know the song melody or the words. So I assume the lyrics are by someone named Cobain. I think I’ve heard of him. Kurt? Then the real words of the song I don’t know were replaced by AI?

Whether I listen or not listen, does not have anything to do with this. Nor is that a solution to anything. It appears that folks here do not know what respectful and disrespectful even mean.

It can be a tough crowd around here!

Correct. The song is Smells Like Teen Spirit, which was a huge hit for Nirvana, Kurt Cobain's band.

Someone arranged it in the style a Nelson Riddle, big-band arrangement. That's a feat in itself. I'm guessing they build the arrangement from samples of actual Frank Sinatra songs and sounds from sample libraries. With programs like Spleeter, it's possible to disassemble songs into parts (bass/drum/vocal/keys/other). Using pitch and time shifting, it's possible to re-assemble these parts into a reasonably good pastiche of a new song.

They then sang the song in the style of Frank Sinatra, and used RVC (Retrieval Based Voice Conversion) replace the timbre of their voice with Frank Sinatra's.

Edit: After some research, it looks like the material was lifted from Paul Anka's version of the song, and it's likely Paul Anka's voice was used as well.

RVC is a neural network-based tool that, when given audio input, identifies the phonemes of the input and stores that data. It can then change the timbre of any voice to the trained target. It only changes the timbre of the voice, so for the illusion to work well, the source voice should have the same style, inflection, and general performance attributes of the target.

Thanks David for the update and clarification. I was aware that it was Cobain's song, but certainly not aware of some of the technology available to create the cover.

Good grief.

Thanks, David.

I’m more disturbed that someone could reconstruct a big band. I would think copying the arranging style of Nelson Riddle is much more difficult than faking the voice. So, the song itself with different lyrics does not exist sung by Frank Sinatra?

Perhaps the greatest +++Sinatra tribute+++ ever.

Yeah, that was my thought as well, but I was in a hurry to get somewhere when I made the post.

Paul Anka released the song with a big big band arrangement.

Which I should have known (or at least taken a moment to check), because I'm pretty sure have that CD... somewhere.

Or I could have paid more attention when I opened up the details on the YouTube post, since it had the CD displayed there (although they didn't say they had used it).

Using Spleeter to split off the vocals and backing tracks is a trivial drag-and-drop operation if you've got the Windows GUI version.

From there, it's just a matter of taking Paul Anka's vocal through an RVC trained on Frank Sinatra's voice. If there's no existing RVC model of Frank Sinatra handy, you can again use Spleeter to isolate vocals from Frank Sinatra songs and train an instance of RVC yourself.

Yes, that is making more sense. All this business about un-baking the cake is new to me. I did know about Spleeter and derivatives, but I didn't think they had progressed so far.

Spleeter is especially good if you want to listen to an album without the vocals.

The old method of vocal removal relied on removing the centered audio. It would invert the left channel and sum it with the right channel. Unfortunately, it would also remove anything else that was centered.

Being able to pull out the vocals is really useful for someone like me who's trying to mix tracks.

The isolated vocals makes it easier to hear what sort of doubling and background vocals there are, and how they are balanced.

Isolated backing tracks means getting a better feel for how the instrumentation. Plus, I can finally hear all those lovely instruments that are buried in the mix!

I am afraid there is nothing that can be done to stop this. Too many folks with too much time on their hands and nothing better to do. Plus, notice how the "ads" have jumped on this stuff.

But surprizingly I find this less objectionable. First, it really doesn't sound like Frank in this higher register and second, I like this song much better than Smells like teen spirit.

If that is true, then everything ever posted in the User Showcase should offend you, as people are "reconstructing" bands to back them. "Reconstructing" bands is exactly what this software is intended to do. There are styles that specifically state "Big Band", which I know because I used it to do "As Long As You're Gone."

I actually wish I knew somebody who did this deep fake process because the aforementioned song was made in the style of Sinatra, as well as my song "I Hope Somebody Cries" was done to channel my inner Johnny Cash. I would love to hear those songs sung by the artists I had in mind when I wrote them.

This AI technology is actually an opportunity for musicians with no band to have one. Isn't that exactly what PG music says when they market their product? The vocals in the AI deep fakes are astoundingly accurate and the nerd side of me absolutely loves it. The Fairlight CMI had the technology to sample vocal characteristics and sing lyrics you typed in with that voice. It was rarely used to my knowledge but I once booked an hour of studio time just to get a demo of the Fairlight and saw it used. (The guy ended up not charging me for the hour. He was actually proud of the flex opportunity to say "I have a Fairlight.")

As to the "disrespectful" aspect.... please. Respect is earned. I OWE nobody respect and nobody owes it to me.

And to take that "faking it" aspect one step further, you could also say that every band in the 60s and 70s was faking it by using The Wrecking Crew, which you could also say was the humanized version of what we now call AI generated music. Was The Wrecking Crew disrespecting the musicians they replaced on every song recorded over 25 years, which not by word but by deed implied "We are better than you. You all suck."? How do you think Peter Tork felt when he showed up with his bass to record and they told him "We don't need you today. We got this covered. You can just go home. We got it."

"Disrespect" is a nonsense word like "empower". Empowered implies that you need permission to do something. I am empowered to do anything I want to do. "We are going to empower children to learn about this." whatever "this" is. They are already "empowered" to learn it. They just don't do it.

So, disrespectful to Sinatra? No.

I guess David Seville was disrespectful to chipmunks! LOL!!!

I would like the AI SInatra Bohemian Rhapsody if the harmonies were there.

Yeah. It appears that some don't....

Although technically impressive, I put this usage of AI in the same category as a gimmick or low-brow entertainment.
So what if someone painstakingly copied a Sinatra sound, reproduced his voice and changed the lyrics for a single song. The real "gold" is for an application to seamlessly do this for any song with the user having full edit control of the output; and my prediction is that some company will provide exactly this within 5 years.

BiaB and AI could be a perfect marriage to allow the creation of fully editable custom styles and chord sheets (much like we have today) based on giving it a target audio sample. So if you want a Sinatra style, feed it a Sinatra audio sample. If you want a Frank Zappa, Chicago, chanting Monks meditation music or any other style, just supply the appropriate audio sample and the output would be a ready-to use SGU file. The days of scrolling thru and auditioning hundreds of pre-programmed styles would be replaced with laser focussing on the style you want.

This technology needs to move from the novelty/consumer/listener domain to the creative/editable/musician domain.

Real-Time Style Creation

In 2005, Paul Anka released on the album "Rock Swings". The album is covers of rock songs done in the big-band style.

The arrangement you're hearing in the video was created Brad Dechter, and played by studio musicians.

All that was done without AI.

The only thing that AI did was:

• Split out the vocal from the backing track; and
• Replace the timbre of Anka's voice with Sinatra's.

While that's technologically impressive, it's orders of magnitude less than the sort of thing you are describing.

You've got expertise in the field to base this prediction on?

Because that's not the sort of predictions I hear from people who are actually trying to accomplish this. They're saying there's a world of difference between working with temporally stable pixels and time-changing audio.

Predicting that audio will follow the same timeline as AI generative graphics ignores the huge difference between them.

David answered my question. The band in this example is a real band with real musicians and a very talented arranger who has mastered the style of Nelson Riddle. Not easy!

Eddie, I should have been more precise. It is this thought that AI emulated a real big band playing a Nelson Riddle arrangement that I was skeptical of. Apparently we are not there yet. I’m glad because I write big band jazz charts.

You DID.... LOL

I wonder... what musicians of 1890's thought of Pianola(Player piano)?
Was it something similar to what some people tend to think now about effects/influence of AI on music?

"an inventor in Germany, Edwin Welte, was working on a player which would reproduce all the aspects of the performance automatically, so that his machine would play back a recorded performance exactly as if the original pianist were sitting at the piano keyboard. Known as a Reproducing Piano, this device, the Welte-Mignon, was launched in 1904. It created new marketing opportunities, as manufacturers could now get the foremost pianists and composers of the day to record their performances on a piano roll. This allowed owners of player pianos to experience such a performance in their own homes on their own instruments, exactly as the original pianist had played it." - Wiki.

1904!

I certainly admire those skills that came before our time, but somehow would not be so certain that their delivery matched to quality of today's delivery.

Regardless, given time, these things will just continue to improve.

I play music live for a living in a duo with Mrs. Notes.

Most of my competitors are using Karaoke tracks. I make my own backing tracks. Sometimes with the help of BiaB, sometimes from scratch, depending on the song and how close I want to cover it. It's a lot more work than downloading a K track.

I guess I could accuse the karaoke track duos of cheating, but what would be the point?

My goal is to sound better than my competitors. I think my tracks sound better. That's my story, and I'm sticking to it.

Regarding bands using The Wrecking Crew, The Swampers, or other studio musicians to record their songs, I suppose that could be cheating too. But it sells records. The bands then have to learn the licks the studio musicians played when they go live. Even The Beatles couldn't reproduce many of their songs live. Is that cheating?

I've done some, but not a great deal of studio recording, but enough to know there is a difference between playing in the fish tank or playing in stage. Some bands also have the studio skills, but not all do.

I'm amazed at the deep fakes, and feel that ethically, there should be some notification. But I know even if one was required, it wouldn't be universally used, and there is no way to enforce it.

Respect? Isn't imitation the sincerest form of flattery?

When The Knickerbockers recorded “Lies”, it was so close in style that most people thought the song on the radio was actually recorded by The Beatles. Actually, I thought the band was tighter than the early Beatles.

We aren't going to answer the ethical questions here, although it's interesting to discuss them.

Me? I'm going to be amazed by the technology, and make a mental note not to believe what I see or hear. Then I'll go about my daily life, trying to get as much pleasure out of each day, for however many I have left. I have gigs to play, new songs to learn, food in the fridge, a roof over my head, a wonderful wife, I woke up this morning, and everything else is extra.

Insights and incites by Notes ♫

If one form of backing tracks is cheating, then ALL forms of backing tracks is cheating. You making your own doesn't elevate you to some higher plane of "cheating", despite your high opinion of yourself. (You failed to brag about your 15-20 gigs per month. You slipping in your old age? And the cruise ships. Don't forget the cruise ships.) Backing tracks in any form still results in keyboard and horn players being unemployed.

There was a female duo here that sang over tracks. Nobody cared because they were both very good singers, and both extremely hot to a level where none of the men in the audience actually came to hear them sing. Their tracks were nothing more than a mutual producer friend removing the vocals with software, so they literally sang over the original band sans vocals. Cheating? Or just a shorter way to get to the same finish line every backing track group gets to eventually?

Several years back I was meeting a guy who wanted me to do some keyboards for him. The place we met had a small stage and there was a guy setting up. I paid no attention to him beyond noticing movement on that little platform. When he started I looked over and saw he was playing bass. BASS!!! How DARE he present himself as a solo act when all he did was play bass along with backing tracks? I mean guitar solos were tracks!! Short of putting a bass solo into every song, what is entertaining about watching somebody play the easiest instrument in the world of backing tracks? I mean, it's BASS! As I once told the bass player from one of my bands when he started to get too full of himself about how good "we" sounded, "Dude, WE (gesturing in a circle toward the actual real, actual, playing pieces like the horns and the keyboards) sound good. YOU are just here. If you were to have a stroke and die before a show, there are probably 5 or 6 people right in the audience who could come up and play your parts." A bass player can be replaced by the keyboard player's left hand. But by my own logic, that would also be cheating too, because iIf you want bass, hire a bass player (and pay him 3/4 scale). If you want horns, hire horns.

It becomes clearer and clearer why I can't keep a band together, doesn't it?

You beat me to the punchline!

David, I once fired my girlfriend from the band we started together because she wasn't cutting it.

AND, somehow, we stayed together for about 6 months after that. Then the crazy became more obvious and I bailed.

Now that we got the Bass Players on the run, lets really humiliate them...

Well that was actually pretty amazing to watch ..

WOW that girl can play!

It shoots the hell out of Eddie's version of bass players doesn't it

Wow.

Y'all ain't seen her yet? Wait till you see her jamming with Steve Vai. On stage.

That girl was not in my band!

Yep, she's certainly talented alright.

I posted this one of her in the Best of YouTube section a few weeks ago.

That little girl is amazing.

BTW Eddie, since you asked. When we did cruise ships, we used the backing tracks I made myself, and set all-time revenue records in the lounge we played in. We were rewarded with a passenger sized cabin with a porthole, allowed to sell cassette tapes to passengers, and lasted 3 years on a 3-week contract which ended when Mrs. Notes' mom got sick, and we needed to jump-ship to become her caretaker.

But that's all off-topic

I mentioned various things that could be considered cheating and answered, “what would be the point?” That was to compare it with subbing Sinatra's voice over Anka's tracks.

Cheating or not isn't the question. My conclusion was:

So I guess you missed the point.

OT, years later when they sold the ship, the entertainment purchaser on the ships told me that our revenue record stood the test of time, and was the highest on that ship, and her two sister ships, ever. He asked if we wanted to come back on. I had so many land gigs, and we were still taking care of mom, so I had to say, "no"

Really? One cherry picked video negates what I said? Why not post Jaco himself? Or Vic Wooten? Or Larry Graham? Or go back to Stanley Clark, Charles Mingus, James Jamerson... That one kid who knows how to play doesn't undo how many horrible bass players I have seen in my decades of doing this, most of whom got into bands to drink on a band tab and get girls. (I mean, that's why I got into it....)

About 8 or 9 years ago I was at a wedding. The band played their dinner set and then took a break before the dance music started. The bass player ate a big chunk of a 6 foot sub that had mortadella in it. He apparently didn't know mortadella has pistachios in it. He DID know he was deathly allergic to nuts, and 2 songs into their show he started to swell up like a mutant from the planet Glorp and had to be rushed to a hospital. These guys were FRANTIC because the keyboard player couldn't play left hand bass while also playing the other side of the page too. I went up and told them "I can play the bass parts. Let's go on like nothing happened." When we finished about 3 songs the singer leaned over and told me "Wow. Not only CAN you play the bass parts but you are actually better than ________." Whatever his name was. I just smiled and said "I've been at this a little while."

I will double down on what I said. Most bass players that are not featured players like Jaco and Victor (to name just 2) are subliminal and it's the easiest instrument there is to play. It's only 4 strings (I am aware you can 5 or 6.) and the parts are not complicated. You CAN play complicated, but a bass player is supposed to play with the drummer and stay out of the way, so basics will do. Toss in a 3rd and a 5th every now and then (and just for Thump, maybe a 7th) as part of the measure before the chord progression changes and then get back to root notes. Those 2 sets I played had about 12-15 songs I did not know, but I told them to give me a key and count it in and I will feel where the changes are. You could do that too, Mario. Herb, David, EtcJoe... Any guitar player worth his weight in salt could do that. Now, with some experience in Motown and funky bands, I also tossed in some thumbing and in the songs that needed it some string snapping. Played a really nice Guild Starfire bass through a Hartke rig. With an MXR flanger and a Big Muff distortion box (that was set up WAY wrong).

And as a bonus for "whatever his name was", I didn't accept their offer to pay me. I told them to pay the guy his full share and consider it a sick day. The best laugh of the night was when we were talking during a break between the 2 sets I played and I told them some of the bands I had played in one of the young men said "Oh yeah! My father mentioned that band a few times." And in my mind I thought "Thanks a lot ya little [*****]!" He called his father and told him who was playing bass and he said "I remember him being the keyboard player in that band." and then of course I had to talk to dad for a minute and when the kid showed me a picture I did remember the guy despite that being 20 years prior.

Better and better condescending flexes every time you tell us how great you and that we all suck!

Happier and happier you didn't come to Herbstock.

You said that, not me.

Oh well, I'm having a very happy life. I hope you are, too.

Easy my friend, I was just pulling your leg on your previous bass is easy comments. Sorry if I ruffled your feathers.

I hear you about bass players. My wedding band had three long time members, two different drummers, but we went through about 10 or more so called bass players. We fired most because of your last sentence in the above message, i.e. they couldn't stay in the pocket with the drummer and they all tried to be a lead instrument!

How did we get from a computer mimic of Frank Sinatra to trashing bass players and people who use backing tracks is beyond me. Eddie wins for grumpiest poster again, though. The technology of deep fakes is going to be troubling if not already. Too many people believe everything they see and hear on their phones and elsewhere. The movie studios would gladly use AI to write their scripts and computer faked actors to play the parts if they were good enough and they will be someday.

I'm afraid on how deep fakes will be used during political campaigns. Its bad enough that they take snippets from a speech to make a candidate look bad but with deep fake they can post things that was never said. This is true for all political parties.

Now back to the main thread.

Repeat winner! So many people to thank....

My concern as well.

That's a lot worse than making Sinatra sing a Nirvana song.

There should be severe penalties for politicians who do that. Any falsification of your opponent(s) should result in immediate disqualification. But as we know, that will never happen.

Back on topic.

And yes, writers, actors, songwriters, musicians, and others will be replaced by AI, along with accountants, investors, news pundits, and so many others.

My question is this: “When AI and robots put half the work force out of work, who is going to have enough money to buy the products?”

Insights and incites by Notes ♫

You are right . . . as well as a profession that cuts close to home . . . engineers!!
For less money than the salary of a single engineer, you'll buy an AI bot that will do the work of an entire engineering department.

"The times they are a changin"

One of the many problems with AI is that it doesn't actually create - it merely recycles.

For example, it may begin an image as white noise, and then progressively refine it so that the error measure of the image falls below some given level.

The valuation of the error value is based on labeled training data that was supplied by people, using artwork created by people.

Often, that artwork was flat out stolen, with no profits being returned to the creators of the original artwork.

This holds true for AIs like code assistants, which have trawled through huge amounts of code, regardless of the license associated with that code. And it's the same for programs that correct grammar, or write term papers. The data comes from somewhere, but there's no attribution on the final product.

Expertise in a field comes not just from knowledge of what has happened, but what direction that points. I've got plenty of books on orchestration. While they have some value, the problem is that styles change, and they no longer reflect current practice.

By embracing AI of this sort, we devalue the works of artists who have innovated, and hand profits over to corporations who only seek to reap profit off the works of those artists while returning none of those profits to them.

Today, this may or may not be universally true; I'm thinking it's not true but of course it comes down to how one defines "create". And with advances in AI coming ever faster and faster, even if this is true today, it will unlikely be true "tomorrow".

There are AI systems in existence today that are designed to create/discover scientific hypotheses that humans alone haven't been able to create. Today's scientists and engineers collect far more data than is possible for humans (alone) to analyze and interpret. Properly trained and programmed computers can crawl thru terabytes of data and discover relationships amoung the variables at ever increasing speeds and usefulness. In fact, AI systems capable of generating hypotheses go back to the 1980s.

It has also been suggested that a category of Nobel (or other) Prize be set up specifically to award AI bots or AI/human collaborations when meaningful discoveries are made . . . stay tuned.

"In science, experimentation and hypothesis generation often form an iterative cycle: a researcher asks a question, collects data and adjusts the question or asks a fresh one. Ross King, a computer scientist at Chalmers University of Technology in Gothenburg, Sweden, aims to complete this loop by building robotic systems that can perform experiments using mechanized arms. One system, called Adam, automated experiments on microbe growth. Another, called Eve, tackled drug discovery. In one experiment, Eve helped to reveal the mechanism by which a toothpaste ingredient called triclosan can be used to fight malaria."
https://www.nature.com/articles/d41586-023-03596-0?utm_source=Live+Audience&utm_campaign=61e730f65f-briefing-dy-20231121&utm_medium=email&utm_term=0_b27a691814-61e730f65f-49938064

I predict we will see many more applications similar to this in the near future; this is the good side of AI.

PS> You never know what might be lurking in your toothpaste

I define create as something different than minimizing an error function.

We've had these for sometime now.

These systems aren't "creating" anything. They're evaluating. The solution is a result of running an algorithm. It's not like these systems have any abilities beyond running these algorithms.

That's hugely useful, but it's not the same are creating.

It seems to me that people making those sorts of suggestions don't have a good grasp of what they are talking about.

And what does it mean to award a computer program a Nobel prize? How do you give a computer program money?

These are only tools used by people to perform functions. They're not self-aware entities that have any understanding of what they are doing, and it's not a good idea to encourage people to think of them in those terms

These are programs that are designed by people to look for patterns in data. While they are called AI, the are not "artificially intelligent".

Executing an algorithm requires no intelligence.

They are capable of learning and classifying, but they are certainly not intelligent.

You can say that about quite a few songwriters and bands. Through my years of performing in cover bands, I've noticed a lot of bands and single artists doing 3 versions of the same song. 3 seems to be the magic number.

When I was working with Motown, Berry Gordy gave us the same talk he reportedly gave to others, I'll paraphrase. Don't try something new, write what is already out there and already a hit.

It has to be similar enough to be recognized and immediately be liked by the audience. While it has to be predictable, it must be different just enough so it isn't predictable 100% of the time. That little twist makes all the difference in the world.

What A! might not have yet, and might someday, is that little human spark of creativity that recycles what went before, but add just a bit of something new.

But when we write pop music, are we really doing something new, or a mash-up of more than one old thing?

It's doubtless that AI is coming after a lot of jobs. I just hope it doesn't take mine soon. I'm not a songwriter, I'm a performer. DJs have already taken too much of my market, I don't need A! taking the rest.

Notes ♫

“Something different than minimizing an error function”? This could mean almost anything. An apple is different than minimizing an error function.

How about this for a working definition of “create” in this context?

Create: To produce, reveal or give rise to something unique or sufficiently dissimilar from that which came before. To bring into being or awareness a novel thing, idea or relationship which did not exist prior.

“Create” and “discover” can be very close cousins.

Generative AI creates new content like images, text and relationships based on information it has learned from a training dataset and a set of rules. And it is a tool and has already surpassed the capabilities of a human. Tools that surpass human capability is nothing new in our history.

There are tools that surpass our crushing power; think sledgehammer.
There are tools that surpass our cutting power; think scissors.
There are tools that surpass our killing power; think rifle.
There are tools that surpass our lifting power; think hydraulic systems
There are tools that surpass our ability to move on the surface of water; think speed boat.
There are tools that surpass our ability to fly; think airplane.
There are tools that surpass our vision; think micro/telescope.

It is easy to connect the dots. Why couldn’t our brains be surpassed?

There are tools that surpass our ability to store and retrieve info; think database.
There are tools that surpass our ability to compute; think hand calculator.
There are tools that surpass our ability to strategize; think Deep Blue vs Gary Kasparov

And there is a tool that has already proven it can discover and create new and relevant relationships between variables that humans haven’t seen, we call this tool artificial intelligence and we ain’t even seen the tip of the iceberg yet. Drug discovery and protein folding are just two areas where this has and will shine.

And for the first time in our history we have a tool that can create a tool. ChatGPT4 can write software (in more than one language).

No doubt, this kid has me beat.
I wonder if anybody here can play a 5-string fretless as well as her.

PS> I like the new name Doctor, too bad "Steely" is already taken

All in fun! I wish we had a trophy to give out!

Joe, the laughs I get from these forums are trophy enough. I enjoy everybody here but one person. (The Flexor as I call him.)

Yes, this is indeed good news. Dan has graduated from a Student to a Doctor in a single day. Fast work, Dan! Congrat's.

Yep, I won't disagree about this.

Very little of what we do is original. Even things that are new - like inventions and discoveries - are found to develop in parallel in separate places.

But something that people can do - that AI can't - is understand which of their decisions are derivative, and which are not. For example, you can write a song in the style of another song, and know which elements can be safely retained as non-infringing (because they may be common across the genre) vs. things that need to be modified, because they hew too closely to the original.

AI can't currently do that, because the underlying technologies don't have a way to evaluate what they are "creating". There's no self awareness of either the process, or understanding of how to evaluate the results.

Training neural networks is based on minimizing error values. Iteratively generating images is based on minimizing error values. The core of neural networks is generating results with the lowest error value.

What I mean by "creates" is that there's an awareness of what's been created. Neural networks don't know about the algorithm they're running, or the data they are processing. It doesn't "know" if it's generating an image, or writing code. It's just applying learned weights to data.

"Rules" are typically inferred from the training data. Or they're hard-coded into the design of the neural network.

The training data is created by people, the valuation of the training data is created by people, and design and evaluation of the neural network is also supplied by people.

I also suggested that these systems are only just tools.

Because these machine allow us to exceed what we could do without them doesn't mean that these machines exceed us. A hammer without a person can't pick itself up and drive nails. Nor would it right to say that a sailing boat has "pluck and courage".

Then why attribute neural networks with the ability to create?

Yes, but that's not a difficult ability to surpass, wouldn't you agree?

I'm not saying that our brains can't be surpassed. I'm saying that neural networks aren't intelligent.

ChatGPT can just as easily "hallucinate" a complete false answer as generate a correct one. It has no ability to discriminate between the two, or even comprehend the meaning of its output.

We called lots of things "artificial intelligence", but that doesn't make the manifestations of "intelligence".

You mean, like a machine that can build a hammer? Because that already existed.

ChatGPT doesn't decide for itself that it wants to write a program, any more than a hammer-building machine decides that it wants to build a hammer.

Thanks David for bringing some perspective to the AI conversations. It's always worthwhile to take all sides of these situations into account.

That's a good assessment.

At least until AI becomes sentient. (Perhaps I've read too much science fiction)

Also, humans exhibit something called “taste” (the ability to discern what is of good quality or of a high aesthetic standard). Well, at least some do.

Example: Will this chord change sound better than this other one, in the context of the lyrics I wrote?

But for so many pop songs that use the same chord progressions, I guess that doesn't come into play. But perhaps the lyrics do. A little twist of the lyrics can make a big difference.

And even humans have their limitations. I've heard what scholars created for Beethoven's 10th symphony. They used notes and notation sch etches he had. The result sounded a lot like Beethoven's work, but paled in comparison because it didn't have the spark of creativity that he did with every symphony.

But AI is still growing. Who can predict how fast and how far? Will it someday write a symphony as creative as the masters and not be obviously derivative?

I guess we have to 'stay tuned' to find out.

I'm going to keep gigging until either it replaces me, or when Rap becomes the choice for the retirement audience here in South Florida. Then, I'll retire.


Insights and incites by Notes ♫

I know it may seem like I have come full circle with this topic. But it is clear to me now that the initial feelings I voiced regarding "disrespect" had nothing to do with AI or playing covers, it was entirely the selection of music. Frank singing Teen Spirit was simply wrong to my ear. Here we have an example of the same thing and this rendiction almost brought tears to my eyes. Somethink about the right song for the right singer.

The good news is that, so far, everyone that has contributed to this discussion has been "adults in the room".; knock on wood. Ever attend a dinner party where you are engaging someone at an intellectual level on a fascinating subject but then chaos breaks out because someone decided to bring their 10 year old twin boys who are wrestling under the tables? I have. Better for such children to play with other children so the adults can engage with one another.

Back to the subject.
If an artificial sweetner is a sweetner;
If an artificial hip is a hip;
If an artificial smile is a smile;
If artificial turf is turf;
And if artificial insemination is insemination;
Then, I claim artificial intelligence is intelligence, and this form of intelligence is getting stronger by the year. We can only guess at what has already been developed behind those thick 3 inch doors; the kind where you need to swipe your badge to get in.

To me this is reminicient of when some folk were convinced that the earth is at the center of the solar system; they were wrong.

Or "humans are the only creatures that can communicate"; wrong.

Or "humans are the only creatures that can fashion tools"; wrong again.

Note that intelligence is not equivalent to awareness, likewise intelligence is not consciousness. This may be the source of our differing views.

I've given you my definition of "create".
Care to give me your definition of "intelligence"?

Good point.
If I put my romantic, poetic, musical and perhaps mystical hat on, I hope that an AI will never create a symphony as good as Beethoven or any other master. I revere these brilliant musical geniuses; I can't see myself revering an AI bot.

But when I put my technocrat/engineering hat on; then logic, cognitive reasoning, an understanding of history and a quest for truth prevent me from putting my head in the sand and adopting a wishful thinking stance. Clear-eyed I must be.

Herein is the quandry. I suppose this is one price I pay for being a thinking, pondering human . . . or am I really a bot ? I wonder if anyone else has this split-thinking going on.

[Actually, I can't be a bot, a bot would play much better bass ]

It's been said that Shakespeare is regarded as a master of human emotion. I wonder what he and Einstein would have to say on this subject.

Did you know Ludwig very well?

You need to be able to separate the dancer rom the dance. Many people don't know what that means so I will explain.

You can go to a dance performance and it can be a sloppy dance routine, but the dancer performs it well.

You can go to see a band and they do a horrible original song, but the singer performs it will.

The bad song doesn't mean it's a bad singer.

That works the other way too. You never met Beethoven. Maybe he was a gigantic [*****]. That doesn't take away from his creation.

So what you should REALLY "revere" is his music. If an AI composing robot can create music of a caliber as the great composers, why doesn't it matter to you that bits of the master's phrasing, chord structure and pacing are referenced in creating a new piece of music?

You are FAR too hung up on a concept that I have yet to see you display a true definition or understanding of. If there is such a thing as artificial intelligence, doesn't logic say that there should also be artificial stupidity, (See: chatGPT)?

Look at a packet of lemonade mix. Everything on that packet is artificial flavoring, artificial sweetener... Yet look at a can of Lemon Pledge. It says "Contains real lemon."

So if AI can give a generation of people who loved Sinatra's work, and I am speaking of his work, not of the man, why deny those people more work that sounds exactly like SInatra despite him having died 25 years ago? You really want to get into some conversation about royalties due to a dead guy when no one cent of it would have gone to you? Why not just enjoy artificially created art rather than lament not having it available anymore?

I don't know what area of engineering you work in, but I see very little structural or electrical engineering done on a drawing board anymore. Only people approaching 100 years of age who have never accepted computers continue to hate them. I will never forget the day in 2002 when a managing partner at a law firm, then 73 years old, who called me on "new computer day" and told me "You have 5 minutes to get this appliance from hell off my desk." I was tempted to ask, though I wisely declined to do so, "So where do you park your horse and buggy anyway?"

Embrace technology and the possibility it offers rather than resist it. And keep using BIAB, which is essentially an AI composition tool.

Sure - "awareness" is a key element. There are many useful behaviors that have utility that don't necessarily demonstrate intelligence.

For example, my cat Nigel will perches on the edge of the litter box and poops onto the mat outside the box. He'll then paw some of the kitty litter into a pile, and leave the box.

He's clearly got some awareness of what's happening, but it's obviously more instinctual than intelligent.

Similarly, neural networks are trained to be able to solve specific problems in specific ways. Vision processing reduces the problem space down to pixels, while text reduces the problem space to words. Each requires a specific architecture, and is custom designed to work in the problem domain.

This is to some degree modeled after what designers of neural networks have found in nature. For example, there's a lot of processing that takes place in the eye before the information is passed onto the brain. And it's not simply a one-way trip - there's a lot of interaction between the layers.

But despite all this useful activity taking place - by my definition - the eye is not "aware".

Programs like ChatGPT can be trained to manipulate information, but they have no "awareness" of what that information is.

People used to thing that if a computer could play chess, it would then be artificially intelligent. There's an embedded assumption here: something must have intelligence to play chess.

But it turns out that you can create a computer program capable of playing chess without there being any artificial intelligence involved.

Or consider Arthur C. Clarke's Third Law: “Any sufficiently advanced technology is indistinguishable from magic.”

"Sufficiently advanced technology" is not the same as magic, it's just "indistinguishable" from magic.

Similarly, you can have a program capable of passing the Turing test, but that doesn't mean it's artificially "intelligent", merely that it can pass the Turning test.

I have a fairly solid understanding of how neural networks work. They are neither intelligent nor creative - their works are only derivative.

That's not to say that these systems aren't useful. But consider if you'd written a program to discover all the prime numbers between 1 and n. It's tedious work, but fairly easy to implement. The resulting system would be a useful tool for performing a task, but it's got no understanding of what it's doing. It's merely running an algorithm.

Similarly, AI tools are useful in performing much more complex work, but although they are much more sophisticated, they are still no more "intelligent" than a program that lists prime numbers.

I'll grant that my definition of "intelligence" is limited to my personal experience of intelligence. I'll also grant that I don't really know how my mind actually works, or if these mental models are just illusions.

On the other hand, simply because tools are useful and allow me to do things I couldn't do without them doesn't mean I'm willing to consider them partners in the work. The idea of granting a program co-ownership of a Nobel Peace Prize seems to me a fundamental misunderstanding of the current state of AI, granting it attributes that don't yet exist.

Intelligence is a hard nut to crack, and I don't think we're anywhere near a good understanding yet. Knowing what the current AI programs can't do helps us understand what we don't know about intelligence.

Hopefully that helped answer the question.

Unfortunately it didn't answer the question of what you think intelligence is. You provided several dots to connect but I'd rather not connect them and end up making inaccurate assumptions.

You did provide some interesting statements and examples, some of which I agree and others I don't. But we need a crisp working definition of intelligence to locate where our views diverge.

Maybe this will help. Here is my crisp definition for "create":

Create: To produce, reveal or give rise to something unique or sufficiently dissimilar from that which came before. To bring into being or awareness a novel thing, idea or relationship which did not exist prior. “Create” and “discover” can be very close cousins.

I'd request a similarly crisp 1, 2 or 3 sentence definition of what intelligence is from your perspective; not what it isn't, but what it is. Think of the form used by Merriam Webster or other but don't necessarily copy from them unless that is what you believe.

This is important because I believe you have an overly restrictive idea of what intelligence is, but I can't be sure until you articulate [crisply] what that idea is.

Interesting. Noting that there are only 12 notes (see what I did there?), the case could be made that everything since the people from the Paleolithic Period made tones through bone flutes has been derivative. If Grog the Caveman arranged those notes in a different pattern than what he had heard before, he "created". Now some mathematician could probably calculate the number of combinations of those 12 tones and come up with a huge but finite number. (Google says that number is 479,001,600.) Constrained by that, there could only be 479,001,600 completely unique songs that are not copied yet created, even if somewhat derivative. Is there a way that AI can squeeze out more than that finite number of combinations? Logic would say no. To follow that thought to conclusion, we have to ask if to this point in time there have been more than, less than, or exactly 479,001,600 songs written.

And does it matter is real intelligence (humans) or artificial intelligence (computer models) wrote them? Unless an AI model can be programmed to add inflection, that music that lacked those dynamics that a human performer can add, (I don't know if that is possible) that music would lack character and feeling.

I love this topic! I hope Frank would love it too.

Remember, somebody (Louis Washkansky) was the first human to receive another human's heart and a December day in 1963. That man lived 18 days and did NOT die from a problem with the heart, but from a lung infection and pneumonia. Remember that we now have artificial hearts that act as a bridge while a patient waits for a donor heart. If there can exist an artificial heart, why not "artificial" music?

Remember, in the early days of heart transplants 56 years ago, thinking, sane people wondered if putting a female heart into a male body would result in that man becoming a woman. (Insert your own Bruce Jenner joke here.) There is no evolution in any field without exploration of possibility. If somebody can create a "new' Sinatra song, bring it on. And Beatles fans who can't accept that John Lennon is dead can delude themselves into thinking that Now And Then is really a Beatles song. I didn't like "the song" itself, but I loved the concept and technology that allowed it to be created.

So use AI to bring back Frank, bring back Freddy, bring back John Entwistle, being back Buddy Rich, hell, go back to 1791 and bring back Mozart. Just do it well.

For the purposes of AI, I consider this to be an insufficiently "crisp" definition, because it lacks intent.

For example, and earthquake can "create" chaos, a flood can "create" ruin, and a river can "create" a canyon.

Yet none of these forces have intent to create any of those things.

Similarly, AI doesn't have the intent to "create" art - it is the result of running an algorithm. The intent was put there by the creator of the AI. The only way that the AI "recognizes" that it's created art is by applying criteria given though training (again, error minimization). But that criteria didn't come from the AI, it came from the people who chose the training data.

Replace the training data with randomly generated values, and the neural network will generate garbage. The program can't discern the difference, but the person monitoring the program will notice that it can't seem to bring the error term down. (It's hard to predict what something should be when the data is random).

I think you can safely believe I've got an overly restrictive idea of what intelligence is.

I've already given a key criteria for intelligence. That's not a negative definition, but it's also not a complete definition.

Intelligence isn't a single thing. I'll go back to the chess analogy. At one point, people thought only intelligent things could play chess. There are plenty of elements of the game that require intelligence. But it turned out that playing chess could be encoded into an algorithm. But having computers that could play chess did little to advance AI, other than make computer scientists realize the problem was much more difficult than they thought.

The ability for a tool to "create" useful outputs is a credit not to the tool, but the developer and the data. You're no doubt familiar with Musikalisches Würfelspiel, where music is composed with a lookup table and dice. Would you say the people rolling dice "created" music, and give co-authorship to the lookup tables? (Actually, it's very much like BiaB, but that's a different tangent).

I've been following AI for long enough remember SHRDLU, Shakey the Robot, and ELIZA. Remember when Prolog was going to take over the world? Makes me feel old.

AI has come a long way since then, but no amount of semantics is going to convince me that neural networks are intelligent. There's just too much that's invisibly supplied by people.

Ooh Whee you better believe I am going to read up on that and experiment with it.

Much like the time I got a paint by number kit with 40 colors and I threw them all into a bag, then made slips of paper 1 thru 40 and then picked a number slip and then a paint color and said "Okay. Now paint with the new numbers." The painting looked ridiculous but it hung on my office wall for like 3 years. With the orange grass and trees many shades of yellow, blue and red. That's just the kind of weirdness that I do, so dice deciding musical things is right up my alley.

Eddie, I would imagine some of those 479 million and change songs are probably awful!! There is a reason there are standards, standard chord progressions etc. Some things work really well and are pleasing to a large number of listeners while some are not. I am stating the obvious, of course. AI is only as good as the programmer and what they give it to "learn". Garbage in, garbage out still applies.

Yep! And I wrote most of them!!!

So what I'm observing is that despite the very real fact that you are unable to define "intelligence" you somehow can claim whether or not something is intelligent. I find this quite baffling. This is no different if I were to claim whether or not that object is a pop-up toaster but when pressed to define what a pop-up toaster is I throw up up my hands and say "I don't know !"

Vocabulary terms are the "atoms" of logic, and logical ideas are the "molecules" of science, and scientific principles and observations give rise to the "compounds" of truth. Without an agreed upon vocabulary you're stuck at the logic-less state. Key criterias are far less valuable than crisp definitions. With observations, this is how we know the earth isn't the center of the solar system. We crisply define earth, we crisply define center and we crisply define solar system.

Several years ago I was asked to write a book review for the National Science Foundation or some Academy of Science; I frankly don't remember (not important). But the book was titled "Writing in the Teaching and Learning of Mathematics" by Meier and Rishel; the book was written to help college-level mathematics professors build better curicula for their students. I wrote a very positive review and the upshot of the book might best be summarized by their words.

"Writing requires thinking; and thinking is what we, as teachers want to encourage. In what follows, we hope to convice you that to get students to absorb mathematics, or any other subject, better, you need to have them think about, then write about that subject".

Later in the book, they imply that if you can't describe (in writing) the complex ideas associated with multivariate calculus or differential equations in a way that your high school educated grand mother can understand, then you really don't understand the material. Clearly, vocabulary is "atomic" in this process. This little paperback was an enjoyable read all those years ago and has paid me dividends throughout my career and beyond. If you can't articulate it, you don't understand it.

You and I threw around some BIG terms like create and intelligence; and to ensure we could effectively communicate I met you halfway and gave my definition for create. When asked to define "intelligence" you are unable to. As we say in aerospace "We are experiencing a massive flame-out"; meaning it's time to pop the canopy and eject.

But before we end this and FWIW, since my college days I 've been facinated by psychology, brain science, and complex systems. During my career I trained multi-layer ANNs (artificial neural networks) and began studying emergent properties, cellula automata and complex adaptive systems. I am now interested in today's rendition(s) of AI, animal psychology, music and music therapy. I'm also interested in how all of these subjects interact on various continuums with what we call morals and ethics and what the near to medium term future might look like in these fields. As such, I was hoping that someone here with overlapping interests might want to explore these subjects on a semi-formal basis.

This is the second problem I have. Nothing can convince you. Not semantics, not logic, not truth. A calcified brittle-bone approach is not how good exploration is done, neither how good science is done. The fields of thermodynamics or structural mechanics for example are centuries old and can be considered mature. AI on the other hand is far from mature and might even be considered infantile. At this time, there is little convergence among theorists who have begun to study complex systems as a class. It is not a field in which a crisp and unified theory has already been developed; and I don't expect to see one in the next few years. But if you'd like to take a week, a month or even a year to ponder and articulate what you think intelligence is, I'd be happy to resume this conversation publically or privately. Some may come to this forum to die, I come to this forum to grow

One thing is for sure. In the big scheme of things, your opinon doesn't matter, neither mine. But the work of Sam Altman, Microsoft, Google and other movers and shakers does matter. And I'll be having fun watching and pondering what emerges. We are living in a truly exciting time.