Artificial Intelligence: Discussion and demonstrations

[Lord Ratner]

It's getting a bit difficult to keep up with The many new manifestations of artificial intelligence, but if you do there's a new one out for open testing that is fairly astonishing, if or nothing else than for the simplicity of producing something functional.

 

The service is called Udio, and right now you can make a whole bunch of 30-second clips just to play around with the technology. Here is a song clip I created with my standard AI testing subject: a dinosaur that falls in love with a toaster.

https://www.udio.com/songs/hveSQTSVkKQ5NnfCUx5Q2s

 

However as I was playing around with this thing, and based on the market action over the past couple years, I'm coming to the conclusion that AI is going to have big and profound social effects on the world (fake news will have a whole new meaning), but I don't think it's going to be the economic game changer everybody is making it out to be, and I think the Nvidia stock is going to be the poster child of this frenzied bubble. That's not to say I think Nvidia is going anywhere, I just think they're going to be the next Cisco of the early dot com bust.

 

This technology is amazing, but even before we get into the inevitable landscape of government regulation, it just doesn't seem to me that it will be capable of creating something new, which is ultimately what produces generational leaps forward in wealth. What we have right now is a hyper sophisticated search engine that can output the results in a wide range of multimodal formats.

Anyways, if you have some time put together some songs and share them here, and if people are interested this thread can be a repository for the many different types of AI products that are at least amusing, if not particularly useful.

[Danger41]

Your description of a a "hyper sophisticated search engine" is very accurate. I've worked on a bunch of AI things and the annoying part about AI is the assurance that people think it will get to "general" artificial intelligence because it's growing so fast. It's very similar to saying we'll have time travel in the next 10 years because tech is changing so fast. 

https://www.hup.harvard.edu/books/9780674278660 - "The Myth of Artificial Intelligence" is a good read to understand this subject. 

[Lord Ratner]

2 minutes ago, Danger41 said:

Your description of a a "hyper sophisticated search engine" is very accurate. I've worked on a bunch of AI things and the annoying part about AI is the assurance that people think it will get to "general" artificial intelligence because it's growing so fast. It's very similar to saying we'll have time travel in the next 10 years because tech is changing so fast. 

https://www.hup.harvard.edu/books/9780674278660 - "The Myth of Artificial Intelligence" is a good read to understand this subject. 

Yup. There's some interesting discussion about using the "holes" in the weights to use AI to discover new things. Basically, the AI uses data to find the most common connections and build answers based on those weights. Reverse it and you can find "answers" that are the opposite... the least common connections. Problem is, how do you filter good things from an infinite supply of "wrong" answers?

 

Anyways, using Udio I finished a full-length song that I think is better than most country music produced today. Enjoy!

 

Crumbs and Chrome Hearts
https://www.udio.com/songs/5g1enYTaTmMCYMbrAxzDVi

 

 

[Standby]

17 hours ago, Lord Ratner said:

Crumbs and Chrome Hearts
https://www.udio.com/songs/5g1enYTaTmMCYMbrAxzDVi

 

Wow, a real tear jerker tune there. This particular lyric really spoke to me: “In this chrome heart, you’ve got dino-sized room I'll be your knight in scaly armor, dispelling the gloom Through the steam and crumbs, our hearts beat as one It's a Jurassic love that can't be undone.”

Really though…this tech is wild. That song is better than 90% of the trash played on the radio, and it’s about a dinosaur that loves a toaster ha!

[ViperMan]

@Danger41 is spot on. There is a lot of hype and misunderstanding about what AI is and about what it can do. The discussion about artificial general intelligence (AGI) is more theoretical. IMO AGI is not possible, because no matter how sophisticated the output seems, a computer is still just a really really fast abacus. In order to admit a computer into the realm of the "intelligent" you simultaneously need to admit an abacus into the same category. I don't think many people would be comfortable with that leap.

Really the whole discipline suffers from having ever been associated with the word "intelligence" in the first place as it begets consciousness, which a computer can never be.

What AI is going to do is make a lot of previously seemingly intractable problems solvable, but all it really is at rock bottom is advanced math (statistics) being applied to lots of high-dimensional data. Computers are good at solving things like that. People not so much. Once you understand that, the magic disappears.

[Swizzle]

6 minutes ago, ViperMan said:

@Danger41 is spot on. There is a lot of hype and misunderstanding about what AI is and about what it can do. The discussion about artificial general intelligence (AGI) is more theoretical. IMO AGI is not possible, because no matter how sophisticated the output seems, a computer is still just a really really fast abacus. In order to admit a computer into the realm of the "intelligent" you simultaneously need to admit an abacus into the same category. I don't think many people would be comfortable with that leap.

Really the whole discipline suffers from having ever been associated with the word "intelligence" in the first place as it begets consciousness, which a computer can never be.

What AI is going to do is make a lot of previously seemingly intractable problems solvable, but all it really is at rock bottom is advanced math (statistics) being applied to lots of high-dimensional data. Computers are good at solving things like that. People not so much. Once you understand that, the magic disappears.

Perhaps because AGI needs an input, rather than already having a purpose. A purpose which is its reason or input-into-life's-abacus. 

Also, any self-fulfilling purpose of an AGI is just scary because of its potential, speed, and limited understanding/control...which is what makes Terminator series so entertaining.

Begs the question: does a street-smart AI = AGI? Or book smart?

[ViperMan]

Neither. Intelligence requires consciousness. It's nonsensical to call an abacus conscious.

[Lord Ratner]

10 hours ago, ViperMan said:

Neither. Intelligence requires consciousness. It's nonsensical to call an abacus conscious.

The current "AI" iterations are not AGI, however they are closer to functioning like a human brain than we have ever gotten before. The irony here is that we didn't accomplish this by figuring out how the brain works, quite the opposite, we created an array of associations that is as mysterious to us as the individual neural pathways of a human brain is. 

 

We know that the brain has a combination of biologically-arranged pathways (e.g. for walking, breathing, visual identification of faces, eye position, etc.) and experience-formed pathways (math, music, flying a plane). Right now the AI models crunch tons of mostly-unfiltered data into a model that we do not have the ability to directly adjust because of the sheer volume of parameters, then an overlay is used to do things like prevent swearing, giving directions for bomb making, etc. But this is in it's infancy. Once you can pre-program certain behaviors into the actual model, then leave the rest of the model to continuously adapt the weights based on new data, we will take another big step to AGI. 

But since we have precisely 0% knowledge on what is or what causes consciousness, it is entirely possible that we reach a point where AGI is achieved simply by running the models with enough horsepower that we stumble into the solution. That is, incidentally, remarkably similar to how evolution works. 

 

Are humans the only animals that are conscious? What about dolphins, octopi, crows, or chimps? If not, does that mean there is no intelligence other than human intelligence? That seems like an arbitrary definition. Where is the line, and how smart does a computer need to be before it is considered intelligent? Smarter than all humans, or just smarter than any human? Do people born with Down Syndrome have consciousness? What if an AI surpasses the intellectual ability of someone with DS?

 

Calling even the current models an "abacus" is like calling the human brain a glutamate sensor. Sure, it's kind of true, but it's the scale of the apparatus that makes it interesting. 

Edited April 18 by Lord Ratner