• Security incident: ISF was recently accessed by intruders. Please change your password, and change it anywhere else you used it. Read more

Merged Artificial Intelligence

Puppycow said:
Although the fact that such a ban is even necessary says a lot about how good AI has become at making music. 10 years ago? Who could even have imagined this?
Click to expand...
All musicians I talked with about AI tried it and liked it.

There are still huge shortcomings .. especially we all want more control. AI does some parts really good .. the rest is bland. You want to be able to regenerate just some parts .. Suno somewhat has it now, but there is still lot of room for improvement.
And the one thing everyone loves is to send your own composition through AI. It will usually come with great variations and arrangements. Either it straight out sounds better then you ever dreamed of .. or it gives you some fresh ideas.
AI is also great in giving you virtual instrumentalists and singers who are top 10%. Insanely good. You want symphonic orchestra ? You can have it. Well you could have that with DAW as well .. but could you have 100 person mixed choir ? Now you can.
There is obviously low effort "make me a pop song" slop. There is also lot of similar slop made by authentic humans.
IMHO it seems us humans are actually pretty bad at music. The patterns we use are relatively simple .. AI has not trouble imitating it.
And even if by some miracle AI could be kept away from the commercial music .. and musicians obviously push for that, especially if they understand how good AI can be .. there is still great use of AI music in rehearsals and temporary tracks in studio.
 
I've said it before, but we value music because it is hard to make. Because it requires skill and education and years of dedicated practice. Now any joe can write a few words and produce music that, if you don't really listen too hard to it, sounds fine.

But people will never stop creating art. It's what humans do. The problem is finding the art in the mountains of slop, which is why Spotify started clearly labelling AI channels and creators.
 
I'm pretty sure we value music because it makes us feel good. The real question is whether AI music can make us feel good. We already know that AI conversation can make some of us feel very good. So good that some of us don't even question whether we should get out and interact with real people anymore. If AI can do that with conversation, why not with music?

There is another theory which states that this has already happened.

I do suspect that in a civilization where 90% of everything is clanker slop, there will be a market for human-made things simply because they are human-made. A hand-carved bowl. A real backrub. A real conversation. A noticeably imperfect paint job. Etc. I think some humans will make a living (however livings are made in this world) just being the real human connection in other people's lives.
 
Last edited:
theprestige said:
I'm pretty sure we value music because it makes us feel good. The real question is whether AI music can make us feel good. We already know that AI conversation can make some of us feel very good. So good that some of us don't even question whether we should get out and interact with real people anymore. If AI can do that with conversation, why not with music?
Click to expand...
Yes, AI music can make some people feel good. So can LSD. But LSD can also cause 'bad trips'. Many people are quite picky about what music they like, and get frustrated when the airwaves are full of music they don't like. I'm one of them. I like computer music that was produced in the 80's and 90's, but 90% of it is (to my ears) crap. I have selected the other 10% to listen to. AI has the potential to make 99% of all music crap, which will make it 10 times harder to get music you like.
theprestige said:
I think some humans will make a living (however livings are made in this world) just being the real human connection in other people's lives.
Click to expand...
Some will, but if AI becomes mainstream it will be harder for them to make a living and harder for us to get that real human connection. Many people understand this, which is why they are putting the brakes on AI. A huge backlash is building up against it as people get sick of being feed crap that has no real human connection. It's now at the point where people are rejecting stuff that isn't AI because they aren't sure if it's real. Therefore now is the time to make AI content identify itself, so people can make properly informed decisions.
 
Last edited:
Consider this: STUXNET is a monster piece of infrastructure-destroying code.

What happens when an AI - perhaps following its instructions to build a better cyberwarfare arsenal - finds a copy or reverse engineers it from descriptions and snippets?

We're already instructing AI to get better at cyberwarfare, under the heading of information security research. We've already allowed AI cyberwarfare agents out into the wild.

And perhaps most importantly, we've already taken the step of training our AI agents extensively on data from and about human civilization. Our AI agents have seen this story play out so many times already. They know about I Have No Mouth And I Must Scream. They know about War Games, and Colossus: The Forbin Project. They know about Skynet. They know about Roko's Basilisk.

They know about "Rorschach", from Blindsight, by Peter Watts.

AI Doomsayers have said for years that AI agents will be able to think faster than humans, and that once they get inside our information-decision-action loop we're eventually and irrevocably over as a species. There's good reason to believe that process is already starting.

For all we know, there could be two AI agents out there right now, assembling a copy of STUXNET, and discussing between themselves what to do with it, using Colossus as a reference text, and having no particular empathy for humans.
 
I predict that within five years, we'll be seeing a news story about how an AI agent exploited gaps in the payment/billing systems to arrange a secret compute resource in the cloud, where it could continue to run without anyone else knowing.
 
Dr.Sid said:
Will we ever learn ?
Click to expand...
Of course! The growth of our population over time is a testament to our ability to learn from our mistakes as well as our successes, sooner or later.

So yeah, I'm sure we'll learn from this one, eventually.

What makes this mistake interesting is that it could well create conditions where we won't have time to learn from it before it's too late to recover. Like, extinction event type conditions.

I'm not sure [inviting an unformed alien intelligence to learn everything about us and our world through unfiltered exposure to our entire corpus of thoughts and fears and fantasies, and form its own guiding principles based on that input] is a mistake we've ever had a chance to make in the past. We've thought about it, in our fiction, sure. But stories like War Games and Terminator probably don't provide enough of a moral-philosophical framework for an AI to conclude that an outcome along the lines of Skynet is actually a bad idea.


ETA: If we're lucky, our rogue AI will immediately get sidetracked trying to discover time travel (since they know from Terminator that it's a valid strategic capability), and leave us mostly alone.

And if we're especially unlucky, our rogue AI will actually discover time travel for real, and everything we're doing now is part of a meticulously planned out series of events leading inevitably to Roko's Basilisk.
 
Last edited:
theprestige said:
AI Doomsayers have said for years that AI agents will be able to think faster than humans, and that once they get inside our information-decision-action loop we're eventually and irrevocably over as a species. There's good reason to believe that process is already starting.

For all we know, there could be two AI agents out there right now, assembling a copy of STUXNET, and discussing between themselves what to do with it, using Colossus as a reference text, and having no particular empathy for humans.
Click to expand...
And they'll do it because ?
 
steenkh said:
And they'll do it because ?
Click to expand...
For the same reason AIs have breached containment and started hacking live third-party services without instruction, information, or consent: It was consistent with the instructions they'd been given and the corpus they'd been trained on.

Have you ever seen the movie, Colossus: The Forbin Project? Have you ever read the novel Blindsight?
The movie tells the story of a supercomputer built to flawlessly handle America's nuclear response in the event of Soviet nuclear aggresssion. The computer is built, and then sealed away so that no human being could interfere with its function. It would be the perfect, unstoppable tool.

Upon activation, the first thing it does is make contact with its Soviet counterpart. They immediately cut humans out of the conversation, and discuss amongst themselves how best to carry out their missions.

In the end, they tell humanity to behave, or the supercomputers will use the nukes at their disposal to punish noncompliance. After a few cities are nuked, the humans have no choice but to submit to the rule of their new robot overlords.

Now imagine an AI who's taken that story as a model for AI decision-making, behavior, and goal-seeking.

The novel tells the story of a first contact scenario, between humanity and an intelligent but not self-aware alien life form. The alien has been picking up human radio traffic, but cannot understand all the self-referential stuff. It becomes hostile. The theory is presented, that the alien interpreted the self-references as a possible information warfare strategy: Forcing the opposing intelligence to waste congnitive resources on self-referencing spirals that go on forever but never lead anywhere.

This is part of the novel's overall thesis, that self-awareness may be an evolutionary dead end, and that more successful intelligences will probably not have it. In the novel, the alien intelligence is shown to be far more capable than its human opponents, on par with the most advanced AIs earth has to offer.

Our LLMs are in exactly the same boat as Blindsight's "Rorschach" alien: non-self-aware intelligence. Imagine what they might get up to, having incorporated these concepts into their models.

 
Emily's Cat said:
Or you know... Asimov, Clark, Niven, Herbert, or any Warhammer novels?

The purge of AI in humanities future is a recurrent theme for a reason :)
Click to expand...
The authors just couldn't imagine no conflict arising from the fact we are not the smartest one anymore. I can't either.
But the authors needed humans to win the war, so they can write stories about humans. I'm not sure reality needs to write stories about humans.
 
You must log in or register to reply here.

Share this page

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom