Members of SAG-AFTRA and the National Association of Voice Actors united at San Diego Comic-Con to address artificial intelligence and how it can harm creators.
People don’t understand that it’s not AI, it’s machine learning. Imagine you have someone alone in a room. He takes the experience and knowledge of people and puts it into his library. When someone asks for something, it uses what it’s got in its library to answer that person’s question. But that person stays in his room, he doesn’t experience life.
These AIs are like that. They feed on human creativity. Human experience of life produces creativity. These AIs do not experience life and do not think, so they can cannot replace humans.
I mostly agree with you but think it’s important to clarify that even with machine learning many humans can be replaced.
To extend your metaphor, that library has always had a bunch of clerks sitting inside of it. They’ve been handling requests, finding books, and organizing them into a system that works to best serve that information.
Now with machine learning, instead of having all of those clerks making the library run smoothly, they’ve effectively replaced 99% of all of the humans with an organizational system that serves content and helps find books even faster than a human would be able to.
Slightly deeper: this machine learning replacement can also now mix and match bits of content. The human system before might have a request that looks like this - “I want information on Abrahamic Religion in Western Culture” so they’d gather up a ton of books and pass them to the person that requested info.
In the new replacement system, the request could take bits and pieces from all of those books and present a mostly comprehensive overview of Abrahamic Religion in the West without having to run and fetch all of the books.
Deeper yet, and the scary iceberg - today, someone still needs to write all of those books and we as a society tend to trust information gotten from those books (cited sources and all that) so humans are safe as the content authors right? We’ve basically just made a super efficient organizational and content delivery system. But as we start to trust the new system and use it more, we’re potentially seeing the system reference its own outputs as opposed to the source material…which creates a recursive, negative feedback loop.
We still need human content creation today, but the scary part (IMO) is when we treat these LLMs as generative general AI. The LLMs are fallible and can be incorrect and often hallucinate - so when most people start blindly trusting these systems (they already do - look no further than general confusion on the terms AI and machine learning and LLMs), we’re going to get increasingly further away from new knowledge generation.
I think we are on the same page. There is a sociological concept of Generic Worker and Self-Programmable worker by the sociologist Manuel Castell. The self-programmable workforce is endowed with the ability to retrain and adapt to new tasks, new processes and new sources of information, as technology, demand and management accelerate their pace of change. Generic labor, on the other hand, is exchangeable and disposable, and coexists in the same circuits with machines and unskilled labor from all over the world.
Generic workers are already being replaced by automation (robots), but now LLMs are threatening self-programmable workers. The only way to adapt to the new reality is to become indispensable in training LLMs. It will completely upend the current job market as we know it. And as you said, the danger is if we treat LLMs as generative AIs.
The fact that this is upvoted so much is just sad.
While at face value it appears to be critical of AI, and thus bandwagoning on a very popular slant these days online, the inherent anthropomorphizing of the model in question is extremely wrong in so many ways.
LLMs are trained to complete human thought. And as a result, that very narrow class of machine learning ends up being oddly good at seeming human in responses.
But a diffusion model for generating images? Or text to voice generation?
To anthropomorphize these models is like saying that your cell tower triangulating your position won’t care about you as much as your mother would.
It’s just incredibly bizarre.
It is going to get better and better at replicating human speech patterns, and is going to be able to be further customized in how it expresses sounds mimicking human emotions. Already it can get uncannily good off just a few seconds of a sample.
As for the actors - as soon as residuals get figured out such that they get paid per hour of secondary usage of their recordings, they are going to go from “I’ll never deign to let AI replace me” to “yes, of course I’ll let you pay me more for me to do less work.”
The creativity of Matt Mercer in deciding on as frightened goblin voice for an innkeeper is going to be years before an AI successfully replaces that contribution.
But for Matt Mercer to provide samples of many different voices to an AI which pairs with GPT-5+ to DM your DnD campaigns with that voice pack for a monthly fee he gets a large cut from?
That’s not only going to be extremely possible sooner than you might think, but you’ll be seeing serious voice actors falling over themselves to directly market their voices to main street for personalized content.
It’s all about economical fairness, and those rigidly protesting change that endangers the status quo are very much like the MPAA fighting Napster instead of funding its successor - who as a result left a clear victory open to Apple and then Spotify and others by resisting change rather than embracing it.
People don’t understand that it’s not AI, it’s machine learning. Imagine you have someone alone in a room. He takes the experience and knowledge of people and puts it into his library. When someone asks for something, it uses what it’s got in its library to answer that person’s question. But that person stays in his room, he doesn’t experience life.
These AIs are like that. They feed on human creativity. Human experience of life produces creativity. These AIs do not experience life and do not think, so they can cannot replace humans.
I mostly agree with you but think it’s important to clarify that even with machine learning many humans can be replaced.
To extend your metaphor, that library has always had a bunch of clerks sitting inside of it. They’ve been handling requests, finding books, and organizing them into a system that works to best serve that information.
Now with machine learning, instead of having all of those clerks making the library run smoothly, they’ve effectively replaced 99% of all of the humans with an organizational system that serves content and helps find books even faster than a human would be able to.
Slightly deeper: this machine learning replacement can also now mix and match bits of content. The human system before might have a request that looks like this - “I want information on Abrahamic Religion in Western Culture” so they’d gather up a ton of books and pass them to the person that requested info.
In the new replacement system, the request could take bits and pieces from all of those books and present a mostly comprehensive overview of Abrahamic Religion in the West without having to run and fetch all of the books.
Deeper yet, and the scary iceberg - today, someone still needs to write all of those books and we as a society tend to trust information gotten from those books (cited sources and all that) so humans are safe as the content authors right? We’ve basically just made a super efficient organizational and content delivery system. But as we start to trust the new system and use it more, we’re potentially seeing the system reference its own outputs as opposed to the source material…which creates a recursive, negative feedback loop.
We still need human content creation today, but the scary part (IMO) is when we treat these LLMs as generative general AI. The LLMs are fallible and can be incorrect and often hallucinate - so when most people start blindly trusting these systems (they already do - look no further than general confusion on the terms AI and machine learning and LLMs), we’re going to get increasingly further away from new knowledge generation.
I think we are on the same page. There is a sociological concept of Generic Worker and Self-Programmable worker by the sociologist Manuel Castell. The self-programmable workforce is endowed with the ability to retrain and adapt to new tasks, new processes and new sources of information, as technology, demand and management accelerate their pace of change. Generic labor, on the other hand, is exchangeable and disposable, and coexists in the same circuits with machines and unskilled labor from all over the world.
Generic workers are already being replaced by automation (robots), but now LLMs are threatening self-programmable workers. The only way to adapt to the new reality is to become indispensable in training LLMs. It will completely upend the current job market as we know it. And as you said, the danger is if we treat LLMs as generative AIs.
The fact that this is upvoted so much is just sad.
While at face value it appears to be critical of AI, and thus bandwagoning on a very popular slant these days online, the inherent anthropomorphizing of the model in question is extremely wrong in so many ways.
LLMs are trained to complete human thought. And as a result, that very narrow class of machine learning ends up being oddly good at seeming human in responses.
But a diffusion model for generating images? Or text to voice generation?
To anthropomorphize these models is like saying that your cell tower triangulating your position won’t care about you as much as your mother would.
It’s just incredibly bizarre.
It is going to get better and better at replicating human speech patterns, and is going to be able to be further customized in how it expresses sounds mimicking human emotions. Already it can get uncannily good off just a few seconds of a sample.
As for the actors - as soon as residuals get figured out such that they get paid per hour of secondary usage of their recordings, they are going to go from “I’ll never deign to let AI replace me” to “yes, of course I’ll let you pay me more for me to do less work.”
The creativity of Matt Mercer in deciding on as frightened goblin voice for an innkeeper is going to be years before an AI successfully replaces that contribution.
But for Matt Mercer to provide samples of many different voices to an AI which pairs with GPT-5+ to DM your DnD campaigns with that voice pack for a monthly fee he gets a large cut from?
That’s not only going to be extremely possible sooner than you might think, but you’ll be seeing serious voice actors falling over themselves to directly market their voices to main street for personalized content.
It’s all about economical fairness, and those rigidly protesting change that endangers the status quo are very much like the MPAA fighting Napster instead of funding its successor - who as a result left a clear victory open to Apple and then Spotify and others by resisting change rather than embracing it.