There’s promise of modern technological progress that echoes the Industrial Revolution. That era promised we could will into existence machines that would make our lives better, or at least more productive. The reality was most inventors were the wealthiest in our society.
Most inventors weren’t George Washington Carver or Nikola Tesla, with fairly altruistic visions for helping the most vulnerable Black and immigrant communities. They were Thomas Edison and others like him who could afford to create “automatons” to improve efficiency in factories and farms and exploit the labor of their employees and Black prison slaves. That reality still holds true for most software inventions today.
The promise of software automation is fundamentally the same as the industrial era. Somehow, by automating our world, we expect new software will be a force multiplier, making us more productive and our lives easier. We expect this will create gains for us – less strained work weeks and improved working conditions – and help strengthen the case for workers’ rights.
Those rights are central to training data, a fundamental component of today’s Artificial Intelligence (AI) software. ChatGPT, Midjourney, Stable Diffusion, and many other popular new AI products all rely on training data collected from the published works of human creators. The labor of people working in journalism, publishing, and digital and analog art generated the data to produce these AI software.
Artificial Intelligence in a Minute: A Quick Primer
AI software attempts to approximate human intelligence to the degree that a reasonable person cannot distinguish between it and a human. Machine Learning (ML) is focused on cognition – learning, memory, knowledge, and the like. There are other types of AI, but we’re focused here on specific types of ML. There’s a debate whether Artificial Intelligence or Machine Learning are more appropriate terms for the software being developed. There’s an argument for both, and AI is defined so broadly it includes ML within it.
ChatGPT, Stable Diffusion, and similar are categorized as Generative Artificial Intelligence (GAI), which is largely learning from training data. Though my personal opinion is that these products are ML, the marketing and general discussion around them uses the term AI. It is important to make the distinction since marketing is such a source of confusing terms.
Marketing for these products implies they are something like human intelligence, with all the implications of full human thought. This is to raise the perceived profile of these products in the eye of the public and potential investors, leading people to value these products more than just next-generation machine learning.
Almost all machine learning works by taking in a data set, using to statistics to weigh various unique parts of the data, and storing those values. This is an over simplification, but it’s essentially how Netflix knows which shows to suggest, YouTube swirls you into an autoplay drain of videos, and Facebook influenced Myanmar.
The origin of this data is particularly important. The first few generations of machine learning crowd-sourced the data from interactions with their software platforms. We engaged in social media, watched movies in their apps, and sent tweets or comments to each other all day long.
The Next Generation
This next generation GAI still crowd-sources from publicly available information, but it doesn’t come primarily from our interactions with their non-AI software product. Now it comes from all over the internet ‐ online art galleries, social media, blogs, magazines, newspapers, and more. Little, if any, of the data is owned by the creators of the GAI products.
It’s important to understand that this generation of machine learning is a huge leap. The data is not the only special part of these products. However, arguing about the mechanisms that make these AI special is a distraction from the problematic source of the training data and the problematic ways the wealthy can take advantage of these AI products.
Much like Edison stealing credit from all the scientists he worked with (most famously Nikola Tesla), these GAI products were developed with no way to credit human creators. GAI constructs whatever it is prompted, drawing from the training data. Today there is no way to know what sources ChatGPT used to generate that blog article or what art contributed to the beautiful generated picture in Stable Diffusion.1
Real people spent time, effort, and money to think deeply about their craft and produce something valuable to them in some way. GAI hijacks their labor to copy and paste small pieces of data into something else. Much like the wealthy inventors of the Industrial Revolution, GAI exploits others’ labor in order to create value, with little or none of that value offered back to the laborers.
This labor exploitation happens in two different ways. The first exploitation is the training data – remember GAI is not true intelligence but instead machine learning. It further exploits human labor by requiring prompts to generate from the training data. Without people spending time thinking about what they want from the GAI, it cannot generate anything.
Generative AI bakes labor exploitation into the software. More and more businesses are moving to fire people paid to create art and writing and replace them with the machine learning that uses their work as training data. This is not in any way ethical. Maybe it’s legal, I am not a lawyer, and maybe the laws need to change.
Workers’ rights are central to all of this. Especially when you consider companies making people compete against generative AI. Companies are already firing large numbers of software engineers, copywriters, journalists, artists, and graphic designers in favor of using GAI to replace their job functions
This is why I started out talking about the Industrial Revolution. This period of labor exploitation led to civil unrest and actual, literal blood-shed between employers and workers. It eventually resulted in organized labor, government action, and changed laws. Blood was spilled to secure the 40 hour work week and recognize real rights for workers.
We cannot afford systemic classism, racism, misogyny, fascism, and all manner of labor exploitation to be automated by generative AI. It has already happened once with ChatGPT and Stable Diffusion. These early predecessors were very prone to generating Nazi imagery and racist text, for example. We should be protecting workers’ rights here.
This generation of AI, and whatever comes next, is not designed to amplify the abilities of existing and future workers or to secure their economic future. Instead, it’s being used to extract value from their past or unpaid labor in order to replace their current labor. This is mostly due to current attitudes about capitalism, market forces, creative destruction, and how companies create value for shareholders.
That is the current reality, but it doesn’t have to be this way. Instead, this kind of AI could be used as assistive technology. It could be trained by people who consent to participate in the data and operated by people who benefit from a productivity boost for either themselves or their employer.
We need to work towards a future where AI isn’t used as an excuse to make human beings redundant. We need to work with each other and advocate that local and federal government recognize this threat against workers’ rights. Like the song:
It is we who plowed the prairies; built the cities where they trade;Pete Seeger, Solidarity Forever
Dug the mines and built the workshops, endless miles of railroad laid;
Now we stand outcast and starving midst the wonders we have made;
But the union makes us strong
If we don’t take collective action, we will rapidly find ourselves “outcast and starving midst the wonders we have made.” Make no mistake, it will happen rapidly because we’re facing 50 years of conservative political action to erode workers’ rights and a machine learning product capable of learning faster than human beings. Finding ourselves in poverty and homeless, if a real threat, will be sudden and unexpected.
Remember though that today these AI models require prompts. Laborers still have to turn the wheel of the product. The song continues:
They have taken untold millions that they never toiled to earn
But without our brain and muscle not a single wheel can turn
We can break their haughty power, gain our freedom when we learn
That the union makes us strong
In our hands is placed a power greater than their hoarded goldPete Seeger, Solidarity Forever
Greater than the might of armies, magnified a thousand-fold
We can bring to birth a new world from the ashes of the old
For the union makes us strong
- As of February 7th, 2022, a company named Chroma began work on a way to identify artists contributions to generative AI images. It is in early stages.
- Christian Power - March 7, 2023
- “We Know Best”
Abuse-Enabling Christian Leaders Playing God - February 10, 2023
- A New Industrial Revolution
Artificial Intelligence And Modern Labor - February 7, 2023