There are countless legal challenges being fought around the globe between IP holders and AI development firms. At this point, it must feel like a rite of passage when an AI company gets sued for IP theft for allegedly using copyrighted material without permission to train models.
Unfortunately, however, the discussion over whether it’s okay for an AI company to train models on IP without permission is moot. Sure, there are legal issues to be sorted, damages to be awarded or dismissed, and lawyer fees to be paid.
But you can’t untrain an AI model. Big tech followed the old adage that it’s better to beg forgiveness than ask permission — and it worked. The models are here to stay.
I could cite precedent, analyze the ongoing court cases, and seek out a legal expert to explain the real cost of these lawsuits. But why waste my time and yours?
Here’s what you need to know:
As of Nov. 14, the five most valuable companies in the world, Amazon, Apple, Google, Microsoft, and Nvidia, were all entered into individual strategic partnerships with AI industry leader OpenAI.
Any threat to OpenAI’s development paradigm is a defacto threat to the entire LLM industry. There are no alternative methods for developing ChatGPT-level AI models without using IP or powering them with models that were previously trained on IP.
In other words: If OpenAI and all the companies it is partnered with have to shut down their models or operate under a business structure that requires fair use payouts to artists, then the entire tech sector is likely to suffer a massive market crash.
The combined worth of these five firms is equal to around 15% of the total global GDP.
To put it mildly, it seems unlikely that the courts will ultimately forbid big tech from training models on IP without permission or that governments will step in to intercede on artists’ behalf. And that means AI-generated art and all the ethical implications that come with it are here to stay.
The core ethical question we should ask then, in my opinion, is whether AI-generated art is copied from or inspired by humans.
Here, I’ll argue that AI cannot create. It can only copy.
AI models are trained on massive datasets full of human-generated data. They silo that data into tensors that are weighted with preferences during a pre-training transformer cycle and activated using brute force mathematical processing.
It works because math is immutable. If you created an AI model that was exactly like ChatGPT in every way except it could only output the words “yes” and “no,” it would be a trivial challenge to determine exactly which word it would output to any given prompt. At four symbols, the challenge becomes increasingly difficult. At around 10 symbols, humans need tools to perform the math.
When you get to the scale at which large-language models function, with trillions of parameters, the potential permutations become mathematically intractable for humans, even with tools. The math is too big for us, but it’s still perfect.
Human creativity isn’t as predictable. Maybe the math underpinning our brain’s functions is perfect, and maybe it follows different rules. The important distinction is that our individual architectures are bespoke and no two humans think alike. Essentially, our thought processes are naturally encrypted.
According to the evidence, every human experiences stimuli differently. Scientists have narrowed down specific brain functions to certain regions, but synaptic responses to stimuli are nonuniform.
Thus, theoretically-speaking, if you were to copy information directly from one human brain to another, there’s a good chance the data would be unreadable.
When we experience and process art, we’re not copying data. We’re responding to stimuli. And that response may very well cause our brains to generate new data. If so, that data is unique; it cannot be mathematically predicted.
But all AI functions are intrinsically mathematically tractable — with sufficient binary processing power.
According to the Church-Turing thesis, this means that anything AI developers can do with LLMs — today and in the future — can also be accomplished using seashells (and a sandy beach of sufficient size).
Yet, no amount of seashells can predict the outputs of a given human brain. Our intellectual complexity belies the binary.
We can debate the philosophical nature of creation, ownership, and art until the cows come home. But, if we define human creativity in the context of our brain’s mathematical complexity and unpredictability, then truly human-level art can only be generated by a machine capable of general human-level reasoning.
Everything else is just more seashells.
The questions around AGI ethics are complex and evolving. Get thoughtful, well-researched commentary and the latest developments delivered straight to you by subscribing to AGI Ethics News.
A veteran AI journalist, Tristan was the creator and managing editor of The Next Web’s “Neural” imprint. He is one of the first and most prolific reporters to cover artificial intelligence as a full-time beat.