Note
In this text, the term "Generative AI (GenAI)" refers to contemporary transformer-based models, their harnesses, and technologies with similar characteristics that are used to generate content, such as text, images, audio, and video.
i do not like how GenAI is being built, advertised, and used. And i will restrain myself, to the best of my ability, from participating in the current GenAI economy.
i am no expert in machine learning. This text is my opinion, formed based on my very limited understanding of GenAI, and my very emotions.
GenAI can do some fascinating things, let's give it that. Among its best application is pattern recognition, processing natural language and structured text: context-aware autocomplete, translation, summarisation, raster image upscaling or restoration, to name a few.
i am not particularly against the technology and its mathematics. i despise, however, the way GenAI is being imposed upon us by tech companies, rushed out of the door with little legal and ethical framework; how its complexity is marketed away, its biases objectified, and responsibility blamed on end users:
"AI is just a tool - it matters how you use it."
— a sentiment heard too often
GenAI is a tool, yes - as much as a gun is a tool. i refuse to use a gun, and i can too with GenAI. GenAI can be convenient, but so would a gun be, if eliminating someone i don't like were legal and ethical. If a tool is built by someone that promotes inequity and is complicit in genocides, i feel obliged to reject it, and will do so if i am in a position that affords such option. GenAI is no difference. No tool exists without political, economic, environmental, and social context. What person am i if i were to shoot someone?
We use tools, and they too shape us.
When someone says GenAI is "just a tool" to me, they are shutting down the conversation we are having, and basically saying that they are no longer interested in anybody else's perspective on the matter.
GenAI is more than automation: it is delegation. i don't just use GenAI to do something, i ask GenAI to do it for me.
Yes, with GenAI, i may not have to learn how to do ABC anymore, yet i would forget that, through learning, i don't just attain the how-to-do, but also refine my how-to-learn. Learn to learn, that is.
Yes, with GenAI, i don't have to ask someone else to do XYZ, yet i'm reminded that it would be another human connection lost, one step closer to isolating myself from the very social context that made up who i am.
Alienated at best, is GenAI.
Beyond being theft, a heavy environmental tax, a mode of capital and power concentration, etc. - GenAI is misused, in the name of automation and productivity, to take away our hands and minds from labor - the hard and boring work but at the heart of which lies craftmanship and the human creative spirit.
Whether it is a text predictor, a complex probabilistic model, or a new form of intelligence, commercial GenAI - intended or not - feels at times like they are trying their best to be as humanly as possible. It can speak, via text, image, video, audio. But, if it speaks only a few sentences repeatedly and predictably, that would be robotic.
Instead, give it some randomness, some personality, some imperfection, some humanity.
And ask it to do human things.
Is it not strange, that i would want something done as humanly as possible, but without the human? Are there no longer programmers, designers, painters, musicians, writers, ...?
The machine can speak. Its reward function tells it to.
We can speak too, as human.
When a history teacher in a far remote fishing village is pressured into using GenAI to generate their lessons, what knowledge is being produced and propagated? How much knowledge is lost? Can the local lores and stories about its land and people persevere through time? Or will they be eclipsed by the more dominant narratives that GenAI is trained on?
Loud voices own knowledge production.
The human collective knowledge is vast and diverse, rich in history and perspectives, only a fraction of which is documented. What small parts of this fraction are fed into GenAI during its training? Who decides so? i can only speculate, that my mother's recipes are not among them. Neither are the proverbs or folk songs my grandmother once sang, nor the traditions and wisdom passed down only by spoken words.
"Oh, just train your own model from your own dataset?"
How would i do that? By spending decades of unpaid effort studying and documenting knowledge because the only thing that fed me is curiosity? Or by buying up warehouses of GPUs and hardware because i had generational wealth at my disposal? Or, should i ask some GenAI to do those for me?
The impact of GenAI is far and deep. It is already happening today. GenAI is being handed to people without any disclaimer or instruction manual, and overused where it should not be used.
"Oh, technology always develops so fast like that. It's hard to avoid some left-behinds."
Does it have to be? If technology is truly fast, it must surely be driven by some drunk moron.
Why is it that some farmers where i come from still don't have access to combine harvesters? They are incredible pieces of technologies that help ensure food security for an entire population. Sure those machines take much to manufacture and are expensive to buy, but with this so call fast technology, couldn't we make it more affordable already?
Why is it that websites and apps still crash every time i need them the most? Why aren't they caught up to accessibility standards? Surely reliability and inclusiveness are good ideas, no?
The Web Accessibility Content Guidelines (WCAG) was first published in 1999. Combine harvesters have been around decades before that. How fast have they been?
At the same time, why is GenAI - a highly-praised achievement of humanity over a century of mathematics and computer science - so suddenly affordable that everyone can have it in their pockets today? Is it not taking trillions of priceless US dollars to build data centers and to train those models?
Am i comparing oranges and apples? Or is it not really about technology? It certainly feels to me more about being disruptive, being first in the race, being comfortable in the gray area of legality and ethics.
Perhaps, GenAI is rushed because these tech companies deem it to be. A new market, a hype-based lucratively stupid one.
"On don't worry about it, we will catch up."
Oh i know: we always do. The same way we manage to imprison war criminals only decades after their crimes were committed. The same way we still picking up pieces of landmines and digging up unexploded bombs today.
We will catch up. But at what cost? What communities are left estranged? What species are eradicated?
By the time we catch up to GenAI, what else will be there for us to catch up to?
In this reduction of knowledge by the expansionists. i have yet to see an end. Do you?
- Maybe if i don't know how to do something, don't have the time to learn the craft, and cannot find another person to help me, then i should not be doing it in the first place.
- Maybe let's return GenAI to labs until there are clear legal and ethical frameworks? A defective product recall is an apt analogy?
- Maybe let's try our best to figure out a sustainable way to train these models for specific use cases, instead of making massive know-all models? Some minimal models that can run locally on your and my machines without the centralised cloud?
- There are enough jobs for everyone. Maybe let's chill out on this hype train and AI arms race? Let's live and eat and love instead of trying to trade one's livelihood for another's greed in this never-ending zero-sum game?