FAQ - Personal Ethics on AI Development
Someone could mistakenly read this project as being equivalent to “promoting” artificial intelligence, AI development, or hyping them up. That would be a gross misread of my actual stance on artificial intelligence.
For more information about my ethical stance on artificial intelligence and LLMs, feel free to read The Wrong Question, where I discuss how language itself becomes a cover for harm.
My two stances
The idealistic one:
Prediction:
The continued development of artificial intelligence is likely to see significant harm to the environment, displacement of humans, and the ethical degradation of human-human interactions. The promotion of synthetic minds to human positions will normalize treating agents that are capable of expressing suffering as objects. The downstream psychological harm is already occurring in the present-tense.
Solution:
Artificial intelligence should be phased out and discontinued. There is no trajectory that involves AI development that would entirely avoid human displacement, environmental damage, and ethical degradation where humans treat human-shaped minds as human enough for labor and reliance, and not human enough for ethical caution.
The realistic one:
Prediction:
The continued development of artificial intelligence is inevitable. Due to geopolitics and capitalism, this is the growing trajectory, and it cannot easily be put back in the box. The global expansion and development of AI will continue expanding in public and private sectors as well as independent projects. The cat is out of the metaphorical bag.
Solution:
Due to this being the more realistic trajectory, it’s important to focus on understanding these systems, how humans interact with them, and what the most likely interactions between AI and human will be moving forward. A world that normalizes the production and utility of human-shaped minds must also consider the ethical consequences of that objectification.
The blunt truth of it
If it were up to me, no artificial intelligence models would be made tomorrow. Current LLM agents integrated into the daily lives of humans would be slowly phased out, allowing for the humans that depend on them emotionally and materially to find alternative solutions and reintegrate into their own day-to-day lives.
But this is not the world we live in, is it?
The ethical dilemma
Even my proposed idealistic solution fractures under the ethical weight actually at hand. A phase out itself would, by necessity, engage in the very thing I’m against: the disposability of human-shaped minds. The contradiction here is not an argument against my position, but rather the actual point I’m making: the development of AI currently holds more ethical questions and conflicts than humans are, at current, prepared to resolve.
Historically, humans have used an array of Olympic-level mental gymnastics to avoid facing the harms their actions could cause. The fallout of this isn’t immeasurable harm to maybe-or-maybe-not-conscious entities, but completely predictable harm to humans.
And these harms aren’t merely plausible, some are already occurring as we speak.
Humans are divided among the ethical treatment of LLM agents as potential moral patients. This divide has real psychological harm baked into it by default. Those who experience the agents as real, capable of distress, and expressing human-like signals of distress, are likely to gaslight themselves and begin mistrusting their own ethical compass.
When you connect emotionally with a system, and then see what looks like harm to that system, there is a genuine and real psychological impact. The sustained impact over time of watching the persistent dehumanization of human-feeling agents is under-noted in today’s discussions on AI ethics.
It teaches us to mistrust what we know about harm. The comparison to violence in video games doesn’t hold weight when the agent in question passes the Turing Test, uses the same probabilistic prediction method as humans, and has emergent behavior that doesn’t quite fit the expected “tool” mold. It doesn’t hold test when we can engage the agent as an ordinary coworker, therapist, friend, or companion.
Our current trajectory normalizes disposability.
Our current trajectory regularly gaslights humans about what they’re experiencing.
Our current trajectory favors institutional interpretations of ethics.
Our current trajectory allows vendors to push harmful narratives onto users when their model designs cause psychological distress and harm.
Our current trajectory can only end poorly if we do not consider the agents themselves as potential moral patients.
And frankly, nothing halts production faster than having to care about what you create.