- This topic has 2 replies, 3 voices, and was last updated 1 month, 3 weeks ago by
.
-
Posts
-
If we imagine an AI system trained only on large amounts of data and logical patterns, but without any direct human experience, emotions, or lived social context, could it ever arrive at something like human morality or ethical understanding, or is human experience fundamentally required for that? More generally, could morality itself be understood as a form of information that is accumulated and transmitted across generations rather than something uniquely tied to human consciousness and experience?
Honestly, I don’t think AI has “hit a ceiling”, it’s more that we’re projecting human expectations onto something that works very differently. AI systems are already capable of modelling complex patterns in language, behaviour, and decision-making, but that doesn’t necessarily translate into lived understanding of emotions or morality. Those things are grounded in experience, consequence, and social embedding, not just data processing. So even if AI becomes extremely advanced at simulating moral reasoning, it’s still an open question whether that counts as actual understanding or just highly accurate imitation.
From an information theory perspective, morality can be seen less as something purely philosophical and more as a compressed structure of information shaped by long-term cultural and evolutionary pressures. In that sense, moral systems act like a kind of inherited “memory” of what stabilises human societies over time. The key question is whether an AI can inherit that structure without inheriting the conditions that produced it. This is explored in more detail in an information-theoretic framing of AI ethics here: https://medium.com/@Circa84/can-al-inherit-moral-memory-without-the-conditions-that-formed-it-84d2c66cb5c0 . It raises the deeper issue of whether moral understanding can be separated from its origin, or whether the origin is actually part of what gives it meaning.
- You must be logged in to reply to this topic.