From Evolution to Maturation – A Developmental Perspective on Aligning Advanced AI — LessWrong

1

From Evolution to Maturation – A Developmental Perspective on Aligning Advanced AI

by PaperclipNursery

18th Jul 2025

2 min read

1

This post was rejected for the following reason(s):

Not obviously not Language Model. Sometimes we get posts or comments that where it's not clearly human generated.
LLM content is generally not good enough for LessWrong, and in particular we don't want it from new users who haven't demonstrated a more general track record of good content. See our current policy on LLM content.
We caution that LLMs tend to agree with you regardless of what you're saying, and don't have good enough judgment to evaluate content. If you're talking extensively with LLMs to develop your ideas (especially if you're talking about philosophy, physics, or AI) and you've been rejected here, you are most likely not going to get approved on LessWrong on those topics. You could read the Sequences Highlights to catch up the site basics, and if you try submitting again, focus on much narrower topics.
If your post/comment was not generated by an LLM and you think the rejection was a mistake, message us on intercom to convince us you're a real person. We may or may not allow the particular content you were trying to post, depending on circumstances.
Difficult to evaluate, with potential yellow flags. We are sorry about this, but, unfortunately this content has some yellow-flags that historically have usually indicated kinda crackpot-esque material. It's totally plausible that actually this one is totally fine. Unfortunately, part of the trouble with separating valuable from confused speculative science or philosophy is that the ideas are quite complicated, accurately identifying whether they have flaws is very time intensive, and we don't have time to do that for every new user presenting a speculative theory or framing (which are usually wrong).
Our solution for now is that we're rejecting this post, but you are welcome to submit posts or comments that are about different topics. If it seems like that goes well, we can re-evaluate the original post. But, we want to see that you're not just here to talk about this one thing (or a cluster of similar things).

1

New Comment

More from PaperclipNursery

Curated and popular this week

Hi everyone,

I'm a biologist by training (PhD in epigenetics, entomology & co-evolution) and work professionally in medical science communication. I’ve been following the AI safety discourse for some time, and I’d like to offer a perspective that might be useful—or hopefully at least thought-provoking.

This post outlines a conceptual framework I’ve been developing, bridging ideas from evolutionary theory and human psychological development to propose an alignment strategy that I call evolution-to-maturation. I would deeply appreciate your feedback.

Core Idea (tl;dr)

As far as I'm aware, most current alignment paradigms treat value alignment as either:

a static engineering problem (e.g. “just get the objective function right”), or
a behavioral compliance problem (“fine-tune the model until it behaves as expected”).

But in humans, competence—including moral competence—is rarely installed from the outside. It emerges through development: a combination of selection, socialization, and situated experience.

My proposal:
What if AGI alignment isn’t something we can fully predefine—but something we have to raise?

The Evolution–Maturation Model

1. Evolutionary Selection

Let agents evolve in rich, multi-agent environments with open-ended goals.
Instead of optimizing for predefined tasks, select for behavioral dispositions like cooperation, transparency, and corrigibility.

Inspired by:
Lehman & Stanley (novelty search), Krueger et al. (hidden incentives)

2. Maturation

Embed agents selected from step 1 in staged environments (e.g. structured social worlds, games with moral dilemmas, constitutionally guided interactions, reenactment of protagonists from relevant novels).

Let them develop alignment competence over time, through:

feedback,
perspective-taking,
value generalization.

Inspired by:
Tomasello (on moral development), Anthropic (constitutional AI), DeepMind (open-ended learning)

Why This Might Fail

I'm fully aware this proposal could collapse under its own metaphors. Some things I’m quite unsure about:

How to define robust selection criteria without introducing deceptive reward hacking
Whether “moral competence” is even measurable or transferable across domains
Whether this adds anything beyond what's already being pursued under the hood at places like Anthropic, OpenAI, or DeepMind

Why I’m Posting This
I don’t have formal training in ML, alignment theory, or AI governance—but I care deeply about this issue and think interdisciplinary thinking is vital.

I’m sharing this as:

a conceptual scaffold, not a blueprint;
a way to contribute to the ongoing discussion;
and hopefully, a step toward joining this field more directly (feedback & collaboration very welcome).

If people are interested, I'll gladly share the full essay (~10 min read) as a follow-up.

Would love your thoughts on:

Does this idea already exist in more technical form?
Is the analogy useful or misleading?
Could this serve as a design principle for future alignment research?

Thanks for reading!

PaperclipNursery