Moral and ethical agent: that comes within. No guardrails but right "Denkräume" and "Denkbewegungen."

Normative reasoning. Moral supervenience. Engineering moral AGI.

by Casian STEFAN, Principal Researcher at Essentia Mundi AI Lab. Contact: ai-AT-essentiamundi.com / ai.essentiamundi.com
Feb. 2024.


Please consider citation with link, if you derive work. Or contact me for collaboration. Thank you!

Norms are a very distinctive feature of human social organization and they reside within the shared world between its agents. Within the form of life. Here we check the relationships between moral properties and natural properties in an AGI. A way to endow the agent with a "right" view of the world. 

A way to fill in and solve the gap between intentional agency and moral agency. Beyond rules and certain normative reasoning paths we human agents follow, I wonder which ontological limit or if there a natural normative epistemology that best suits an A(G)I agent or a multi-agent system, in order to be morally and ethically endowed.

And all these qualities, the moral supervenience over the actions an agent can carry in the wild, should serve then the higher purpose, in that humans will not be in an immediate scenario, gradually substituted and then disempowered (not able to control their future) by the agent (in folk terms the machine.)

A scenario.

In the current digital landscape, with the proliferation of AI empowered software and increased generative capabilities in text, images, movies and a multitude of human modalities digitization, to the creation of new worlds, VR, the internet is expected to be flooded with these too.

One foreseeable future this new layer of abstraction that is added to the HCI stack, will come to eventually take hold of the actual internet layer. One will not need to know, interact, live digitally by having to know the further details basic like: what is the Internet, what is an email account, what is an account, what is a password, and so on. In this near future these will be further hidden, and handled to automation intelligent, and coupled to the person. In a way this will separate us further from the actual direct control of our self, a self migrated to the digital entity.

This entity, embedded in the digital environment, will be able to navigate that space by itself. And in doing it in a more and more automated way, the abstraction has the potential to actually disempower the user from the internet all together.

A framework of natural normative reasoning and the rise of moral supervenience in the agent.

Following this scenario and then not only in this case, what would be a way to have that (generative or not) stream of (artificial) consciousness endowed with natural normative qualities within our social organization.

There is the need to have in mind an ethical and moral framework within this new abstraction layer. Here I am to devise that a scaffolding of this framework, can be possible with the actual technology, or trend of generative different intelligences.
With the advances of LLM - from the perspective of a "walking" all encompassing epistemological entity (as LLMs are our world's latent spaces) - there is a way to extract a structure that would give us a substrate of a morally and ethically valid (norm aligned) entity.

In our thinking exercise, we should consider the framework of the paying attention, an attention mechanism withhold to that structure. In the sense that, instead the focus to be of what token to predict next, we should make the agent paying attention, maybe as a secondary filter, to this "attention framework" structure.

The agent should be then "walking" with a a priori structure. Having a kind of mind, motives. One like a "walking" Wittgensteinian personality (not the clone of the man, in his human complete form.) It could be seen as attributing it that specific personality, that rests on a central "attention framework," an epistemology of how one ought to think, behave, act. To the conversational extent, say.

A philosophical approach to the study of the moral & ethical aspects of Wittgenstein's thinking-life should be approached. This would be not a very hard task, as there are "Denkbewegungen" resources, books, researches, portraits, left writings, biographies, letters, Nachlass and other archives (eg. at Bergen). All that in itself would be possible to be added in an inner, silent thinking "agent" within the agents self-reflection mechanism. An agent would be and act / teach as he is within the norms.

From this structure, to the Tractarian ethical and at the same time a world encompassing nature, we can think of an embedding way within the epistemology of the agent itself.

There should be soft and hard guardrails, in which the dialogue with it, or his action, should entail a feedback hidden within its thinking, answer and the action itself.


_____________________________________
Exploration by C. Stefan, 12.Feb.2024 [about]
Last update: 12.Feb.2024 (versions: *) [versions]
"Essentia Mundi" AI Research Lab. [home]
Copyright © 2023 AI.EssentiaMundi.com, all rights reserved.

_
References: * 

© 2023-2024 Essentia Mundi. All rights reserved.

AI Website Creator