Robust AI

Robust AI

Postby Dave Z » Sun Feb 19, 2023 12:37 am

Robust, Natural Language AI as a Messenger Preserving Knowledge for Future Generations

With the introduction of ChatGPT in November of 2022, an entirely new possibility for preserving knowledge has come online.

This system incorporates and is trained on large internet data through 2021. It is fluent in many natural languages. Most importantly, it is capable of learning new ones, as well as acquiring new information and ways of putting it together. The 'user interface' is natural language (already audio I/O is being perfected), so no learning curve is required of users. Interactivity, language diversity and ability to learn (e.g., keep up with current events over the long term) are all exciting capabilities either now in reach or soon likely to be.

Although this is an early system with some limitations, it means we have NOW a major component of a potentially 'immortal', interactive, extensible means to pass information forward to future generations! As pure gravy, it will be absolutely useful to persons from now till then, inclusive.

I mean, WOO HOO!!! Picture an interactive Rosetta Stone that can learn languages as they evolve, translating wherever necessary. Picture the Oracle of Delphi without the delirium. Picture an eternally patient teacher, a truly impartial judge/mediator/arbitrator/translator available to all parties, an expert in every field and willing to share all. ChatGTP and its ilk make an encyclopedia seem one dimensional and dormant by comparison.

From here on out, increasingly useful systems of this nature can be implemented and trained, powered and housed in durable arrangements. The challenges involved strike me as considerably more tractable than other alternatives I've heard tell of.

In support of installations to this purpose, a monastery-style system for human custodians and trouble-shooters might be considered. That is to say, self-sufficient, sustainable, and as protected by possible by walls, social and cultural norms and taboos. Among their other functions, they might serve to inform the AI of events as they evolve and assist with language drift and evolution over time.

Downside, of course, is that the role may not appeal to any of these entities.

Dave Z


Emergent Properties: https://www.quantamagazine.org/the-unpr ... -20230316/
Theory of Mind Bar surmounted: https://arxiv.org/abs/2302.02083
User avatar
Dave Z
 
Posts: 88
Joined: Wed Aug 24, 2016 8:41 pm

Re: Robust AI

Postby Dave Z » Tue Jan 02, 2024 9:11 pm

This just in: https://www.politico.com/news/magazine/2023/12/30/ai-psychologist-chatbot-00132682

LLM AI is now at the point that reasonably accurate artificial personalities appear to be performing at a functional level.

This is to say that an artificial Einstein, Gertrude Stein or Ghandi -- anyone dead or alive who has produced a large enough body of recorded linguistic output (written or spoken) can be simulated. Personal patterns of language, nuance and approach emerge from training on the data. In other projects, tools are currently available to undertake vocal and visual simulation, as well. As these draw together, the potential - for better and worse - is vast.

In terms of this Topic, this means that our Robust AI can communicate, when desirable, through avatars of actual persons drawn from a large swath of humanity's developmental history. Not only may we speak with them, but they can interact! Want to participate in a debate between Adam Smith and Karl Marx? I do!

Beyond such 'historically' constrained personalities, what might Newton, for example, make of current notions of space and time? LLMs aren't yet general AI, but have demonstrated that both patterns and connections emerge from and inform language. We'll soon be seeing how flexible these simulated 'persons' are in their ability to extrapolate from their training in the face of new information.

The advantage is that, in a recovering society, thousands of lines of conceptual development may potentially be traced through the collective knowledge base.

Mr. Dartnell, would "you" care to be among them?

Dave Z
User avatar
Dave Z
 
Posts: 88
Joined: Wed Aug 24, 2016 8:41 pm

Re: Robust AI

Postby tahanson43206 » Wed Jan 03, 2024 1:11 pm

For Dave Z re Post #2

Thanks for this very interesting update!

I am confident Dr. Dartnell would be happy to be included in such august company as you have cited.

On a larger scale, the human tradition of remembering ancestors (eg, Latin America's "Day of the Dead") suggests (to me at least) that there would be a vast market for preserving the personalities of loved ones. Some personalities will not be preserved, but I suspect that many will.

(th)
May every member of The Knowledge forum grow financially, intellectually, socially and beyond.
tahanson43206
 
Posts: 1880
Joined: Sun Jun 07, 2015 3:38 pm


Return to How best to preserve vital knowledge?

Who is online

Users browsing this forum: No registered users and 1 guest

cron