Seeking feedback on idea of Effective Oracle - truth and correct decision making maximizer.

Hi everyone! I need to get feedback on a project i am working on - Effective Oracle.

Summary: I'm proposing the creation of an "Effective Oracle" - a collection of four interconnected systems that optimize for uncovering critical truths, spreading these truths widely, enhancing logical reasoning, and making decisions based on these truths. Leveraging AI and collaboration, this self-improving system aims to address the imminent challenges posed by superintelligent machines. I am initiating this vision by developing a decision-making software tool. Feedback on this idea is sought for further refinement and implementation.

Its just a google doc, you can read it in couple minutes. I would greatly appreciate if you give your feedback.
https://docs.google.com/document/d/1YWpBiT1fmxQT0GS7NMUjO7APP-REBD2J9Q7mEfe9slQ/edit?usp=sharing

Google Docs

Gathering feedback on Effective Oracle

Gathering feedback on Effective Oracle Author: Percy Otebay zhumaotebay@gmail.com Twitter: @Radlib4, Reddit: Radlib123, Discord: percybay Contents: Summary Problem and a need for solution Solution, a system that maximizes truth and better decision making. Many people with different interests ca...

dual-salmon•10/22/23, 7:07 AM

@krantzbhanzinpantz This seems relevant to what you are working on (similar goal, completely different mechanism).

dual-salmon•10/22/23, 7:14 AM

The idea of "Oracle AI" has been floated around a bunch over the years. Here are some of the key challenges:

How do we make an AI that values truth? Basically, this requires solving a substantial subset of alignment itself. It's also something I would assume groups like OpenAI are already working on, given that "hallucination" is bad for sales.
An Oracle, even if built, does not eliminate the incentives to build agentic AI--and even could be used to help accelerate its appearance.
The line between a "safe" Oracle and a "dangerous" Agent can be blurry. If a truth-seeking AI wants to maximize its ability to answer questions, for example, it turns the world into computronium if it is asked a question that requires infinite compute.

Not saying these are knock-down objections, but they are issues this proposal should recognize and address.

inadequate-blush•10/25/23, 4:26 PM

I agree. I spoke to @Percy a little about it. Here was my feedback:

Percy, I read your paper. I like your ambition. I completely agree with everything. That is what we need to build. Keep that spirit.

Now, for the hard part. How do you actually build that thing?

This is a good description of the program I began trying to build back in 2010. I quit my career as a casino director and went back to school to study philosophy for 8 years and learned math, physics and computer science independently. I rigorously studied artificial intelligence and collective intelligence for more than a decade.

Good news. Now I see an actual way to build a system that does this.
Like, step by step.

Absolutely nobody is doing anything like it. It's a game changer.

We are going to need someone better connected to others in the field with access to more resources.

Got any ideas on how to get anyone's attention?

inadequate-blush•10/25/23, 4:31 PM

I think the key towards guiding AI to seek truth lies in how we as a society decide to record truth. I think the hard part is finding a decentralized way to actually create a 'general ledger' of truth.
I'd be very interested to speak with you again about more of how to do this explicitly. Not sure how much you remember from our conversation or how well I expressed the strategy, but since Doug's passing, my moral has dropped significantly. I very much enjoyed our conversation though and would love to have another if you could find the time.

Iill-bronze Hi everyone! I need to get feedback on a project i am working on - Effective Ora...

faint-white•10/31/23, 11:24 PM

I read the document in full, it's all useless. Basically a crackpot proposal.

Basically, there is little to no mention of how each of the four components will work. This is not a plan, it's a wishful dream.

(Also, it has very little relationship with Oracle AIs, it's just a misnomer.)

Iinadequate-blush I think the key towards guiding AI to seek truth lies in how we as a society dec...

ill-bronzeOP•11/1/23, 6:05 AM

Of course!

Seeking feedback on idea of Effective Oracle - truth and correct decision making maximizer.

Similar Threads

Similar Threads

Similar Threads