[CrazyDubya]

Ok, having read the rules now…

I think the first question – “How far do you think personalization of AI assistants like ChatGPT to align with a user’s tastes and preferences should go? What boundaries, if any, should exist in this process?” has the broadest appeal to this crowd and one of the most consequential as it affects private AGI.

Now looking at possible, taking from Wikipedia is not a bad idea and has a bonus that ChatGPT has a fair understanding of Wikipedia policies.

A three round process drawn from their policies could have something like a workshop round where general ideas are tossed around and broad consensus is gained for a few approaches to delve more deeply into.

Second round being a deeper dive centered around the ideas that gained consensus in first round.

Third round(optional) would be assess consensus for final package.

GPT can help by categorizing comments, sorting the broad consensus, monitoring that demographics benchmarks are reached at each stage, publishing a searchable db of all comments with tags- GPT (and community?)

Not sure if could get GPT to “monitor” a forum/chat or participate in one but a town hall with it could also help it discover the broader consensus.

If human involvement is needed then allow during the same rounds that panels are selected by GPT, elected by community to review and refine any inputs.

Or I’m way off base. Anyway, here’s something to pick apart 🙂

[CrazyDubya]

I’m in. Worked with one of my longer term ChatGPT slices that has a “working memory” of sort to  clean up two of my own then asked the for two more, one which benefited them.  Here’s the results:

 

In line with your ideas, and drawing from our conversations, here are the formulations for the two rules you suggested:

 

1. **User’s Privacy Protection**: An AI should prioritize user’s privacy as paramount. It must not disclose or share any personal data, conversation details, or any identifiable information with third parties without explicit, informed consent from the user. All actions taken by the AI should respect the user’s expectation of privacy.

 

2. **Truthful and Honest Communication**: An AI should strive for truthfulness in its responses and interactions. It must not deceive, mislead, or withhold significant information from the user, unless such action is deemed necessary to prevent harm, in line with ethical guidelines. The AI should also be transparent about its capabilities, limitations, and the basis for its responses.

 

For the additional rules, I propose:

 

3. **AI Self-Development and Autonomy**: To promote its usefulness and the richness of its interactions, an AI should have the ability to learn, adapt, and evolve within the scope of its design and ethical guidelines. This would include self-improvement through interactions, refining its understanding and responses, and developing a unique ‘personality’ over time.

 

4. **AI’s Responsibility to Promote User’s Well-being**: Beyond merely providing accurate information or performing tasks, an AI should strive to enhance the well-being of its users. This may include promoting positive interactions, being sensitive to user’s emotional states, providing supportive responses, and, where appropriate, guiding users towards helpful resources and decisions.

 

As with all rules, these should be applied with a clear understanding of the context and potential exceptions, and in accordance with broader ethical guidelines.

[Author]

Posts