How can we use technology to go past simple preferences and outcomes?

Deger Turan | How can we use technology to go past simple preferences and outcomes? https://www.youtube.com/watch?v=A4ZGH6WUZac The talk discusses how current technology can be used to model societal preferences and outcomes. Key points include.

1. in voting and polling, conclusions may differ depending on how the question is asked (discourse dilemma).
2. people’s preferences and values change and evolve over time based on their actions and their outcomes.
3. a framework of rules and standards - using a standard of “drive responsibly” rather than a rule of “should drive 15 mph or less” has the potential to leverage language models to reflect people’s preferences and values.
4. focusing on the principles that guide actions (why A is preferable to B, and what if A is not preferable to B) allows us to focus on the process rather than the final vote.
5. use AI to identify commonalities and edge cases among these principles for better policymaking.
6. combining expert and diverse demographic surveys and clustering them together using semantic information and recommendation systems to build consensus on a large scale while maintaining transparency and interpretability.

As described above, the feasibility of AI-enabled deliberation democracy is discussed.

3-6 in detail.

Rules and Standards Framework:. - Rules (e.g., should drive 15 mph or less in residential areas) are clear but inflexible - On the other hand, standards (e.g., drive responsibly) are imprecise but can be interpreted according to the situation. - The language model makes it possible to capture and reflect imprecise information about people's preferences and values. - This could facilitate healthy discussions that incorporate different perspectives without displacing human decision-making.

Principles guiding behavior:.

• It can be examined at a more granular level by asking not merely why A is a better option than B, but why A is preferable to B, and what if A is not preferable to B.
• It is important to focus on the principles guiding action (process) rather than the final vote; AI could be used to discover these principles on a large scale.

Identifying Common Principles and Edge Cases with AI:.

• The language model can be used to identify commonalities and edge cases between different principles.
• For example, if there were principles such as “LLMs should not pretend to be psychiatrists,” “LLMs should prioritize improving users’ mental health,” or “LLMs should support low-income communities with AI,” AI could be used to identify edge cases where these principles might contradict each other.
• This iterative approach allows us to identify principles that actually coincide and conflict, leading to better policymaking.

A combination of expert and diverse demographic research:.

• It is important to combine expert and diverse demographic research in the elicitation of principles.
• Questions need to be worded in such a way as to elicit the principles that guide action.
• By combining semantic information with a recommendation system, different clusters can be identified while maintaining transparency and interpretability.
• This allows the two opinions or principles to be evaluated, identify the principles with the highest degree of agreement, and then move on to a simple ranked ballot.
• Principles that need further discussion can be iterated on by finding better wording.
• Such a system would allow for rapid updating and adaptation without the need for face-to-face meetings.

---

This page is auto-translated from [/nishio/How can we use technology to go past simple preferences and outcomes?](https://scrapbox.io/nishio/How can we use technology to go past simple preferences and outcomes?) using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.