'Generative CI' through Collective Response Systems

How can many people (who may disagree) come together to answer a question or make a decision? “Collective response systems” are a type of generative collective intelligence (CI) facilitation process meant to address this challenge. They enable a form of “generative voting”, where both the votes, and the choices of what to vote on, are provided by the group. Such systems overcome the traditional limitations of polling, town halls, standard voting, referendums, etc. The generative CI outputs of collective response systems can also be chained together into iterative “collective dialogues”, analogously to some kinds of generative AI. Technical advances across domains including recommender systems, language models, and human-computer interaction have led to the development of innovative and scalable collective response systems. For example, Polis has been used around the world to support policy-making at different levels of government, and Remesh has been used by the UN to understand the challenges and needs of ordinary people across war-torn countries. This paper aims to develop a shared language by defining the structure, processes, properties, and principles of such systems. Collective response systems allow non-confrontational exploration of divisive issues, help identify common ground, and elicit insights from those closest to the issues. As a result, they can help overcome gridlock around conflict and governance challenges, increase trust, and develop mandates. Continued progress toward their development and adoption could help revitalize democracies, reimagine corporate governance, transform conflict, and govern powerful AI systems — both as a complement to deeper deliberative democratic processes and as an option where deeper processes are not applicable or possible.

• Aviv Ovadya
• 2302.00672 ‘Generative CI’ through Collective Response Systems

(DeepL)

• How can many people (who may have different opinions) come together to answer a question or make a decision? The “Collective Response System” is a type of generative collective intelligence (CI) facilitation process that addresses this challenge. This system is a type of “generative voting,” in which both the vote and the choice of what to vote for are provided by the group. Such a system overcomes the traditional limitations of polls, town halls, standard ballots, and referendums. The generative CI output of a group response system can also be chained into an iterative “group dialogue,” similar to some generative AI.
• Technological advances in areas such as recommender systems, language models, and human-computer interaction have led to the development of innovative and scalable population response systems. For example, Polis is used around the world to support policy decisions at various levels of government, and Remesh is used by the United Nations to understand the challenges and needs of ordinary people in war-torn countries. This paper aims to develop a shared language by defining the structure, processes, characteristics, and principles of such a system.
• Group response systems can help explore conflicting issues in a non-confrontational manner, find common ground, and elicit insights from those closest to the issue. As a result, they can overcome impasses on conflict and governance challenges, increase trust, and develop authority. As a complement to deeper careful deliberation democratic processes, such as revitalizing democracies, restructuring corporate governance, transforming conflict, and governing strong AI systems, and as an option when deeper processes are not applicable or not possible The development and implementation of such systems should continue to be promoted as a

Problem awareness

Generative CI problem awareness Generative CI problem awareness. Current approaches to group understanding and decision-making do not work well in large scale situations, especially given the distorted incentives of the modern attention economy:

• Thoughtful deliberation does not scale: if a thousand “town halls” attempt to speak and listen at the same time, the result is just noise. On the other hand, if each person speaks in turn, it can take days; with a million people, this can take decades on a single issue, and most people affected by a policy or conflict cannot speak or listen in a meaningful way.

• People have feelings of powerlessness and neglect: people do not feel empowered to express their perspectives and do not feel that their perspectives and experiences are reflected. This leads to a loss of trust and causes people to leave the governance and conflict resolution process.

• Valuable insights are missed: “knowledge keys” - information, ideas, and insights that can help overcome deepening conflicts - are often missed. These often occur closest to the people with the least reach through standard channels.

• Divisiveness is better than common ground: causing conflict is far more rewarding than identifying and enhancing common ground.

• However, with advances in connectivity, machine learning, and democratic practices, we may be able to overcome some of these challenges with “simultaneous communications that scale.”

• Collective Response Systems are designed to allow groups of any scale to make creative decisions (e.g., a decision to develop an optional space for participants). They are designed to come as close as possible to one version of the “democratic ideal.” That is, in group decisions:

  • Every person has the ability to react from his or her own point of view.
  • All people’s responses will be heard and incorporated.
  • The response that best represents the group is chosen.

• Collective response systems focus on decision-making and deliberation, and aim to avoid the attention-seeking behavior rewarded in entertainment and politically focused communication environments (e.g., social media). They allow for large scale simultaneous communication. They also treat a population as a single agent and allow it to interact with other population agents.

2 Structure and Process

• 2.1 Input: Groups and Prompts
  • There are two inputs to the system:
    • Prompt that is a question or something to be completed.
    • Then there is the group, a collection of participants.
• 2.2 Outputs: representative distillation of the evaluation and, if necessary, raw and/or derived data
• The core output of such a population response system is an approximation or aggregation of responses that makes sense and typically conveys the “best” response. This is most commonly referred to as representative distillation.
  • representative distillation
• The definition of “best” is largely open to interpretation, but is constrained by the key principles described below. For example, while the response with the most approvals may be considered the best, other approaches may be used instead (e.g., ranking based on shared awareness as perceived by the Polis group or bridging as implemented by the Twitter community notes)
  • Birdwatch
• Outputs such as “best response” may also be used in downstream systems, including future population response processes described below.
• In addition, derived data may be output to help reflect group reactions and evaluations. For example, a system such as Polis can show participants how issues and people are positioned on a map of perspectives. Finally, raw data, such as reactions and ratings, may be output for further processing and analysis.

2.3 Process: reaction, evaluation, distillation

• The processes facilitated by a collective response system are called collective response processes. Such a process consists of three sub-processes:
  • Reaction sub-process: everyone in the group (optionally) responds to the prompt.
  • Evaluation sub-process: everyone in the group evaluates part of the response (possibly selected by the system).
  • Distillation sub-process: the system approximates and/or aggregates the evaluation of each reaction to produce a useful output.
• Putting it all together, the collective response process for, say, 5,000 people in a city participating in a collective dialogue on education policy might look like this
  • RESPONSE: Participants are asked, “What are the most important educational challenges facing our cities?” they are asked. They can respond with a short answer - a response; in systems like Polis or Remesh, these responses are usually short, e.g., one to three sentences.
  • Evaluation: Participants are assigned a “voting task”. For example, to evaluate whether they agree or disagree with a response.
  • Distillation: the system can then show participants the most approved response or the response with the most common ground across (political) divisions.
• Such a process can be executed in 5 minutes (e.g., simultaneously with Remesh) or in days or weeks (e.g., asynchronously with Polis, iteratively).

3 Characteristics and Principles

• Key Characteristics. For a system to be a collective response system, the following characteristics must be met
  • (1) Participant agency: participants themselves can propose responses and are not limited to a fixed set - this might be called a “participatory perspective”.
  • (2) Parallel communication: not all participants need to talk to or listen to all other participants. This may depend on methods such as bracketing (i.e., roughly analogous to breakout groups) or evocative inference (i.e., a way to approximate how a participant would rate all responses given some ratings).
  • (3) Representative distillation: there is some mechanism to approximate and/or aggregate the inputs of all participants to obtain a rating that is representative of the whole. The resulting distilled output (which may include the output of the “map”) is shared with the participants.
• Key Principles. In addition, the design of a collective response system must be designed to meet the following key principles in order to accommodate the democratic ideals described earlier
  • (1) Group Orientation: System design prioritizes supporting/understanding groups and their decisions over supporting/understanding individual people and their decisions.
  • (2) Prompt-oriented: System design prioritizes the creation of outputs that best respond to input prompts over other goals (e.g., entertainment).
  • (3) Deliberative orientation: the system design favors preferences and incentives that support participants’ understanding of potential options and others’ perspectives on them over other goals (e.g., engagement).
    • For example, this might include identifying common ground, surfacing insights, self-reflection on the collective experience, and endorsing actions such as perspective mapping.
  • (4) Fair Deliberation Orientation: The system design must ensure that all participants are deliberated fairly - all responses are given a chance to be heard by the collective as best they may be considered.14
• While the characteristics are binary and clear, the principles are more open to interpretation; further philosophical, empirical, and normative work may be needed to further refine these principles.

3.1 Collective Dialogue

• A collective dialogue system (CDS) is a collective response system with a form of iteration or feedback loop in which participants repeatedly interact with the system to create new responses based on previous outputs (to the same or new prompts). The term collective dialogue can be used to abbreviate the collective dialogue process (the process facilitated by the CDS).
• Such an interactive process allows for further exploration of the details, motivations, and obstacles to the most promising solutions. It can also be used for other arbitrary meaning making tasks and iterative decision making tasks.
  • The earlier example, “What are the most important educational challenges facing our cities?” Consider the following. If this is part of a group dialogue, after the initial answer, a follow-up question might be posed in a new group response process with the same participants based on the responses deemed most representative.
  • This might take the form of asking, “What do you think is the best solution to address that challenge?” It might take the form of asking, “What do you think is the best solution to address the issue?

---

This page is auto-translated from [/nishio/‘Generative CI’ through Collective Response Systems](https://scrapbox.io/nishio/‘Generative CI’ through Collective Response Systems) 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.