Literature Review Direction
I will answer the following questions for each of the papers:
- How did they define deliberation?
- How did they use that definition to frame the deliberative practice they are researching?
- How many are involved in the deliberation? Is it human-to-AI? AI-to-AI? or AIs-to-human? Humans-to-AI? AIs-to-AIs? human-to-AIs?
- How do they employ the technology into the deliberation, and how does that relate to their definition of deliberation?
- i.e., does it advance the definition’s ideals?
- What were their findings?
- What were their conclusions, and work that remains to be done?
Papers
Machine-assisted deliberation.base
My thoughts
I like the critique from hernandezAutomatingDeliberationIdea2025 that approaches like tesslerAICanHelp2024 and bakkerFinetuningLanguageModels2022 assume static preferences, and that when you do, you begin to see people engaging in deliberation as “political patients” than “political agents”. So I like the idea of giving humans more agency within the deliberation, and assigning LLMs to the task of training them to form better arguments, as well as posing questions. This is also very much in line with lazarCanLLMsAdvance2025.
Some of these are narrow problem spaces. papachristouLeveragingLargeLanguage2025 is looking at decision making, and tanRobotsMiddleEvaluating2024 is looking at disputes. huaAssistiveLargeLanguage2024a and sheaACELLMbasedNegotiation2024 are looking at
People try to adopt social science theories and apply them to LLMs. argyleAIChatAssistants2023 with restatement, validation, and politeness to improve perception of feeling understood, kambhatlaPromotingConstructiveDeliberation2024 with six strategies for signaling receptiveness.
Review papers are super useful. zhangDesigningAIInfusedInteractive2025 contains two tables, first being categorization of papers into groups like AI motivational support, ability support, and coordination, and second being how they evaluate desired outcomes like willingness to contribute, and collaboration.
lamPolicyMapsTools2025? yangTestersBiasesGuiding2023? wallachPositionEvaluatingGenerative2025?
Important considerations:
- Bidirectional feedback between human and AI
- Designing for perspective shifts, or static preferences