← Explore Capabilities
Dimension: Learning Speed Read about assessment rating system
Simulate prototyping
Ability to run trials that are good enough to learn from, and fast enough to enable rapid testing of new methods and process comparisons.
MINIMAL
EXTREME
Maturity
Opportunity
Importance
Neglectedness
Transnational
How this is performed now...
Process organizers
Process organizers typically rely on real-world experimentation to learn from. Existing examples of simulated trials are primitive agent environments and do not track the complexity of end-to-end processes.
Related Product Gaps
There are no related product gaps at the moment.
Related Goals and Research Questions
Goal: Designers who understand the validity boundaries and appropriate use cases for deliberative process simulation
What simulation fidelity level (agent realism, dialogue authenticity, decision distributions) accurately predicts outcomes for specific deliberative formats under real-world constraints, and where does increased fidelity stop improving predictive value?
Urgent
Simulation & Modeling
Related Capability: Simulate prototyping
For what uses, in what contexts and with what level of faithfulness is it helpful or appropriate to use simulations, and what are the philosophical, moral, and political implications?
Urgent
Simulation & Modeling Ethics & Philosophy
Related Capability: Simulate prototyping
Goal: Deliberative processes that can be tested and refined before implementation with real participants
What design variables in deliberative formats can AI systems reliably identify as leverage points for optimization through automated multi-agent simulation?
Urgent
AI/ML Simulation & Modeling
Related Capability: Simulate prototyping
Can AI generate its own suggested changes and test them to search the latent space for optimal solutions?
Urgent
AI/ML
Related Capability: Simulate prototyping
How can lessons from speculative execution and speculative decoding help increase the availability of deliberative processes through reduced costs?
Urgent
AI/ML Simulation & Modeling
Related Capability: Simulate prototyping
What are the key technical blockers (agent behavior calibration, emergent group dynamics modeling, preference faithfulness) to effective and trustworthy multi-agent simulation, and which are tractable with current methods?
Urgent
AI/ML Simulation & Modeling
Related Capability: Simulate prototyping
Related Resources
Human-centred mechanism design with Democratic AI
Simulated participants built with imitation learning have been used to provide high-volume feedback on possible income redistribution policies (DeepMind, 2022).
Experimental Practice
DeepMind
Deliberativa deliberative labs
Deliberativa headed 'deliberative labs' ahead of both global citizens' assemblies to test out methodological questions of time e.g. best method for multi-lingual deliberation or data capture.
Experimental Practice
Deliberativa
Language Agents as Digital Representatives in Collective Decision-Making
Paper proposes training LLMs as "digital representatives" that can stand in for individual humans in collective decision-making processes, expressing their preferences in group interactions like consensus-finding. It explains the concept of digital representation, defines metrics for evaluating ...
Experimental Practice
DeepMind
Plurals: A System for Guiding LLMs via Simulated Social Ensembles
Introduces Plurals, a Python library for pluralistic AI deliberation using simulated social ensembles with LLM Agents, customizable Structures inspired by deliberative democracy, and Moderators overseeing discussions. The system integrates government datasets to create nationally representative p...
Research
Ashkinaze et al.
Artificial Utopia: Simulation and artificially intelligent agents for exploring Utopian and democratized futures
This paper propose a novel research agenda focusing on ‘utopian’ democratization efforts with formal and computational methods as well as with artificial intelligence – this agenda is labelled ‘Artificial Utopia’. Artificial Utopias provide safe testing grounds for new political ideas and economi...
Research
Yannick Oswald
Democracy on Mars 3: New Tools for Popular Sovereignty
Several have explored versions of this idea, under the keywords 'AI-as-representative' (Collins, 2023), 'voting avatars' (Grandi, 2018), and 'virtual democracy' (Kahng et al., 2019), 'plurals' (Ashkinaze et al., 2024) and 'simulated deliberative democracy' (Leike, 2023).
Research
Tantum Collins
A proposal for importing society’s values
Proposes using imitation learning on large language models trained on recorded deliberative democracy sessions to answer value-laden questions at scale. The approach involves recording human 'mini-publics' deliberating complex value questions with AI assistance, then training models to simulate t...
Research
Jan Leike