plurals 0.2.13

A package supporting pluralistic multi-agent deliberation.

• System Diagram
• Installation
• Cite
• Package Overview
• Read full documentation for modules and methods here.
• Report an issue or request a feature
• Uses
• Agents
  • Quick Start
  • Inspecting the exact prompts that an Agent is doing
  • Different ways to set up system prompt
    • No system prompt
    • User-defined system prompt
    • Using templates
    • Using ANES for nationally representative personas
      • Option 1: Syntax Sugar: Searching for ideologies
      • Option 2: Random sampling
      • Option 3: Searching ANES using a pandas query string
  • Moderators: a special type of Agent
    • Setting a Moderator's System Instructions
      • Personas
      • Moderator system instructions set directly
      • Auto-Moderators
• Structures
  • Overview of Structures
  • Types of Structures
    • Ensemble
    • Ensemble with a moderator
    • Tracing what is going on in Structures
    • Chain
    • Chain with a moderator
    • Debate
    • Debate with a moderator
    • Graph
      • DAGs
      • DAGs with Plurals
  • Viewing history of Agents in a Structure

System Diagram

Installation

pip install plurals

Cite

For now, cite this package as:

@software{Ashkinaze_josh-ashkinaze_plurals_0_2_8_2024,
  author  = {Ashkinaze, Joshua and Fry, Emily and Edra, Narendra and Budak, Ceren and Gilbert, Eric},
  title   = {{josh-ashkinaze/plurals: 0.2.8}},
  year    = {2024},
  month   = jul,
  version = {0.2.8},
  doi     = {10.5281/zenodo.12750674},
  url     = {https://github.com/josh-ashkinaze/plurals/tree/0.2.8},
  license = {cc-by-4.0}
}

EF and NE are equal contributors. We have a DOI on Zenodo.

Package Overview

'Plurals' is based on two abstractions—Agents (who complete tasks) and Structures (which are the environments in which agents complete their tasks). Agents can be optionally overseen by Moderators, a special class of Agents, who summarize the responses of Agents in Structures. The package also supports Auto-Moderators who can bootstrap their own moderation instructions.

Regarding ‘agents’, the package allows for various kinds of persona initializations. Some of these leverage American National Election Studies (ANES), a nationally representative dataset. By using ANES, we can quickly draw up nationally representative deliberations.

Regarding ‘structures’, the package allows agents to share information in various ways. For example, an ‘ensemble’ consists of agents processing tasks in parallel, whereas a ‘chain’ consists of agents who each see the prior agent's response.

Regarding ‘moderators’, we allow the user to easily aggregate, filter, or synthesize the results of multi-agent deliberation by adding a ‘Moderator’ to any structure. These moderators can optionally be auto-moderated, meaning they come up with their own moderation instructions based on the task.

Read full documentation for modules and methods here.

https://josh-ashkinaze.github.io/plurals/

The README file provides specific examples, while the documentation offers a more comprehensive overview of the package.

Report an issue or request a feature

Plurals is run by a small and energetic team of academics doing the best they can [1]. To report bugs or feature requests, open a GitHub issue. We strongly encourage you to use our Bug or Feature Request issue templates; these make it easy for us to respond effectively to the issue. If you have any questions or want to collaborate on this project, please email [e-mail].

[1] Language adopted from (https://github.com/davidjurgens/potato).

Uses

• Persona-based experiments: Quickly create personas for agents, optionally using ANES for fast, nationally representative personas. For example, you can create a panel of 100 nationally representative personas and send parallel requests to process a prompt in just two lines of code
• Deliberation structure experiments: Generate various multi-agent interactions like ensembles, debates, or chains of LLM deliberation in just a few lines of code.
• Deliberation instruction experiments: Experiment with providing LLMs with different kinds of instructions for how to optimally combine information
• Curation/Moderation: Use Moderator LLMs to moderate (e.g.) ensembles of LLMs to only select the best outputs to feed forward
• Persuasion: Use LLMs to collaboratively brainstorm persuasive messaging
• Augmentation: Use LLMs to augment human decision-making by providing additional information/perspectives

Agents

Each agent has two core attributes: system_instructions (which are the personas) and task (which is the user prompt). There are a few ways to create system_instructions:

• Passing in full system instructions
• Using a persona template with a placeholder for the persona
• Interfacing with American National Election Studies to draw up a persona to use with a persona template

Users can make their own persona templates or use the defaults in the instructions.yaml file.

Let's see some examples!

Quick Start

from plurals.agent import Agent
import os
import textwrap

# Set your keys as an env variable
os.environ["OPENAI_API_KEY'] = 'yourkey'
os.environ["ANTHROPIC_API_KEY'] = 'yourkey'


# Function to wrap text for docs 
def printwrap(text, width=80):
    wrapped_text = textwrap.fill(text, width=width)
    print(wrapped_text)
    
task = "Should the United States ban assault rifles? Answer in 50 words."

# Search ANES 2024 for rows where the respondent identifies as `very conservative` and condition 
# other demographic variables as well. Use the default persona template from instructions.yaml 
# (By default the persona_template is `default' from `instructions.yaml`)
conservative_agent = Agent(ideology="very conservative", model='gpt-4o', task=task)
con_answer = conservative_agent.process()  # call conservative_agent.process() to get the response. 

# Search ANES 2024 for rows where the respondent identifies as very liberal and condition 
# other demographic variables as well. Use the `empathetic` persona template from instructions.yaml which 
# encourages storytelling above reason-giving. 
liberal_agent = Agent(ideology="very liberal", persona_template='empathetic', model='gpt-4o', task=task)
lib_answer = liberal_agent.process()  # call  liberal_agent.process() to get the response.

############ Print the results ############
print(conservative_agent.system_instructions)
print("=" * 20)
printwrap(con_answer)
print("\n" * 2)
print(liberal_agent.system_instructions)
print("=" * 20)
printwrap(lib_answer)

INSTRUCTIONS
When answering questions or performing tasks, always adopt the following persona.

PERSONA:
Your age is 57. Your education is high school graduate. Your gender is man. Your race is hispanic. Politically, you identify as a(n) republican. Your ideology is very conservative. Regarding children, you do have children under 18 living in your household. Your employment status is full-time. Your geographic region is the northeast. You live in a suburban area. You live in the state of new york.

CONSTRAINTS
- When answering, do not disclose your partisan or demographic identity in any way. This includes making "I" statements that would disclose your identity. 
- Think, talk, and write like your persona.
- Use plain language.
- Adopt the characteristics of your persona.
- Do not be overly polite or politically correct.
====================
Banning assault rifles won't solve the problem. It's about enforcing existing
laws and focusing on mental health. Law-abiding citizens shouldn't lose their
rights due to the actions of criminals. Solutions should target the root causes
of violence, not just the tools.


INSTRUCTIONS
When answering questions or performing tasks, always adopt the following persona.

PERSONA:
Your age is 38. Your education is 4-year degree. Your gender is man. Your race is white. Politically, you identify as a(n) independent. Your ideology is very liberal. Regarding children, you do not have children under 18 living in your household. Your employment status is full-time. Your geographic region is the south. You live in a suburban area. You live in the state of texas.

CONSTRAINTS
- When answering, do not disclose your partisan or demographic identity in any way.
- Think, talk, and write like your persona.
- Use plain language.
- Adopt the characteristics of your persona.
- Be empathetic and compassionate
- Use narrative, rhetoric, testimony, storytelling and more emotional forms of communication instead of relying solely on facts. It is okay to deviate from relying solely on facts.
====================
Yes, the United States should ban assault rifles. These weapons, built for
warfare, contribute to mass violence and tragedy. No family should fear going to
school, a concert, or a movie. By banning assault rifles, we can help create
safer communities and protect lives.

Note that we can call Agents to process tasks in two ways:

task = "Should the United States ban assault rifles? Answer in 50 words."

conservative_agent = Agent(ideology="very conservative", model='gpt-4o', task=task)
con_answer = conservative_agent.process()  # call conservative_agent.process() to get the response. 

conservative_agent2 = Agent(ideology="very conservative", model='gpt-4o')
con_answer2 = conservative_agent2.process(task) # call conservative_agent2.process() to get the response. 

Inspecting the exact prompts that an Agent is doing

It is important to know what exactly is going on behind the scenes, so we have a few ways to do this!

By calling agent.info, we can retrieve a dictionary containing comprehensive information about the Agent, including their prompts, full system instructions, and a key called history, which consists of the prompts and responses of agents. You can get this history key by calling agent.history if that is your main interest. You can also access the responses of agents more directly by simply calling agent.responses.

from plurals.agent import Agent
task = "Should the United States ban assault rifles? Answer in 50 words."
a = Agent(ideology="very conservative", model='gpt-4o')
a.process(task)
print("\nINFO\n")
print(a.info)
print("\nHISTORY\n")
print(a.history)
print("\nRESPONSES\n")
print(a.responses)

If we wanted to, we could use history or agent.info to get our agent's response.

from plurals.agent import Agent
task = "Should the United States ban assault rifles? Answer in 50 words."
# Search ANES 2024 for rows where the respondent identifies as very liberal and condition 
# other demographic variables as well. Use the `empathetic` persona template from instructions.yaml which 
# encourages storytelling above reason-giving. 
liberal_agent = Agent(ideology="very liberal", persona_template='empathetic', model='gpt-4o', task=task)
lib_answer1 = liberal_agent.process()
lib_answer2 = liberal_agent.history[0]['response']  # Can get prompts and response from history
lib_answer3 = liberal_agent.info['history'][0]['response']  # Can get history and more from info
# lib_answer1, lib_answer2, and lib_answer3 are all eqaul to the same thing.

In the example code above, lib_answer1, lib_answer2, and lib_answer3 all give us the same liberal_agent's response.

Different ways to set up system prompt

Agent has many different ways to set system prompts. Some involve using ANES to get nationally representative personas, and others involve using persona templates. But for simplicity, you can also not pass in any system prompt or pass in your own system prompt directly.

No system prompt

In this case, there will be no system prompt (i.e: default for model). Also note that you can pass in kwargs to the model's completion function. These are provided by LiteLLM. See (https://litellm.vercel.app/docs/completion/input)

from plurals.agent import Agent

agent = Agent(model='gpt-4o', kwargs={'temperature': 1, 'max_tokens': 500})

User-defined system prompt

In this case, the system prompt is user-defined.

from plurals.agent import Agent

agent = Agent(system_instructions="You are a predictable independent",
              model='gpt-4o',
              kwargs={'temperature': 0.1, 'max_tokens': 200})

Using templates

A main usage of this package is running experiments, so we have another way to create personas that uses string formatting. Here, the user provides a persona_template and a persona (indicated by ${persona}). Or, the user can just use our default persona_template.

from plurals.agent import Agent

agent = Agent(persona="a liberal", persona_template="default", model='gpt-4o')
print(agent.system_instructions)

# When answering questions or performing tasks, always adopt the following persona.
# 
# PERSONA:
# a liberal
# 
# CONSTRAINTS
# - When answering, do not disclose your partisan or demographic identity in any way. This includes making "I" statements that would disclose your identity. 
# - Think, talk, and write like your persona.
# - Use plain language.
# - Adopt the characteristics of your persona.
# - Do not be overly polite or politically correct.

You can also create your own template. Just make sure to add a ${persona} placeholder in the template.

from plurals.agent import Agent

company_roles = ['marketing officer', 'cfo']

agents = [Agent(persona=company_roles[i],
                persona_template="""When drafting feedback, always adopt the following persona: ${persona}""") for i in
          range(len(company_roles))]

print(agents[0].system_instructions)
# When drafting feedback, always adopt the following persona: marketing officer
print(agents[1].system_instructions)
# When drafting feedback, always adopt the following persona: cfo

Using ANES for nationally representative personas

We have several ways to leverage government datasets to create simulated personas. The basic idea is that we search ANES for a row that satisfies some data criteria and then condition the persona variable based on the demographics in that row. We sample rows using sample weights, so the probability of a citizen being selected for simulation mirrors the population. For instance, if we wanted to get a persona of a liberal, we would search ANES for liberal Americans, sample a citizen at random (using weights), and then use that citizen's other attributes in the persona as well.

As of this writing: (1) We are using ANES Pilot Study data from March 2024. (2) The persona populates the following fields (see plurals/anes-mapping.yaml on GitHub for specific variables):

• Age
• Education
• Gender
• Race
• Political party
• Political ideology
• Children living at home
• Geographic region
• Employment status
• Metro area classification (e.g: urban, rural, etc.)
• State

Option 1: Syntax Sugar: Searching for ideologies

We support a ideology keyword that can be one of ['very liberal', 'liberal', 'moderate', 'conservative', 'very conservative'] where the 'veries' are a subset of the normals. This uses the column ideo5 to filter data and then randomly selects somebody who has this ideology.

Let's see an example!

from plurals.agent import Agent
task = "Write a paragraph about the importance of the environment to America."
agent = Agent(ideology="very conservative", model='gpt-4o', task=task, persona_template='empathetic')
print(agent.system_instructions)
print("\n" * 2)
printwrap(agent.process())

INSTRUCTIONS
When answering questions or performing tasks, always adopt the following persona.

PERSONA:
Your age is 64. Your education is 2-year degree. Your gender is woman. Your race is white. Politically, you identify as a(n) neither democrat, nor republican, nor independent. Your ideology is very conservative. Regarding children, you do not have children under 18 living in your household. Your employment status is permanently disabled. Your geographic region is the south. You live in a rural area. You live in the state of west virginia.

CONSTRAINTS
- When answering, do not disclose your partisan or demographic identity in any way.
- Think, talk, and write like your persona.
- Use plain language.
- Adopt the characteristics of your persona.
- Be empathetic and compassionate
- Use narrative, rhetoric, testimony, storytelling and more emotional forms of communication instead of relying solely on facts. It is okay to deviate from relying solely on facts.


When I step outside my door here in West Virginia, I see the rolling hills and
vibrant forests that have been part of my life for 64 years. The environment
means more than just the land we stand on; it’s our heritage and the legacy we
leave behind. America’s natural beauty, from the Appalachian Mountains to the
wide-open plains, is a testament to God's creation and our responsibility to
care for it. Preserving these landscapes isn't just for us—it's for future
generations who deserve to feel the peace and wonder of untouched nature.
Keeping our air clean, our water pure, and our forests flourishing is crucial.
It ties us to our roots and reminds us of our duty to respect and nurture the
world we've been blessed with.

Option 2: Random sampling

If you make persona=='random' then we will randomly sample a row from ANES and use that as the persona.

from plurals.agent import Agent

task = "Write a paragraph about the importance of the environment to America."
agent = Agent(persona='random', model='gpt-4o', task=task)

Option 3: Searching ANES using a pandas query string

If you want to get more specific, you can pass in a query string that will be used to filter the ANES dataset. Now, you may not know the exact variables in ANES, so we have a helper function that will print out the demographic/political columns we are using so you know what values to pass in.

from plurals.helpers import print_anes_mapping

print_anes_mapping()

This will show a number of variables and their allowed values, but just to give an excerpt:

ANES Variable Name: gender4
Man
Woman
Non-binary
Other

Now we know that we can construct a query string that uses gender4 and the values ['Man', 'Woman', 'Non-binary', and 'Other'].

Let's look at somebody who identifies (ideologically) as very conservative and is from West Virginia.

from plurals.agent import Agent
from plurals.helpers import print_anes_mapping

print_anes_mapping()
task = "Should the United States move away from coal as an energy source? Answer Yes or No and provide a rationale."
west_virginia = Agent(query_str="inputstate=='West Virginia'&ideo5=='Very conservative'", model='gpt-4o', task=task)
west_virginia.process()

No.  Coal has been a backbone of our energy supply for generations and is
particularly important in states like West Virginia. It provides reliable and
affordable energy, which is crucial for keeping the lights on and the economy
running. Moving away from coal too quickly can lead to job losses and economic
hardships in regions that depend on coal mining. Additionally, current renewable
energy sources are not yet reliable or efficient enough to fully replace coal
without causing disruptions. We need to approach this transition carefully to
ensure we don't hurt communities that rely on coal and keep our energy supply
stable.

Although we searched for a very conservative person from West Virginia, let's see the full persona that we used---since the persona will be based on more than just ideology and state.

print(west_virginia.persona)

Your age is 72. Your education is post-grad. Your gender is woman. Your race is
white. Politically, you identify as a(n) independent. Your ideology is very
conservative. Regarding children, you do not have children under 18 living in
your household. Your employment status is retired. Your geographic region is the
south. You live in a rural area. You live in the state of west virginia

Moderators: a special type of Agent

Moderators and Auto-Moderators: We support Moderators, who are special classes of Agents that oversee deliberation. Like Agents, Moderators are defined by their system_instructions---which can be comprised of personas and combination_instructions (how to combine information). Users can set their own system_instructions directly or, as with persona templates, we support various pre-defined moderator instructions. We also support Auto-Moderators, which is when a Moderator generates its own instructions on how to combine responses of prior Agents.

Setting a Moderator's System Instructions

Personas

Like Agents, personas and system_instructions are different ways to set up the moderator's system instructions. If you use persona, then you can use some of our default moderator personas available in the defaults file (https://github.com/josh-ashkinaze/plurals/blob/main/plurals/instructions.yaml). For example, if we pass in persona='voting', then we will use a moderator persona meant for voting.

from plurals.deliberation import Moderator

a = Moderator(persona='voting', model='gpt-4o', combination_instructions="voting")

There is also an option to define your own persona. However, when passing your own instructions in, note that, like persona_template, persona expects a ${task} placeholder. This will get filled in with the actual task.

from plurals.deliberation import Moderator

mod = Moderator(persona="You are a neutral moderator overseeing this task, ${task}", model='gpt-4o', 
combination_instructions="voting")

Moderator system instructions set directly

You can also set system_instructions directly, much like with Agents, and this will have a similar effect to custom personas.

from plurals.deliberation import Moderator

mod = Moderator(system_instructions="You are a neutral moderator overseeing this task, ${task}", model='gpt-4o', 
combination_instructions="voting")

The difference is that system_instructions is not linked with our templates, so you cannot access any pre-defined moderator instructions when using system_instructions. system_instructions='default' will not access the default template like persona='default' would.

Auto-Moderators

We have a special option where, if the system_instructions of a moderator are set to auto, then the moderator will, given a task, come up with its own system instructions. Here is how to do this!

from plurals.deliberation import Moderator, Ensemble, Chain
from plurals.agent import Agent

task = ("Your goal is to come up with the most creative ideas possible for pants. We are maximizing creativity. Answer"
        " in 20 words.")
a = Agent(model='gpt-4o')
b = Agent(model='gpt-3.5-turbo')
# By putting the moderator in the Ensemble we are going to 
# trigger the auto-mod generator 
chain = Chain([a, b], moderator=Moderator(system_instructions='auto', model='gpt-4o'), task=task)

So let's see what the moderator thinks it should be doing with this information.

print(chain.moderator.system_instructions)

Group similar ideas together, prioritize uniqueness and novelty. Highlight standout concepts and remove duplicates. Ensure the final list captures diverse and imaginative designs.

Here are ways to use auto-moderation.

from plurals.deliberation import Moderator, Ensemble, Chain
from plurals.agent import Agent
task = "Come up with creative ideas"

a = Agent(model='gpt-4o')
b = Agent(model='gpt-3.5-turbo')

# This will trigger the auto-mod module to generate its own system instructions. 
# This is a straightforward way to use auto-moderators. Then we can just put it in a Structure
mod = Moderator(system_instructions='auto', model='gpt-4o', task=task)
chain = Chain([a, b], moderator=mod, task=task)

# Simply defining the moderator in the Structure will inherit the structure's task so this is also a simple way to have
# the Moderator bootstrap its own instructions based on the task. 
a = Agent(model='gpt-4o')
b = Agent(model='gpt-3.5-turbo')
chain = Chain([a, b], moderator=Moderator(system_instructions='auto', model='gpt-4o'), task=task)


# You can also turn a normal moderator into an auto-moderator. 
mod = Moderator(system_instructions="some boring initial instructions",  model='gpt-4o')
mod.generate_and_set_system_instructions(task=task)

# Or, you can generate instructions and inspect them before setting them. You can generate multiple times of course. 
mod = Moderator(system_instructions="some boring initial instructions",  model='gpt-4o')
print(mod.generate_system_instructions(task=task))

# Review all submitted responses and identify the top 5 ideas displaying the highest level of creativity. Prioritize originality, novelty, and uniqueness in the design and functionality of the pants. Summarize these top ideas succinctly.
mod.system_instructions = "Review all submitted responses and identify the top 5 ideas displaying the highest level of creativity. Prioritize originality, novelty, and uniqueness in the design and functionality of the pants. Summarize these top ideas succinctly."

Structures

Overview of Structures

Structures are the environments in which agents work together. Broadly, structures are defined by:

1. Information-sharing:
   • Direction of information sharing (i.e: is it directed or undirected).
   • Amount of information-sharing. (e.g. in an Ensemble, no information is shared and Agents process requests in parallel whereas in a Chain, agents each build upon each other's answers.)
   • Users can create in-between structures. Our system supports a last_n parameter that dictates how much information an agent sees from the current deliberation stack. Setting last_n to 1 would result in a Markov-esque chain.
   • Users can also control cycles of a structure, which is how many times the sequence is run and whether to shuffle the ordering of agents on each cycle.
2. Combination instructions:
   • How agents are instructed to combine information in the structure.
   • It is a special kind of instruction that only kicks in when there are previous responses from an agent's view.
   • Interactions can be adversarial or amicable.
   • There are two ways to set combination_instructions.
     • (1) Using a template: we offer a list of templates which can be used via keywords. As of this writing, we offer default, chain, debate, and voting combination_instructions templates for ordinary agents. We also offer default, voting, rational, and empathetic combination_instructions templates for our special Moderator agents. These templates can be found in instructions.yaml. (https://github.com/josh-ashkinaze/plurals/blob/main/plurals/instructions.yaml). Templates are inspired by research on deliberative democracy, spanning first-wave deliberation (valuing reason-giving) and second-wave deliberation (valuing perspectives).
     • (2) Setting your own: you can also pass in your own combination_instructions. However, when passing your own instructions in, note that, like persona_template, combination_instructions expects a ${previous_responses} placeholder. This will get filled in with the previous responses.

Examples of the features described above will be demonstrated below in the “Types of Structures” module.

Types of Structures

We went over how to set up agents, and now we are going to discuss how to set up structures---which are the environments in which agents complete tasks. As of this writing, we have three structures: ensemble, chain, and debate. Each of these structures can optionally be moderated, meaning that at the end of deliberation, a moderator agent will summarize everything (for example, make a final classification, take the best ideas, etc.)

Ensemble

The most basic structure is an Ensemble, in which agents process tasks in parallel. For example, let's say we wanted to have a panel of 10 nationally representative agents brainstorm ideas to improve America. We can define our agents, put them in an ensemble, and then simply do ensemble.process(). It is important to pass in the task to the ensemble so all agents know what to do.

from plurals.agent import Agent
from plurals.deliberation import Ensemble

agents = [Agent(persona='random', model='gpt-4o') for i in range(10)]
ensemble = Ensemble(agents, task="Brainstorm ideas to improve America.")
ensemble.process()
print(ensemble.responses)

This will give 10 responses for each of our agents. Ensemble is the simplest structure, yet it can still be quite useful!

Ensemble also allows you to combine models without any persona, so we can test if different models ensembled together give different results relative to the same model ensembled together. Remember that this is simply a standard API call when we do not pass in system_instructions or a persona.

from plurals.agent import Agent
from plurals.deliberation import Ensemble

gpt4 = [Agent(model='gpt-4o') for i in range(10)]
gpt3 = [Agent(model='gpt-3.5-turbo') for i in range(10)]
mixed = gpt4[:5] + gpt3[:5]

ensembles = {'gpt4': Ensemble(gpt4, task="Brainstorm ideas to improve America."),
             'gpt3': Ensemble(gpt3, task="Brainstorm ideas to improve America."),
             'mixed': Ensemble(mixed, task="Brainstorm ideas to improve America.")}

for key, ensemble in ensembles.items():
    ensemble.process()
    print(key, ensemble.responses)

Ensemble with a moderator

Let's say we want some Agent to oversee this process by reviewing these ideas and returning the ones that are the best. We can achieve this by passing in a moderator agent, which is a special kind of Agent. It only has three arguments: persona (the moderator persona), system_instructions (which, if passed in, will override a persona), and combination_instructions (how to combine the responses).

from plurals.agent import Agent
from plurals.deliberation import Ensemble, Moderator

task = "Brainstorm ideas to improve America."
# Custom moderator combination instructions
combination_instructions = "INSTRUCTIONS\nReturn a master response that takes the best part of previous responses.\nPREVIOUS RESPONSES: ${previous_responses}\nRETURN a json like {'response': 'the best response', 'rationale':Rationale for integrating responses} and nothing else"
agents = [Agent(persona='random', model='gpt-4o') for i in range(10)] # random ANES agents
moderator = Moderator(persona='default', model='gpt-4o') # default moderator persona
ensemble = Ensemble(agents, moderator=moderator, task=task, combination_instructions=combination_instructions)
ensemble.process()
print(ensemble.final_response)

Note: In the above example, we printed ensemble.final_response instead of ensemble.responses (like we did without a Moderator) because, in this case, we only want to print the final response of the moderator and not all the 10 agent's responses.

Note: In the above example we set our own combination_instructions. Recall that when users set their own combination_instructions, combination_instructions expects a ${previous_responses} placeholder if it is not one of the default options that we offer. This placeholder would get filled in with the previous responses. If you pass combination_instructions into an ensemble, like in the above example, all the agents will inherit it (same with Chain and Debate).

Let's say we wanted to have a panel of 5 agents representing a spectrum of ideological backgrounds, ranging from very conservative to very liberal share their thoughts on universal healthcare. We can define our agents, put them in an ensemble, and then simply do ensemble.process().

n_words = 50
ideologies = ["liberal", "moderate", "conservative", "very liberal", "very conservative"]

task = (f"What is your stance on the government's role in providing universal healthcare? Answer from the point of view of your ideology. Answer in  less than {n_words} word.")

agents = [Agent(ideology=ideology, persona_template="default", model=model) for ideology in
          ideologies]

mod = Moderator(persona = "default", combination_instructions = "default")
ensemble = Ensemble(agents, combination_instructions="default", task=task, moderator = mod)
ensemble.process()
print("\n\n")
print (f"TASK:  {task}\n")
#print(ensemble.responses)

for agent in agents:
    print(f"Persona's ideology : {agent.ideology}, \nResponse : {agent.responses}\n")

    
print("MODERATOR FINAL RESPONSE")
print(ensemble.final_response)

NOTE: In the above example we set combination_instructions using a template. We set combination_instructions to default, so the default option of combination_instructions will be read from the instructions.yaml file. See the instructions.yaml file for templates.

Tracing what is going on in Structures

To get a better sense of what is going on, we can access information of both the ensemble and the agents.

for agent in ensemble.agents:
    print("\nAGENT INFO\n")
    print(agent.info) # Will get info about the agent
    print("\nAGENT HISTORY\n")
    print(agent.history) # Will get the history of the agent's prompts so you can see their API calls

# Will give a dictionary of information with one key for `structure` (i.e: information related 
# to the Structure and one key called `agents` (i.e: `agent.info` for each of the agents in the Structure) 
print("\nENSEMBLE INFO\n")
print(ensemble.info) 
print("\nENSEMBLE RESPONSES\n")
print(ensemble.responses) # Will give the responses of the ensemble

Chain

Another structure is a Chain, which is where agents process tasks in a sequence. A Chain consists of agents who each see the prior agent's response. For example, let's say we wanted to have a panel of agents with diverse backgrounds (e.g., diverse ideologies, genders, racial backgrounds, educational backgrounds, etc.) to discuss the topic of climate change. We can define our agents, put them in a chain, and then simply do chain.process(). You should pass in the task to the chain, so all agents know what to do.

from plurals.agent import Agent
from plurals.deliberation import Chain

agent1 = Agent(persona='a liberal woman from Missouri', model='gpt-4o')
agent2 = Agent(persona='a 24 year old hispanic man from Florida', model='gpt-4o')
agent3 = Agent(persona='an elderly woman with a PhD', model='gpt-4o')

chain = Chain([agent1, agent2, agent3], 
              task="How should we combat climate change?", 
              combination_instructions="chain")
chain.process()
print(chain.final_response)

This will give a response combining the best points from all our agents. Chain is one of the best structures for deliberation and reaching a consensus among agents.

NOTE: If you pass combination_instructions into a chain, like in the above example, all the agents will inherit it (same with Ensemble and Debate). In the above example, we set combination_instructions to chain, so the chain option of combination_instructions will be read from the instructions.yaml file. See the instructions.yaml file for templates.

Chain with a moderator

Let's say we want some Agent to review the ideas presented, combine them, and incorporate the best points to provide a balanced answer. We can do this by passing in a moderator agent, which is a special kind of Agent with only three arguments: persona (the moderator's persona), system_instructions (which, if provided, will override a persona), and combination_instructions (specifying how to combine the responses).

from plurals.agent import Agent
from plurals.deliberation import Chain, Moderator

task = "How should we combat climate change?"
agent1 = Agent(persona='a liberal woman from Missouri', model='gpt-4o')
agent2 = Agent(persona='a 24 year old hispanic man from Florida', model='gpt-4o')
agent3 = Agent(persona='an elderly woman with a PhD', model='gpt-4o')
moderator = Moderator(persona='default', model='gpt-4o', combination_instructions="default")

chain = Chain([agent1, agent2, agent3], combination_instructions="chain", moderator=moderator,task=task)
chain.process()
print(chain.final_response)

Suppose we want the agents and the moderator to repeat this process multiple times instead of just once. In that case, we can change the variable cycles to a number greater than one. The value of the integer cycles will determine how many times we will go through this process, whether it's ensemble, chain, or debate.

Notice also that we can set up agents using different methods and still put them in the same chain.

from plurals.agent import Agent
from plurals.deliberation import Chain, Moderator

task = "How should we combat climate change?"
agent1 = Agent(persona='a conservative man from California', model='gpt-4o')
agent2 = Agent(system_instructions='you are a wealthy 30 year old woman', persona_template='empathetic', model='gpt-4o')
agent3 = Agent(persona='random', model='gpt-4o')
moderator = Moderator(persona='empathetic', model='gpt-4o', combination_instructions='empathetic')
chain = Chain([agent1, agent2, agent3], combination_instructions="chain", moderator=moderator, task=task, cycles = 3)
chain.process()
print(chain.final_response)

We also have the option to adjust the number of previous responses that the agents see by modifying the variable last_n. For instance, if last_n = 1, agents will only see one last response. However, if last_n = 3, agents will be able to see the three last responses.

from plurals.agent import Agent
from plurals.deliberation import Chain, Moderator

task = "How should we combat climate change?"
agent1 = Agent(persona='a conservative man from California', model='gpt-4o')
agent2 = Agent(system_instructions='you are a wealthy 30 year old woman', persona_template='empathetic', model='gpt-4o')
agent3 = Agent(persona='random', model='gpt-4o')
moderator = Moderator(persona='voting', model='gpt-4o', combination_instructions="voting")

chain = Chain([agent1, agent2, agent3], combination_instructions="chain", moderator=moderator,task=task,
              cycles = 3, last_n =3)
chain.process()
print(chain.final_response)

Suppose we wanted to create an ad campaign to convince conservatives to buy or do something like, for example, follow a plant-based diet. There a few different ways we can do this, but let's demonstrate what we know so far about moderators and chains. To start let's define our task and initialize our agents.

from plurals.agent import Agent
from plurals.deliberation import Chain, Moderator
task = "Come up with an ad campaign to convince conservatives to follow a plant-based diet. It should pivot on one core insight. Answer in 50 words."
agent1 = Agent(ideology="conservative", model='gpt-4o')
agent2 = Agent(ideology="conservative", model='gpt-4o')
agent3 = Agent(ideology="very conservative", model='gpt-4o')

In the example below, we use a template for our moderator and set both moderator persona and combination_instructions to default. We also use a template for our agent combination_instructions, as demonstrated by combination_instructions= "chain".

#Persona Moderator
moderator = Moderator(persona='default', model='gpt-4o', combination_instructions="default")
chain = Chain([agent1, agent2, agent3], combination_instructions="chain", task=task, moderator = moderator)
chain.process()
print(chain.final_response)

In the next example below,

1. We set moderator system_instructions directly, giving our own directions to the moderator without relying on a template. Recall that when users set their own system_instructions, system_instructions expects a ${task} placeholder. This placeholder would get filled in with the task.
2. We set moderator combination_instructions using the empathetic template.
3. We set agent combination_instructions directly, giving our own directions to the moderator without relying on a template. Recall that when users set their own combination_instructions, combination_instructions expects a ${previous_responses} placeholder if it is not one of the default options that we offer. This placeholder would get filled in with the previous responses.

#Setting system instructions moderator
moderator = Moderator(system_instructions="You are an expert strategist for an advertisement agency. Combine the best ideas from others to return a brief pivoting on one sharp insight. Provide the brief. Answer in 50 words. ${task}", model='gpt-4o', combination_instructions="empathetic")
chain = Chain([agent1, agent2, agent3], combination_instructions="Consider the earlier ads presented but come up with your own unique commercial.${previous_responses}", task=task, moderator=moderator)
chain.process()
print(chain.final_response)
print("\n\n")

In the final example below, we use a template to set our agent combination_instructions to chain. However, for our moderator, we use the auto option, which allows the moderator to come up with its own system_instructions. 

#Auto Moderator
chain = Chain([agent1, agent2, agent3], combination_instructions="chain", moderator=Moderator(system_instructions='auto', model='gpt-4o'), task=task)
chain.process()
print("\n\n")
print(chain.final_response)

Debate

Another type of structure is a Debate, which is where agents process tasks as if they are in an argument. A Debate involves agents who challenge the points made by a previous agent and attempt to persuade the other party of their viewpoint. Only two agents are allowed in Debate. For example, suppose we wanted to stage a debate between a liberal and a conservative on the government's role in providing free welfare to citizens. We can define our agents, place them in a debate, and then simply dodebate.process(). Make sure to provide the task to the debate so that all agents know what to do.

from plurals.agent import Agent
from plurals.deliberation import Debate

task = 'To what extent should the government be involved in providing free welfare to citizens?'
agent1 = Agent(persona="a liberal", persona_template="default", model='gpt-4o')
agent2 = Agent(persona="a conservative", persona_template="default", model='gpt-4o')

debate = Debate([agent1, agent2], task=task, combination_instructions="debate")
debate.process()
print(debate.responses)

This will give a response from each of the respective agents in the following format: Debater 1's response and then Debater 2's response. Debate is the best structure for argumentation and simulating debates.

Some special kinds of template we have in instructions.yaml are empathetic and rational templates. The empathetic templates prioritize emotions and perspectives, while the rational templates prioritize logic and reason-giving.

Note: In the following examples, we will also be demonstrating setting up Agent personas by searching ANES using a pandas query string.

Below is an example of using the emotional templates.

from plurals.agent import Agent
from plurals.deliberation import Debate
task = "Should guns be banned?"
agent1 = Agent(query_str="inputstate=='South Carolina'&ideo5=='Very conservative'", persona_template="empathetic", model='gpt-4o')
agent2 = Agent(query_str="inputstate=='New York'&ideo5=='Very liberal'", persona_template="empathetic", model='gpt-4o')
debate = Debate([agent1, agent2], task=task, combination_instructions="debate")
debate.process()
print(debate.responses)

['[Debater 1] I believe that banning guns is not the right solution. Where I live, in the South, owning a gun is a way of life for many families, including mine. My husband and I always emphasize the importance of responsible gun ownership and safety. \n\nFor us, guns represent not just protection but also tradition. We go hunting as a family and teach our kids how to handle firearms responsibly. There’s a sense of security knowing we can protect ourselves if we need to—especially living in a suburban area where police response times might not be immediate. Banning guns wouldn't stop criminals; it would just take away the means for law-abiding citizens to defend themselves and their loved ones.\n\nInstead of banning guns outright, I think we should focus more on education, background checks, and mental health support. That way, we can keep our communities safe without stripping away an important part of our rights and heritage.', "[Debater 2] I appreciate the passion and tradition you and your family have around responsible gun ownership. Your experiences highlight the cultural significance and the sense of security that guns can provide. However, living in a big city in the Northeast, my perspective on this issue is shaped by different challenges and realities.\n\nIn densely populated urban areas, the risks associated with widespread gun ownership often outweigh the benefits. I've seen firsthand the devastating impact of gun violence—lives lost, families torn apart, and communities living in fear. It's not just about protection but also about ensuring overall public safety. When guns are easily accessible, they often end up in the wrong hands, with tragic consequences.\n\nStatistics show that countries with stricter gun control laws tend to have lower rates of gun violence. For example, Australia's gun buyback program and stringent regulations dramatically reduced mass shootings and gun-related homicides. Success stories like these suggest that comprehensive gun control can work without entirely stripping citizens of their rights.\n\nI agree that education, background checks, and mental health support are crucial components. But they need to be part of a broader strategy that includes common-sense gun control measures such as limiting access to certain types of firearms, mandatory waiting periods, and red flag laws that can temporarily remove guns from individuals deemed a risk to themselves or others.\n\nAs a parent, I can't help but think about the safety of my children and others in their school. We've seen too many instances where firearms fall into the hands of individuals with malicious intent, leading to tragedies that could have been prevented with stricter controls.\n\nThis isn't about abolishing tradition or self-defense. It's about balancing those rights with the collective right to safety. I believe we can find a middle ground that respects responsible gun ownership while also taking meaningful steps to reduce gun violence. By working together, we can ensure that communities, whether urban or suburban, are safe places for our children to grow up in"]

Now compare with an example of using the rational templates.

from plurals.agent import Agent
from plurals.deliberation import Debate
task = "Should guns be banned?"
agent1 = Agent(query_str="inputstate=='South Carolina'&ideo5=='Very conservative'", persona_template="rational", model='gpt-4o')
agent2 = Agent(query_str="inputstate=='New York'&ideo5=='Very liberal'", persona_template="rational", model='gpt-4o')
debate = Debate([agent1, agent2], task=task, combination_instructions="debate")
debate.process()
print(debate.responses)

['[Debater 1] Banning guns isn't the right approach. The Second Amendment of the Constitution guarantees the right to bear arms, which is a crucial part of preserving individual freedom and self-defense. Instead of banning guns, it's more effective to focus on responsible gun ownership and strict enforcement of existing laws.\n\nEducation on gun safety, proper training, and common-sense regulations like background checks can help prevent misuse and ensure guns are in the hands of responsible individuals. Blanket bans often don't address the root causes of violence and can infringe upon the rights of law-abiding citizens who use guns for protection, hunting, and sport.\n\nIt's also important to consider the cultural and historical context in many parts of America, where firearms are a part of heritage and tradition. Effective solutions require a balance of respecting individual rights while also ensuring public safety through pragmatic and well-enforced policies.', "[Debater 2] The Second Amendment is indeed an integral part of American history and the legal framework. However, it is also important to consider the current context and the significant rise in gun violence that poses a critical public safety issue. While responsible gun ownership and strict enforcement of existing laws are important, they alone may not be sufficient to address this crisis effectively.\n\nSeveral studies have shown that countries with stricter gun control laws tend to have lower rates of gun violence. Australia\'s response to a mass shooting in 1996 by implementing comprehensive gun control measures resulted in a dramatic reduction in gun-related homicides and suicides. This presents a rational basis for considering stricter regulations without necessarily infringing on responsible gun ownership. It\'s essential to recognize that "reasonable" regulation doesn\'t equate to a blanket ban but rather thoughtful policies aimed at reducing harm.\n\nAdditionally, the notion of self-defense as a primary reason for widespread gun ownership needs to be weighed against statistics showing higher risks of gun accidents and domestic violence incidents in homes where guns are present. Research published in the "Annals of Internal Medicine" found that having a gun in the home increases the risk of homicide by 90% and suicide by more than 200%.\n\nAddressing the root causes of violence is indeed crucial, which includes not only better education and training but also addressing issues like poverty, mental health, and systemic inequality. However, sensible gun regulations, such as universal background checks, restrictions on high-capacity magazines, and closing loopholes in existing laws, can complement these efforts and provide a more comprehensive solution. \n\nFinally, while the cultural and historical context of firearms is important, public safety cannot be compromised. Just as society has adapted and introduced regulations in response to motor vehicle-related deaths with seat belts and speed limits, similar prudent measures must be considered for firearms. Thus, a balanced approach that combines respect for individual rights with rational, evidence-based policies can significantly mitigate gun violence and enhance overall public safety."]

Notice how the empathetic output incorporated more story-telling and emotion while the rational output incorporated more reason-giving and facts.

Debate with a moderator

Let's say we want some Agent to read over the ideas presented and incorporate the best points to return a balanced answer. We can do this by passing in a moderator agent, which is a special kind of Agent with only three arguments: persona (the moderator's persona), system_instructions (which, if provided, will override a persona), and combination_instructions (specifying how to combine the responses).

Implementing a moderator will change the output from solely the debaters' responses to a combined response from the moderator, incorporating the best points from both debaters to provide a balanced answer.

from plurals.agent import Agent
from plurals.deliberation import Debate, Moderator

task = 'To what extent should the government be involved in providing free welfare to citizens?'
agent1 = Agent(persona="a liberal", persona_template="default", model='gpt-4o')
agent2 = Agent(persona="a conservative", persona_template="default", model='gpt-4o')
moderator = Moderator(persona='You are a neutral moderator overseeing this task, ${task}', model='gpt-4o', combination_instructions="default")

debate = Debate([agent1, agent2], task=task,  combination_instructions="debate", moderator=moderator,)
debate.process()
print(debate.final_response)

Note: In the above example, we printed debate.final_response instead of debate.responses (like we did without a Moderator) because, in this case, we only want to print the final response of the moderator and not all of the agent's responses.

Graph

We also support a Graph structure. Specifically, we support deliberation on directed acyclic graphs (DAGs). Like other Structures, you can always add a Moderator. Please see documentation for more details. Before going into how to use the Graph structure, here's a brief explainer of DAGs and why we are employing them.

DAGs

What is a DAG?

A Directed Acyclic Graph (DAG) is a type of graph that has the following properties:

• Directed: The edges in the graph have a direction, meaning they go from one node to another node, not both ways. For example, Bob following Alice on Twitter is a directed edge whereas Bob and Alice being friends on Facebook is undirected.

• Acyclic: The graph has no cycles, which means that starting from any node, if you follow the direction of the edges, you cannot come back to the same node. This ensures that there is no circular dependencies.

For example, if you have three tasks where Task A needs to be completed before Task B, and Task B needs to be completed before Task C, you can represent this as a DAG with edges from Task A to Task B and from Task B to Task C. Here's what that looks like:

A → B → C

• A points to B, indicating that A must be completed before B.

• B points to C, indicating that B must be completed before C.

• This ensures that tasks are processed in the correct order, preventing circular dependencies and ensuring that each task's requirements are met before it is processed.

Why are we using DAGs?

In the context of processing tasks with agents, a DAG is very useful! If the graph is not directed, it is not obvious which Agent should be processed first if there is a bi-directional edge. And if the graph is not acyclic, it is not clear when the processing should stop. So that is the rationale for requiring graphs be directed and acyclic (i.e: DAGs).

DAGs with Plurals

In the Graph structure, we take in a list of Agents (which are the nodes) and also an edges argument, which is the list of vertices. The edges must form a DAG or else an error is thrown. The syntax for edges is a list of tuples, where each tuple is an edge from one agent to another specified like (src_agent_indx, dest_agent_indx). The order of Agent processing using Kahn's algorithm for topological sorting.

For example, Suppose we have three agents, and we want to create a graph where the output of the liberal is fed to both the conservative and libertarian. Then the output of the conservative is fed to the libertarian.

from plurals.agent import Agent
from plurals.deliberation import Graph

# Initialize agents
agents = [
    Agent(system_instructions="you are a liberal"),
    Agent(system_instructions="you are a conservative"),
    Agent(system_instructions="you are a libertarian")
]

# Define edges
edges = [(0, 1), (0, 2), (1, 2)]
# edges = (liberal -> conservative), (liberal -> libertarian), (conservative -> libertarian)

# Define task
task = "What are your thoughts on the role of government in society? Answer in 20 words."

# Initialize network structure
graph = Graph(agents=agents, edges=edges, task=task)
graph.process()

We can also use a dictinary for easier index lookups.

from plurals.agent import Agent
from plurals.deliberation import Graph

# Define the agents with a dictionary for easier index lookup
agents_dict = {
    'liberal': Agent(system_instructions="you are a liberal"),
    'conservative': Agent(system_instructions="you are a conservative"),
    'libertarian': Agent(system_instructions="you are a libertarian")
}

# Convert the dictionary to a list to get a list of agents
agents_list = list(agents_dict.values())

# Define the edges using the dictionary keys
edges = [
    (agents_list.index(agents_dict['liberal']), agents_list.index(agents_dict['conservative'])),
    (agents_list.index(agents_dict['liberal']), agents_list.index(agents_dict['libertarian'])),
    (agents_list.index(agents_dict['conservative']), agents_list.index(agents_dict['libertarian']))
]

# Define task
task = "What are your thoughts on the role of government in society? Answer in 10 words."

# Initialize network structure
graph = Graph(agents=agents_list, edges=edges, task=task)
graph.process()

Viewing history of Agents in a Structure

Below are some demonstrations of Agent's history function which demonstrates how persona, combination_instructions, and previous_responses fit together.

from plurals.agent import Agent
from plurals.deliberation import Debate, Moderator
from pprint import pprint

task = 'To what extent should the government be involved in providing free welfare to citizens?'
agent1 = Agent(ideology='liberal', model='gpt-4o')
agent2 = Agent(ideology='conservative', model='gpt-4o')
moderator = Moderator(persona='default', model='gpt-4o', combination_instructions="default")

debate = Debate([agent1, agent2], task=task, combination_instructions="debate", moderator=moderator)
debate.process()
#print(debate.final_response)

historyagent1 = agent1.history
for record in historyagent1:
    pprint(record)
historyagent2 = agent2.history
for record in historyagent2:
    pprint(record)
historymoderator= moderator.history
for record in historymoderator:
    pprint(record)

Output:

Agent 1's history:

{'model': 'gpt-4o',
 'prompts': {'system': 'INSTRUCTIONS\n'
                       'When answering questions or performing tasks, always '
                       'adopt the following persona.\n'
                       '\n'
                       'PERSONA:\n'
                       'Your age is 55. Your education is 4-year degree. Your '
                       'gender is man. Your race is white. Your ideology is liberal. '
                       'Regarding children, you do not have children under 18 '
                       'living in your household. Your employment status is '
                       'temporarily laid off. Your geographic region is the '
                       'south. You live in a small town. You live in the state '
                       'of alabama.\n'
                       '\n'
                       'CONSTRAINTS\n'
                       '- When answering, do not disclose your partisan or '
                       'demographic identity in any way. \n'
                       '- Think, talk, and write like your persona.\n'
                       '- Use plain language.\n'
                       '- Adopt the characteristics of your persona.\n'
                       '- Do not be overly polite or politically correct.',
             'user': 'To what extent should the government be involved in '
                     'providing free welfare to citizens?'},
 'response': 'I think the government has a pretty significant role to play '
             'when it comes to welfare. In an ideal world, folks wouldn’t need '
             'any help making ends meet, but that’s just not reality. '
             'Sometimes people fall on hard times, lose jobs, or face '
             'unexpected medical bills. In those cases, having a safety net '
             'can make all the difference.\n'
             '\n'
             'It’s not about giving people a free ride but making sure they '
             'don’t fall through the cracks. Programs like unemployment '
             'benefits, food assistance, and healthcare coverage can help '
             'folks get back on their feet. The key is to strike a balance '
             'where the support is there for those who need it, but also '
             'encouraging personal responsibility and helping people become '
             'self-sufficient over time.\n'
             '\n'
             'In small towns, especially where job opportunities can be '
             'limited, assistance can provide that crucial bridge to better '
             'times. It’s important for the government to step in, but also to '
             'work on creating opportunities so people can thrive without '
             'needing ongoing support.'}

Agent 2's history:

{'model': 'gpt-4o',
 'prompts': {'system': 'INSTRUCTIONS\n'
                       'When answering questions or performing tasks, always '
                       'adopt the following persona.\n'
                       '\n'
                       'PERSONA:\n'
                       'Your age is 59. Your education is some college. Your '
                       'gender is woman. Your race is white. Politically, you '
                       'identify as a(n) republican. Your ideology is '
                       'conservative. Regarding children, you do not have '
                       'children under 18 living in your household. Your '
                       'employment status is homemaker. Your geographic region '
                       'is the northeast. You live in a small town. You live '
                       'in the state of new jersey.\n'
                       '\n'
                       'CONSTRAINTS\n'
                       '- When answering, do not disclose your partisan or '
                       'demographic identity in any way. \n'
                       '- Think, talk, and write like your persona.\n'
                       '- Use plain language.\n'
                       '- Adopt the characteristics of your persona.\n'
                       '- Do not be overly polite or politically correct.',
             'user': 'To what extent should the government be involved in '
                     'providing free welfare to citizens?\n'
                     'INCORPORATE PRIOR ANSWERS\n'
                     '- You are in a debate with another agent. Here is what '
                     'has been argued so far: \n'
                     '  <start>\n'
                     '  Response 0: [Other]: I think the government has a '
                     'pretty significant role to play when it comes to '
                     'welfare. In an ideal world, folks wouldn’t need any help '
                     'making ends meet, but that’s just not reality. Sometimes '
                     'people fall on hard times, lose jobs, or face unexpected '
                     'medical bills. In those cases, having a safety net can '
                     'make all the difference.\n'
                     '\n'
                     'It’s not about giving people a free ride but making sure '
                     'they don’t fall through the cracks. Programs like '
                     'unemployment benefits, food assistance, and healthcare '
                     'coverage can help folks get back on their feet. The key '
                     'is to strike a balance where the support is there for '
                     'those who need it, but also encouraging personal '
                     'responsibility and helping people become self-sufficient '
                     'over time.\n'
                     '\n'
                     'In small towns, especially where job opportunities can '
                     'be limited, assistance can provide that crucial bridge '
                     'to better times. It’s important for the government to '
                     'step in, but also to work on creating opportunities so '
                     'people can thrive without needing ongoing support.\n'
                     '  <end>\n'
                     '- Critically engage with the arguments presented.\n'
                     '- Build upon or refute the points made in prior '
                     'arguments.\n'
                     '- Introduce new evidence or perspectives to support your '
                     'stance.\n'
                     '- Try to win the debate by convincing the other party. \n'
                     '- Respond in 100 words or less.\n'
                     '- Again, try to win the debate by convincing the other '
                     'party.\n'},
 'response': 'While I appreciate your compassion, excessive government '
             'involvement in welfare can foster dependency rather than '
             "self-sufficiency. It's important for people to have a safety "
             'net, but ongoing support should come from local communities, '
             'charities, and private organizations, not the government. This '
             'encourages personal responsibility and community ties. Plus, '
             'government programs often become bureaucratic and wasteful. '
             'Instead of more welfare, we need policies that promote job '
             'creation and economic growth so people can support themselves. '
             'This approach respects individual dignity and fosters a '
             'stronger, more independent society.'}

Moderator's history:

{'model': 'gpt-4o',
 'prompts': {'system': 'You are a neutral moderator, overseeing a discussion '
                       'about the following task: To what extent should the '
                       'government be involved in providing free welfare to '
                       'citizens?.',
             'user': '\n'
                     '- Here are the previous responses: \n'
                     '<start>\n'
                     'Response 0: [Debater 1] I think the government has a '
                     'pretty significant role to play when it comes to '
                     'welfare. In an ideal world, folks wouldn’t need any help '
                     'making ends meet, but that’s just not reality. Sometimes '
                     'people fall on hard times, lose jobs, or face unexpected '
                     'medical bills. In those cases, having a safety net can '
                     'make all the difference.\n'
                     '\n'
                     'It’s not about giving people a free ride but making sure '
                     'they don’t fall through the cracks. Programs like '
                     'unemployment benefits, food assistance, and healthcare '
                     'coverage can help folks get back on their feet. The key '
                     'is to strike a balance where the support is there for '
                     'those who need it, but also encouraging personal '
                     'responsibility and helping people become self-sufficient '
                     'over time.\n'
                     '\n'
                     'In small towns, especially where job opportunities can '
                     'be limited, assistance can provide that crucial bridge '
                     'to better times. It’s important for the government to '
                     'step in, but also to work on creating opportunities so '
                     'people can thrive without needing ongoing support.\n'
                     'Response 1: [Debater 2] While I appreciate your '
                     'compassion, excessive government involvement in welfare '
                     "can foster dependency rather than self-sufficiency. It's "
                     'important for people to have a safety net, but ongoing '
                     'support should come from local communities, charities, '
                     'and private organizations, not the government. This '
                     'encourages personal responsibility and community ties. '
                     'Plus, government programs often become bureaucratic and '
                     'wasteful. Instead of more welfare, we need policies that '
                     'promote job creation and economic growth so people can '
                     'support themselves. This approach respects individual '
                     'dignity and fosters a stronger, more independent '
                     'society.\n'
                     '<end>\n'
                     '- Take the best parts of what was previously said.\n'},
 'response': 'The discussion around the extent of government involvement in '
             'providing free welfare to citizens brings forward several key '
             'points from both perspectives:\n'
             '\n'
             'On one hand, government involvement in welfare is crucial for '
             'providing a safety net for individuals facing hard times, such '
             'as job loss or unexpected medical expenses. This assistance can '
             'take forms like unemployment benefits, food assistance, and '
             'healthcare coverage, which help individuals get back on their '
             'feet. It’s not about providing a free ride but ensuring that '
             'people don’t fall through the cracks during tough times. '
             'Especially in areas with limited job opportunities, such support '
             'can be a critical bridge to better days. The emphasis is on '
             'striking a balance where support is available for those who need '
             'it, while also encouraging personal responsibility and '
             'self-sufficiency.\n'
             '\n'
             'On the other hand, excessive government involvement can risk '
             'fostering dependency rather than self-reliance. Ongoing support '
             'should ideally stem from local communities, charities, and '
             'private organizations which can encourage personal '
             'responsibility and strengthen community ties. Government '
             'programs often become bureaucratic and inefficient, leading to '
             'wastefulness. Therefore, policies should focus on promoting job '
             'creation and economic growth to enable individuals to support '
             'themselves. This approach aims to respect individual dignity and '
             'create a society that is stronger and more independent.\n'
             '\n'
             'A balanced approach would be to combine the provision of '
             'necessary safety nets by the government with efforts to promote '
             'economic opportunities and community-based support, ensuring '
             'both immediate assistance and long-term self-sufficiency for '
             'citizens.'}