L1-3 intro multi-agent sys + agent-based modeling, ABMs in cognitive + computational social science

Module 1. Introduction to Multi- Agent Systems • Module 2. Intro to Agent-Based Modeling • Module 3. On Using ABMs in Cognitive and Computational Social Science

What is an agent?

An entity that has perception-action capabilities

• It can sense its environment and act in it

• ex. humans, animals, robots, kinds of software

What are multi-agent systems?

• A systems or group of (potentially) interacting agents

  • in some environment that they can sense and act in

  • can communicate and solve problems together

• Can form the bases for distributed AI systems

• We also model multi-agent systems that exist in nature to try to understand how they work (ants, birds, economies, etc.)

• Whole (system) is greater than the sum of the parts (agents)

What are examples of multi-agent systems?

• insect colonies

• humans

• organizations

• video games

• economies

• autonomous vehicles

What are 6 characteristics of multi-agent systems?

• agent design

• environment

• perception

• control

• knowledge

• communication

What is environment in MAS?

static vs dynamic

• MAS often dynamic (especially with learning + interactions)

What is perception in MAS?

• Information is distributed in environment

  • Spatially, temporally, semantically

• Partial observability

  • Makes action planning challenging

What is control in MAS?

• Decentralized (emergent, self-organized)

  • Robust

  • Hard to divide decision-making

• Game theory and coordination

• Different control architectures and rules are possible

What is knowledge in MAS?

• Levels of knowledge may differ between agents

• Common or shared knowledge structures are important

  • Knowing what other agents know

  • Shared mental models, situation awareness, transactional memory systems

What is communication in MAS?

• Two-way sender receivers

• Needed for coordination and negotiation

• Protocols for heterogeneous agents

What are applications of MAS?

• E-commerce, trading, auctions

• Robotics

• Computer games

• Social and cognitive science

• Internet

• Human-machine teaming

What are challenges of MAS?

• How can we understand and solve problems with multi-agent systems?

• How can agents maintain a shared understanding of their environment?

• How can we design agents that coordinate and resolve conflicts?

• What kind of learning mechanisms are there for agents?

• How can agents of different types interact effectively?

What is a model?

An abstracted description of a process, object, or event

• Not a perfect representation

• Often wrong in many ways, but still useful

What is a simulation?

Evolution of a model over time

• Often computer based

What is agent-based modeling?

• Agent: Autonomous individual elements with properties and actions in computer simulation

• ABM: World can be modeled using agents, environment, and description of agent-agent and agent-environment interactions

What are equation-based models?

Equation-based models have closed form solutions:

• Continuous

• No local details

• Top-down versus bottom-up (ABM)

• Can be converted to ABM to complement EDMs

What are lab experiments?

• Lab experiments can generate theory

• Lab experiments are rarely scaled up

• ABM can be created for lab experiments

  • Generate new hypotheses

  • Determine sensitivity of results

  • Can compare generative principles from ABM with lab experiments

What are Limitations and Resistance for ABMS?

• High computational cost

• Many free parameters

• Requires detailed individual-level behavioral knowledge

• Resistance

  • Centralized control mechanisms

  • Lack of education about complex systems

  • Expectations of ‘causal’ explanations

Why would you create an agent-based model?

• ABM can provide a description of a real-world (or artificial system)

  • Simplified version → Make the model as simple as possible but not simpler

• Can explain the potential underlying phenomenon that control a system

  • Proof of concept of emergence (rules governing certain behavior)

• Can run repeated experiments varying conditions and parameters and observe changes

• Help us understand other systems with similar patterns of behavior (fish, birds, drones)

What are Netlogo Model properties?

• Agents and patches can all have properties that we can inspect

• We can change those properties over iterations (ticks)

• We can set up conditional model behavior based on these

What are the 4 things considered an agent?

• turtles

• patches

• links

• observer

What are the two key procedures required by Netlogo models?

• to setup

• to go

Which of the following is NOT an example of when ABM is a useful methodological tool for cog sci?

When there is homogeneity within and between groups

What are cognitive methods and models?

• often developed based on cognition in isolation

  • lab experiments, surveys

• analytic models

Describe how human cognition occurs in iteractions

• others influence our attention, learning, decision making…

• social environments are dynamic

  • interaction can create hard-to-predict feedback loops

• do individual cognitive models “scale“?

What is the methodological challenge of cognitive phenomena?

External actions may directly influence cognitive components and the behavior of cognitive entities is not trivially scaled to a population

What are 3 advantages of ABMS for CogSci?

• explore what happens when multiple cognitive entities interact over time and space

• can calibrate (parameterize) and validate (test predictions) cognitive models in complex systems

• encourage model development

When is an ABM a useful methodological tool?

• when cognitive function is dependent on the actions of other agents/environments

  • contingent behaviors (where feedback influences future states)

    • ex. voting behavior that has micro (individual) and macro outcomes

• when there is heterogeneity within and between groups

  • defining groups and setting properties/parameters

    • ex. financial market with regulators and investors

• can be exogenous (starting with different params) vs endogenous (start with same params but diverge over time)

• when time is a factor (change over time)

  • emergence of group dynamics

• when there is spatial distribution

  • ex. navigation, search, foraging…

What are the uses of ABMs in CogSci?

• in CogSci we want to understand how humans acquire knowledge, make decisions, and solve real-world problems

• calibration: provides parameterization

  • compared to known data to set parameters that best replicate and reproduce those data

• validation: tests predictions

  • parameters set theoretically, empirically, or via calibration

• longitudinal questions

  • evolution of cognition, culture, social norms…

• ethical questions

  • scenarios that are not “ethical” to manipulate in real-life (ex. violence, spread of misinformation…)

What’s a more complex definition of agents?

autonomous systems that operate transitions between states of the world, based on mechanisms and representations incorporated into them

• vary in the degree of autonomy, self-interest, sociability, learning, complexity

How are mental representations incorporated (if at all)?

• utilized to reason about the world and other agents, plan, make decisions, communicate

• symbolic (explicitly modeled “in“ an agent) vs sub-symbolic (more implicit associations/relationships)

What is multi-realizability?

• multi-level multi-agent systems can be implemented in different ways at lower levels and still generate the same or imilar macro-level phenomena

• with varying implementations, equivalence testing is possible for structure and mechanisms of the models

What are 2 current opportunities for ABMs?

• how to scale and incorporate real-time simulation with massive amounts of data

  • parallel and computing infrastructure

• model equivalence

  • have to accept multiple paths

  • better to have strong theoretical foundations and real-world plausibility

What is an ABM recipe for model building?

minimality procedure: simplest set of rules required to generate macroscopic effects

• seems to be consensus in model building but it has shortcomings (reduces validity) as it isn’t driven by theory and can create non-plausible models

Describe cognitive models

difficult to control the inner validity and calibration so it reflects the real-world system without also complex real-world data collection

What are generative models?

• generate theory of behavior by accounting for behavior in terms of mechanisms that are supposed to operate while producing it

• requires:

  • external (environmental and social)

  • internal (behavior/cognitive) mechanisms

• aims at finding the general mechanisms yielding the wide spectrum of behaviors of relatively autonomous systems

What are 3 variants of computational social science?

• deductive: explain social science phenomena with math, computer science, and logic/game theoretic models

• generative: ABMs

• complex: combine with complex systems methods, learning, data science, machine learning…

How is this an interdisciplinary foundation for CSS?

• describe the dynamics of a given phenomena using simulated and large datasets

• use ABM to check the internal consistency and resulting states

• apply cross-methodological experimental methods to validate hypotheses against real-world data

• update data-mining methods and models

• use equation-based modeling when possible