The Project
In the late 1990s, Arizona contemplated a rule change for jurors that would allow them to deliberate before all the evidence of a trial was presented, the intent being to reach verdicts quicker and make jury duty more interesting. This project was a simulation of the jury decision-making process to investigate the possible effects on verdict outcomes.
My Role
Principle Investigator and ML Engineer, working as an undergraduate under the mentorship of Edwin Hutchins, PhD, the leading researcher in the field of distributed cognition.
Interesting Challenges
The primary challenge for any simulation is selecting a suitable modeling paradigm. For this project, the system being simulated was the psychological biases and persuasive susceptibility of a group of human actors. At the level of individual actors, a well-known phenomenon called 'confirmation bias' is known to operate on an individual’s perception of real-world evidence. At the level of a group of actors, a hypothetical phenomenon coined 'group confirmation bias' was presumed to operate, and was the intended focus of the simulation.
In order to simulate individual confirmation bias, a programming paradigm called 'constraint-satisfaction network' was employed. Network nodes arranged into a small communicating group are allowed to influence each other, both positively and negatively, until the activation level of the nodes settles to an equilibrium. For this simulation, the initial activations modeled the a priori biases of an individual actor, and external signals fed into the network simulated trial evidence.
To simulate deliberation among individuals, a collection of constraint satisfaction networks was connected in a mesh network to allow each network’s current activation level to affect the others. For this simulation, the timing of inter-network communication was varied as either early-onset while evidence was being fed to the individual networks, or late-onset following the evidence feed.
Results
The simulation results indicated a statistically significant main effect of increased likelihoods of guilty verdicts when only the timing of jury deliberation onset was varied, while holding the sequence and content of trial evidence constant.
This project was awarded a spot in the 2000 Undergraduate Research Conference, where I presented the research methodology and results to a peer group of other similarly-awarded undergraduates and their mentors.
My mentor subsequently used this project and its results as an exemplar for research projects in his distributed cognition curriculum into the early 2010's.
The Project
In the late 1990s, Arizona contemplated a rule change for jurors that would allow them to deliberate before all the evidence of a trial was presented, the intent being to reach verdicts quicker and make jury duty more interesting. This project was a simulation of the jury decision-making process to investigate the possible effects on verdict outcomes.
My Role
Principle Investigator and ML Engineer, working as an undergraduate under the mentorship of Edwin Hutchins, PhD, the leading researcher in the field of distributed cognition.
Interesting Challenges
The primary challenge for any simulation is selecting a suitable modeling paradigm. For this project, the system being simulated was the psychological biases and persuasive susceptibility of a group of human actors. At the level of individual actors, a well-known phenomenon called 'confirmation bias' is known to operate on an individual’s perception of real-world evidence. At the level of a group of actors, a hypothetical phenomenon coined 'group confirmation bias' was presumed to operate, and was the intended focus of the simulation.
In order to simulate individual confirmation bias, a programming paradigm called 'constraint-satisfaction network' was employed. Network nodes arranged into a small communicating group are allowed to influence each other, both positively and negatively, until the activation level of the nodes settles to an equilibrium. For this simulation, the initial activations modeled the a priori biases of an individual actor, and external signals fed into the network simulated trial evidence.
To simulate deliberation among individuals, a collection of constraint satisfaction networks was connected in a mesh network to allow each network’s current activation level to affect the others. For this simulation, the timing of inter-network communication was varied as either early-onset while evidence was being fed to the individual networks, or late-onset following the evidence feed.
Results
The simulation results indicated a statistically significant main effect of increased likelihoods of guilty verdicts when only the timing of jury deliberation onset was varied, while holding the sequence and content of trial evidence constant.
This project was awarded a spot in the 2000 Undergraduate Research Conference, where I presented the research methodology and results to a peer group of other similarly-awarded undergraduates and their mentors.
My mentor subsequently used this project and its results as an exemplar for research projects in his distributed cognition curriculum into the early 2010's.
Skills
Demos / Artifacts
Slide Presentation
The slides presented at the 2000 Undergraduate Research Conference.
View slides