Abstract
AI and Responsibility in Financial Decision Making
This case study examines how the use of AI affects financial decision-making and the
fragmentation of accountability that comes with it. It begins with the Flash Crash of May 2010,
which resulted in the disappearance of nearly one trillion dollars of equity on American stock
markets in mere minutes due to the automated response of trading algorithms. From there, the
research follows the expansion of AI in finance into credit scoring, fraud detection, insurance
underwriting, and risk management, from determining whether a person qualifies for a loan or
how their premium is calculated, to the trillions of dollars exchanged daily without any single
party bearing responsibility when something goes wrong. The framework for this case study is
the Actor-Network Theory associated with Cressman, Latour, and Winner, and particularly the
idea of the black box. By reducing a sociotechnical system to a single object, all its assumptions
remain invisible to the user and the other parties involved. To apply this notion to the financial
use of AI, four groups of actors are analyzed: model engineers, institutions, regulators, and
consumers, who ultimately bear the consequences of the process. All of them operate with
incomplete knowledge and incomplete control, and the choices available narrow progressively as
one moves down the chain. The main opposing viewpoints argue that AI removes bias, that
improved efficiency is always good, and that excessive regulation stifles innovation. In reality,
AI only perpetuates the biases already present in historical data; improved efficiency means little
when there is no way to take action if something goes wrong; and regulation stops being neutral
once applied to opaque processes. The paper concludes by arguing for transparency during the
design phase, human oversight during deployment, and meaningful recourse for those harmed by
automated decisions.
Using Student Projects to Compare Model-Based Systems Engineering and
Human-Centered Design
This paper compares MBSE with HCD based on the experience gathered from two senior
capstone projects and presented at the IEEE SIEDS Conference. The basis of comparison in this
work was provided through the application of both MBSE and HCD in senior capstone projects
by the same team. For the fall semester, MBSE was used for creating a resource management
system for an industrial makerspace, whereas the focus of the spring semester changed to
designing the booking system of the architecture makerspace at UVA through HCD, due to the
increasing demand for 3D printers in the space. The implementation of MBSE included
following a very tight weekly schedule to create SysML deliverables such as block definition
diagrams, internal block diagrams for the identification of runtime data flow, activity diagrams
for user workflow, and eventually a complete system architecture using Papyrus. On the other
hand, the implementation of HCD involved performing contextual interviews and observations,
developing wireframes, conducting test sessions with actual end-users, and creating prototypes
of different levels of fidelity until a website was developed. Progress made in HCD was
measured by insights gained from each iteration rather than completion of diagrams. Debriefing
interviews with an industry MBSE practitioner as well as two academics with expertise in HCD
have been used in this paper's analysis. The analysis shows that MBSE is more effective for
large, complex, safety-critical systems with long timelines, while HCD is better suited to smaller
projects with accessible users and changing requirements, framing the two as complementary
rather than competing approaches.
