SBIR: Predictive Psychological Architectures for Decision-Making (PPADM)

OUSD (R&E) critical technology area(s): Human-Machine Interfaces, Trusted AI and Autonomy

Objective: This effort aims to understand the association between decision-maker characteristics and behavior to improve human and algorithmic decision-making, delegation, and trust. Research will focus on decision-making in difficult domains, where trusted decision-makers disagree; no right answer exists; and uncertainty, time-pressure, resource limitations, and conflicting values create significant decision-making challenges.

Description: Expanded use of artificial intelligence technology within the Department of Defense makes it increasingly important to examine the ability of algorithmic decision-makers (ADMs) and human decision-makers (HDMs) to operate in difficult domains. Such difficult decisions may violate fundamental principles of current decision-making models that center on rules, rationality and the recognition and application of prior patterns. A principal need in the difficult decision-making domain is to understand the human architecture that predicts how individuals will make difficult decisions, and the factors at the individual level (referred to as ‘Key Decision-Maker Attributes’ or KDMAs) that predict meaningful variation in how individuals navigate the decision-making space and what choice they ultimately make. These attributes are indicative of a decision-maker’s outcome preferences and/or their decision-making process. Examples include risk-seeking vs. risk-aversion1 or maximizing vs satisficing2. Decision-maker models must provide an architecture that is resilient to new domains, while accommodating varying degrees of intra and inter-individual consistency.

The intent of this topic is to:

• Investigate the psychological structure of Key Decision-Maker Attributes and the impact of KDMAs on decision-making and trust, such as the willingness to delegate difficult decisions to an ADM or HDM
• Predict meaningful variation of individuals’ KDMAs in the decision-making space and the impact on what choice decision-makers ultimately make
• Determine KDMA drivers underlying the willingness to delegate for both human-to-human decision-maker delegation and human-to-ADM decision-maker delegation
• Validate the scientific hypothesis that it is possible to meaningfully link the decisions that someone decides (and the process through which they make those decisions) to KDMAs and that KDMAs are identifiable and measurable.

This effort seeks to leverage research, technical development, and policy centered on the expansion of “human off the loop” algorithms in decision-making domains where decision-makers do not agree. As an example, see DARPA’s In the Moment3 program which seeks to enable measuring and building algorithmic decision-makers for mission-critical DoD operations such as battlefield triage. A better understanding of the psychological structure of KDMAs will enable the creation of automated systems that can more effectively work with, for, or in place of human decision-makers. Understanding the association between decision-maker characteristics and decision-making behavior is essential for bridging the gap between the current state of the art and the future of human/algorithmic decision-making, delegation, and trust.

Phase I

This topic is soliciting Direct to Phase II (DP2) proposals only. Prior research should have been conducted in the past five years. Reports which provide data, clearly present the analysis done and provide evidence of scholarly impact will be strongly preferred. Research which has been done on military decision-makers, especially those involved in making difficult decisions like those faced in combat, mass casualty, and triage events, will be strongly preferred. Proposers should already have a defensible framework and methodology to apply, and test based on their previous research.

Phase II

Provide an architecture capable of modeling KDMAs for human decision-makers. The models should be able to use KDMAs to predict meaningful variation in how individuals navigate the decision-making space and what choice decision-makers ultimately make. Validate the scientific hypothesis that it is possible to meaningfully link the decisions that someone decides (and the process through which they make those decisions) to KDMAs and that KDMAs are identifiable and measurable. Determine KDMA drivers for delegation for both human-to-human decision-maker delegation and human-to-ADM delegation.

Phase III dual use applications

Understanding the KDMAs associated with decision-making styles will enable development of systems that are in greater alignment with human decision-makers. This alignment will support a better understanding of clinical decision-makers, especially when they are functioning under resource-constrained circumstances where there is no right answer. The KDMA framework developed in this SBIR will have applications for future ADM applications, such as military medical triage, where human decision-makers are often overwhelmed by the operational circumstances but lack a basis for making the decision to delegate to available technology.

This SBIR DP2 effort will produce research that supports DARPA's In the Moment program, particularly the effort to quantify the alignment of algorithmic decision-maker KDMAs with trusted humans in difficult domains such as medical triage.

References

1. Prospect Theory: An Analysis of Decision under Risk. Kahneman, Daniel and Tversky, Amos. The Econometric Society, 1979, Vol. 47.

2. Police Perfection: Examining the Effect of Trait Maximization on Police Decision-Making. Shortland, Neil, Thompson, Lisa and Alison, Laurence. Frontiers in Psychology, 2020, Vol. 11.

3. DARPA’s In the Moment program: https://www.darpa.mil/program/in-the-moment

Keywords

Decision-making, Attributes, Delegation, Trust, Decision theory

TPOC-1

DARPA BAA Help Desk

Email

SBIR_BAA@darpa.mil