A Discrete Formulation of Two Alternative Forced Choice Decision Dynamics Derived from a Double-Well Quantum Landscape

Morgan Rosendahl, Jonathan D. CohenbioRxiv 2021 · DOI

TL;DR

This paper introduces the Multi-Particle and Multi-Well (MPMW) quantum cognitive framework, a novel approach to modeling decision-making that leverages quantum mechanics. The MPMW offers an analytically tractable, discrete-time model that naturally incorporates noise and can be extended to more complex scenarios than existing models.

What Problem Does This Paper Address?

Traditional models of decision-making, like the Drift Diffusion Model (DDM), often treat noise as an external parameter that needs to be fitted. This paper addresses the challenge of modeling decision-making in a way that integrates noise as an emergent property of the system, influenced by factors like stimulus salience and cognitive control. Furthermore, existing quantum models for decision-making, while offering insights into phenomena that violate classical probability laws, often rely on single-particle dynamics. This work aims to bridge the gap by developing a multi-particle quantum framework that is more amenable to analytical solutions and can capture a richer set of cognitive phenomena.

Methods

The core of the paper is the introduction and analysis of the Multi-Particle and Multi-Well (MPMW) quantum cognitive framework. This model conceptualizes decision-making as a process where multiple independent quantum particles, representing stimulus information, are emitted into a "double-well quantum landscape." This landscape, defined by attractor wells, represents the decision alternatives. The dynamics of the decision process are modeled as a discrete-time Markov chain, where the parameters of the wells (width and depth) reflect bottom-up (automaticity/salience) and top-down (control/attention) influences. A key innovation is that stochasticity (noise) arises directly from the quantum landscape itself, rather than being an externally fitted parameter. The paper also explores both single-integrator and multi-integrator versions of the MPMW model.

Datasets

Not discussed in the paper. The paper focuses on theoretical modeling and analysis rather than empirical data fitting.

Key Insights and Ideas

  • Noise as an Emergent Property: The MPMW model elegantly incorporates noise as an inherent feature of the quantum landscape, directly linked to stimulus properties and cognitive control, rather than an arbitrary external parameter.
  • Analytical Tractability: By formulating the model as a discrete-time Markov chain, the MPMW allows for analytical solutions for response time distributions and probabilities, making it easier to study and compare with empirical data.
  • Bridging Classical and Quantum Models: The MPMW provides a framework that mirrors key properties of classical models like the DDM while incorporating quantum phenomena, offering a unified approach to understanding decision-making.
  • Interpretable Parameters: The parameters of the MPMW model (well width and depth) have direct cognitive interpretations related to stimulus salience, automaticity, and attentional control.
  • Oscillations as a Quantum Signature: The model predicts period-2 oscillations in response time distributions and probability of success, which are a consequence of quantum interference effects. While these oscillations may be difficult to observe empirically at biologically realistic scales, they highlight the unique nature of the quantum approach.

Key Results

  • The MPMW model, in its single-integrator form, converges to the behavior of a traditional DDM in the limits of long time or large numbers of integrators.
  • The model produces response time distributions that follow inverse Gaussian curves, consistent with empirical data and previous DDM models.
  • The MPMW predicts that mean response times are faster for incorrect than for correct responses, a common empirical finding.
  • The paper derives closed-form solutions for probabilities of success and response time distributions under both interrogation and free-response paradigms.
  • For multi-integrator systems, the probability of success exhibits period-2 oscillations bound within a collapsing envelope, which decay with increasing numbers of integrators.

Limitations and Future Work

  • Observability of Oscillations: The paper acknowledges that the predicted oscillations in response time distributions and probability of success may not be empirically measurable at biologically realistic numbers of integrators, as they are heavily damped.
  • Extension to N-Alternatives: While the MPMW framework is designed for N-alternative decision-making, this paper focuses primarily on the 2AFC case. Future work will explore the multi-alternative, multi-well formulation.
  • Variable Drift Rates and Landscapes: The current model assumes static landscapes. Future research will investigate the effects of variable drift rates and dynamic landscapes, which could incorporate more complex cognitive phenomena like reallocation of control.
  • Modeling Arousal: The paper suggests that eigenenergies can be used to model arousal, and further work could explore this connection more deeply to provide a unified account of automaticity, control, and arousal.

Summary updated

Abstract

Abstract Tools from quantum theory have been effectively leveraged in modeling otherwise poorly understood effects in decision-making such as apparent fallacies in probability judgments and context effects. This approach has described the dynamics of two alternative forced choice (2AFC) decisions in terms of the path of a single quantum particle evolving in a single potential well. Here, we present a variant on that approach, which we name the Multi-Particle and Multi-Well (MPMW) quantum cognitive framework, in which decisions among N alternatives are treated by the sum of positional measurements of many independent quantum particles representing stimulus information, acted on by an N-well landscape that defines the decision alternatives. In this article, we apply the MPMW model to the simplest and most common case of N-alternative decision making, 2AFC dynamics. This application calls for a multi-particle double-well implementation, which allows us to construct a simple, analytically tractable discrete drift diffusion model (DDM), in the form of a Markov chain, wherein the parameters of the attractor wells reflect bottom-up (automatic) and top-down (control-dependent) influences on the integration of external information. We first analyze this Markov chain in its simplest form, as a single integrator with a generative process arising from a static quantum landscape and fixed thresholds, and then consider the case of multi-integrator processing under the same conditions. Within this system, stochasticity arises directly from the double-well quantum attractor landscape as a function of the dimensions of its wells, rather than as an external parameter requiring independent fitting. The simplicity of the Markov chain component of this model allows for easy analytical computation of closed forms for response time distributions and response probabilities that match qualitative properties of the accuracies and reaction times of humans performing 2AFC tasks. The MPMW framework produces response time distributions following inverse gaussian curves familiar from previous DDM models and empirical data, including the common observation that mean response times are faster for incorrect than for correct responses. The work presented in this paper serves as a proof of concept, based on which the MPMW framework can be extended to address more complex decision-making processes, (e.g., N-alternative forced choice, dynamic control allocation, and nesting quantum landscapes to allow for modeling at both the task and stimulus levels of processing) that we discuss as future directions.
BibTeX
@misc{rosendahl2021discrete,
  author = {Morgan Rosendahl and Jonathan D. Cohen},
  title = {{A Discrete Formulation of Two Alternative Forced Choice Decision Dynamics Derived from a Double-Well Quantum Landscape}},
  year = {2021},
  doi = {10.1101/2021.09.23.461524},
  howpublished = {bioRxiv},
}

Loading related content...

Figure