Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Human Misperception of Generative-AI Alignment: A Laboratory Experiment

Kevin He, Ran Shorrer, Mengjia Xia

Abstract

We conduct an incentivized laboratory experiment to study people's perception of generative artificial intelligence (GenAI) alignment in the context of economic decision-making. Using a panel of economic problems spanning the domains of risk, time preference, social preference, and strategic interactions, we ask human subjects to make choices for themselves and to predict the choices made by GenAI on behalf of a human user. We find that people overestimate the degree of alignment between GenAI and human choices. In every problem, human subjects' average prediction about GenAI's choice is substantially closer to the average human-subject choice than it is to the GenAI choice. At the individual level, different subjects' predictions about GenAI's choice in a given problem are highly correlated with their own choices in the same problem. We explore the implications of people overestimating GenAI alignment in a simple theoretical model.

Human Misperception of Generative-AI Alignment: A Laboratory Experiment

Abstract

We conduct an incentivized laboratory experiment to study people's perception of generative artificial intelligence (GenAI) alignment in the context of economic decision-making. Using a panel of economic problems spanning the domains of risk, time preference, social preference, and strategic interactions, we ask human subjects to make choices for themselves and to predict the choices made by GenAI on behalf of a human user. We find that people overestimate the degree of alignment between GenAI and human choices. In every problem, human subjects' average prediction about GenAI's choice is substantially closer to the average human-subject choice than it is to the GenAI choice. At the individual level, different subjects' predictions about GenAI's choice in a given problem are highly correlated with their own choices in the same problem. We explore the implications of people overestimating GenAI alignment in a simple theoretical model.

Paper Structure

This paper contains 59 sections, 4 theorems, 17 equations, 8 figures, 33 tables.

Table of Contents

  1. Introduction
  2. Related Literature
  3. Theoretical Framework
  4. A Model of Anthropomorphic Projection and Self-Projection
  5. Correctly specified beliefs.
  6. No anthropomorphic projection, some self-projection.
  7. No self-projection, some anthropomorphic projection.
  8. Weighted average of self and GenAI.
  9. Application: Delegation under Misperceived Alignment
  10. When Do Agents Delegate?
  11. Implications of Anthropomorphic Projection
  12. Implications of Self-Projection
  13. Experimental Design and Deployment
  14. Overview
  15. The Decision Problems
  16. ...and 44 more sections

Key Result

Proposition 1

When $\mathbb{E}[b(\omega)]=0$, the agent strictly prefers to delegate to GenAI if and only if $\blacktriangleleft$$\blacktriangleleft$

Figures (8)

  • Figure 1: Experience and attitudes toward GenAI. (a) Distribution of responses to the question: "In a typical week, on how many days do you use generative AI tools?" (b) Percentage of subjects who have used various GenAI models before. (c) Degree of agreement with the statement: "Decisions made by AI are on average similar to decisions made by humans." (d) Degree of agreement with the statement: "On average, AI makes better decisions than humans."
  • Figure A.1: Heterogeneity: Experience with GenAI
  • Figure A.2: Heterogeneity: Attitudes Toward GenAI
  • Figure A.3: Launch Page
  • Figure A.4: Part 1 Instructions
  • ...and 3 more figures

Theorems & Definitions (8)

  • Proposition 1
  • Proposition 2
  • Example 1
  • Proposition 3
  • Proposition 4
  • proof
  • proof
  • proof