Lecture by Mazarine Banagi on AI, Implicit Bias, and Decision Making

Introduction

• Speaker: Mazarine Banagi
• Affiliation: Harvard University, Department of Psychology
• Field of Study: Experimental psychology focusing on unconscious processes in decision making

Key Topics of Discussion

AI in Modern Times

• AI is embedded in various aspects of life.
• Importance of algorithmic decision-making in research and health.
• Example: Algorithms detecting tumors better than radiologists.
• Concerns about untested AI applications, like job interview analysis via facial movements.

Problems and Risks of AI

• **Transparency Issues: **

  • Complexity of algorithms can obscure functionality, even from creators.

• **Corporate Influence: **

  • Risks if AI development is driven by financial interests alone.

Positive Aspects and Addressing Bias in AI

• Growing awareness within the computer science community about biases in AI.
• Emergence of institutes focused on AI ethics and bias reduction.
• Importance of avoiding corporate-driven outcomes in AI development.

Research and Findings on Language and Implicit Bias

Analysis of Large Language Corpora

• Access to digitalized Google Books from the 1800s to present.
• Exploring historical attitudes and implicit sentiments through language.

Findings on Emotional Content in Words

• Some stability in emotional valence (positive/negative) of words over 200 years.
• Internet data analysis (Common Crawl) revealing entrenched stereotypes and biases.

Implicit vs Explicit Bias

• Implicit bias affects decision-making unconsciously, differently from explicit bias.

Importance of Understanding Implicit Bias

• Improved decision-making and alignment of behavior with personal values by becoming aware of implicit biases.

Application in Business and Alternative Measures

Challenges with Traditional Surveys

• Limitations of survey data in capturing complex and subjective evaluations.
• Surveys still valuable but need supplementing with implicit measures for comprehensive insights.

Future Directions and Regulations

Deep Fakes and Ethical AI

• Increasing presence of deep fakes in media, especially in political advertising.
• Legislative measures (e.g., New Mexico law) mandating disclosure of deep fakes in political advertisements.

Balancing Regulation and Innovation

• Need for basic transparency, accuracy, fairness, and accountability in AI systems.
• Balancing regulation with innovation to ensure ethical AI development.