Exploring the Ethics of AI

Summary

The webinar explains how modern AI differs from science-fiction portrayals. AI is a broad collection of techniques that enable computers to perform tasks associated with human thought. Much of today’s AI relies on machine learning: systems learn patterns from data rather than following only explicitly written, step-by-step rules.

Unlike people, many deployed machine-learning models do not continue learning after they have been trained. Changing their behavior typically requires further training.

Where AI Is Used

AI already influences everyday services, often behind the scenes. Examples include video and product recommendations, spam filtering, social-network suggestions, and game characters. AI is also used in areas such as healthcare, education, and government decision-making.

Because these systems affect people’s lives, understanding how they work can help people make informed choices and have a say in how they are used.

Potential Benefits

Panelists describe applications intended to address real-world problems:

• Wildlife monitoring: Identifying animals in images, combined with reports from citizen scientists, can help track species.
• Crop health: Computer vision can help farmers identify diseased cassava plants.
• Aid planning: Analysis of satellite images can help estimate development and infrastructure needs, informing where resources may be useful.
• Science and communication: Protein-structure prediction may support future research, while improved machine translation can help people access information and communicate across languages.

Risks and Ethical Principles

The panel emphasizes that AI can magnify existing social patterns, including harmful bias. Large or poorly understood datasets can make it difficult to determine why a model behaves as it does, and the effects of deployment may not become clear until later.

Risks also depend on how a technology is used. For example, facial recognition for personal device access is different from using it for mass surveillance or persecution.

The speakers identify several principles for responsible AI:

• Patience and caution: Introduce systems gradually, evaluate them, and avoid rushing into broad deployment.
• Fairness: Check whether a system works differently across groups or distributes opportunities unevenly.
• Transparency: People affected by consequential systems should be able to understand why those systems behave as they do.
• Careful choices about purpose and impact: Consider who a system is built for, who may be harmed, and whether its effects—such as job displacement—are being addressed.

Responsible AI work is not limited to computer scientists. Social scientists, ethicists, philosophers, and people from many other fields can contribute to designing, evaluating, and governing AI systems.

Learning Resources and Activities

The speakers recommend learning through hands-on projects and accessible introductory material:

• Kaggle competitions and datasets: Explore a problem and build relevant skills through a project.
• Fast.ai, Coursera, and edX: Take online courses to learn AI concepts and terminology.
• Code.org AI resources: Use videos and lesson plans, including a lesson on AI ethics.
• Code.org “AI for Oceans” tutorial: Explore an activity connecting AI and social good.
• Minecraft Education challenges: Learn AI concepts through problems such as climate change.
• Critical news coverage: Consider who is affected when AI succeeds or fails, and whether a reported problem resembles earlier challenges.

Main Speakers and Sources

• Pat Yongpradit — Code.org Chief Academic Officer and webinar host
• Amanda Askell — OpenAI research scientist on the policy team
• Deb Raji — Algorithmic Justice League fellow and algorithm auditor
• Maron Sahami — Stanford University professor
• Natasha Crampton — Microsoft Chief Responsible AI Officer