AI and Its Societal Impact

Introduction to AI Researcher

• Over a decade of experience in AI research.
• Received an alarming email about AI's potential to end humanity.

AI in the Headlines

• Positive Headlines:
  • AI discovering new molecules for medicine.
  • Public interest in AI (e.g., viral images of the Pope in a puffer coat).
• Negative Headlines:
  • Chatbot advising divorce.
  • AI meal planner suggesting dangerous recipes.

Current Concerns in AI

• Doomsday Scenarios:
  • Existential risks and the singularity.
  • Ongoing discussions and events focused on preventing these outcomes.
• AI's Real Impacts:
  • Contributes to climate change.
  • Uses copyrighted material without consent.
  • Discriminates against communities.

The Need for Transparency in AI

• Importance of tracking AI's impacts on society and the environment.
• Aim for future AI models to be more trustworthy and sustainable.

Sustainability Issues

• Environmental Costs of AI Models:
  • Cloud infrastructure involves significant physical resources (metal, plastic).
  • Each AI query has a planetary cost.
  • Example: Training Bloom model used energy equivalent to 30 homes for a year; emitted 25 tons of CO2.
• Comparison to Other Models:
  • GPT-3 emits 20 times more carbon than Bloom model.
  • Larger models have drastically higher environmental costs.

Addressing Environmental Impact

• CodeCarbon Tool:
  • Estimates energy consumption and carbon emissions of AI training.
  • Helps in making sustainable choices for AI deployment.
  • Suggested using renewable energy to power AI models.

Copyright Concerns in AI

• Difficulty for artists to prove unauthorized use of their work.
• Have I Been Trained Tool:
  • Allows individuals to search datasets to find unauthorized use of their images.
  • Example of a lawsuit by artists against AI companies for copyright infringement.

Bias in AI Models

• Definition and Risks:
  • AI can encode stereotypes and biases.
  • Example: Facial recognition failures for women of color.
  • Consequences of bias include wrongful accusations and imprisonment.

Case Studies of Bias

• Portia Woodruff Incident:
  • Wrongly accused of carjacking due to AI misidentification.
• Stable Bias Explorer Tool:
  • Explores biases in image generation models across professions.
  • Highlights representation issues and stereotypes in AI outputs.

Legislation and Governance

• Importance of creating tools for AI transparency and understanding.
• Need for informed choices regarding AI deployment.
• Collaboration between researchers, legislators, and users to develop effective AI governance.

Final Thoughts

• Addressing tangible impacts of AI is crucial.
• AI's future is not predetermined; society can choose its trajectory.
• Collective decision-making is essential in shaping AI's role in the future.