Overview
This lecture discusses the immediate, real-world impacts of artificial intelligence (AI) on society and the environment, emphasizing practical tools and approaches for measuring and addressing these challenges.
Current Headlines and Public Concerns
- AI is widely discussed in media for both positive innovations and negative incidents.
- There is public fear about AI causing existential threats, but its present impacts are more urgent.
Environmental Impact of AI
- Training large AI models requires significant energy, contributing to climate change.
- Example: Training the Bloom model used as much energy as 30 homes annually and emitted 25 tons of carbon dioxide.
- Larger AI models have grown 2,000 times in size in five years, greatly increasing environmental costs.
- Current lack of transparency from tech companies about AI's environmental impact.
- Tools like CodeCarbon help track AI energy use and carbon emissions and encourage more sustainable choices.
Copyright and Consent Issues in AI Training Data
- AI models are often trained on data (art, books) without creators’ consent.
- Tools like “Have I Been Trained?” allow creators to check if their works are in AI datasets.
- Artists have used such tools to sue companies for copyright infringement.
- New opt-in/opt-out mechanisms for data inclusion have been created to protect creators.
AI Bias and Societal Impact
- AI can encode and perpetuate harmful stereotypes and discrimination (bias).
- Facial recognition systems have misidentified people, leading to wrongful accusations and imprisonments.
- Image generation models often reinforce stereotypes about professions and race.
- The Stable Bias Explorer tool reveals these biases across various professions.
- Understanding AI systems is crucial as they are deployed in essential societal domains.
Solutions and Moving Forward
- There is no single fix for AI’s complex problems (bias, copyright, environment).
- Measuring and disclosing AI impacts enables better decisions by companies, lawmakers, and users.
- Developing accessible tools helps inform policy, regulation, and public understanding.
- Immediate action is needed to address current impacts rather than only focusing on future risks.
Key Terms & Definitions
- AI (Artificial Intelligence) — Computer systems performing tasks that usually require human intelligence.
- Large Language Model — AI trained on vast text data to generate language and answer questions.
- Bias — Systematic favoritism or prejudice embedded in AI outputs.
- Sustainability — Meeting current needs without harming the environment or depleting resources.
- Copyright Infringement — Unauthorized use of someone’s intellectual property.
Action Items / Next Steps
- Use tools like CodeCarbon and “Have I Been Trained?” to assess AI impacts.
- Explore opt-in/opt-out options for creators regarding AI training datasets.
- Stay informed about AI’s environmental, social, and ethical effects.
- Advocate for transparency and regulation in AI development and deployment.