Lecture Notes on AI for Good

Speaker: Jeffrey Hammerbacher

• Early data scientist at Facebook and Cloudera.
• Noted that the best minds are focused on how to get people to click on ads, which is concerning given global challenges.

Global Challenges:

• Millions of children die before age five.
• Climate change affects hundreds of millions.
• 1.6 billion people live with severe disabilities.

AI's Role in Addressing Challenges:

• Organizations and non-profits often lack AI talent to tackle these issues.
• Majority of AI talent is in tech and finance.
• AI can be used in both societal issues and commercial ones; similar analytical approaches apply.

Lessons Learned in AI for Good Space

Lesson 1: AI as a Necessary Solution

• Example: Diabetic Retinopathy (DR) leads to blindness; 450 million people suffer from diabetes.
  • Only 200,000 ophthalmologists globally.
  • AI models can diagnose DR effectively and are already in use.

Lesson 2: Importance of Subject Matter Experts

• Example: Low birth weight prediction shows AI can't determine causality.
  • AI identifies smoking as a factor, but smoking is a cause of low birth weight.
• Partnerships with experts are essential for meaningful outcomes.

Lesson 3: Simplicity Over Complexity

• Humans are drawn to complex projects.
• Simple solutions can have greater impact.
  • Example: SEEDS organization uses simple data collection methods via satellite to address climate change impacts.

Lesson 4: Awareness of Bias in AI

• Left-Handed Dilemma: Misinterpretation in data can lead to false conclusions.
  • Left-handed individuals were found to die younger due to bias in data collection regarding left-handedness.
• Importance of recognizing and mitigating biases in AI datasets.

Examples of Applying AI for Good

• Turkey/Syria Earthquake Response: Collaborated with Planet Labs for mapping affected areas.
• Amazon Deforestation Monitoring: AI models used to assess biodiversity and illegal logging.
• Retinopathy of Prematurity (ROP): AI can diagnose conditions in premature infants, addressing a lack of specialists.
• Seeing AI: App designed for the visually impaired to navigate the world.

Historical Context of Technology and Its Impact

• Discussion on the historical context of infant mortality and the role of electricity.
• War of Currents: ACDC vs DC electricity, highlighting the importance of safe technology.

Potential of Large Language Models (LLMs)

• LLMs democratize access to knowledge and coding capabilities.
  • Non-native English speakers can express ideas fluently.
  • Programming in native languages is now possible.
• Study from JAMA: LLMs provided more accurate and empathetic responses than doctors.

Ethical Considerations of AI

• Future considerations on when it may be unethical not to use AI in healthcare and other critical areas.

Conclusion

• Emphasized the transformative potential of AI and the necessity for responsible application.