Lecture: Conversation with Oren Etzioni on AI and the Allen Institute

Introduction

• Speaker: Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence (AI2)
• Interviewer: John (unknown last name)

Background on AI2

• Founded about 9 years ago with vision and resources from Paul Allen
• Paul Allen's vision: create an institute dedicated to AI research and development
• Oren was initially encouraged to start the project as a tenure professor

Oren Etzioni's Journey

• Gave a notable talk/demonstration in San Francisco (2018-2019) on Natural Language Processing (NLP)
• Demonstrated live models performing tasks but also displayed weaknesses, reinforcing the idea that “this stuff barely works”
• Emphasized the importance of realistic expectations over hype

AI Technologies and Language Models

• Progress of language models (e.g., GPT-3, Lambda)
• Importance of understanding true capabilities and limitations
• Siri/Alexa as examples of both potential and limitations of current AI
• Example: Misinterpretation of historical figures (e.g., president in 1492)

Paul Allen's Influence

• Visionary and intellectual maverick
• Interest in fundamental questions about intelligence and AI
• Desired both scientific understanding and practical utility of AI
• Invested in both AI and brain science (Allen Institute for Brain Science)

Etzioni's AI Philosophy and AI2's Unique Position

• Bridging the gap between academic inquiry and practical technology development
• Universities: limited by student cycles and academic structure
• Companies: profit-driven constraints
• AI2: free to explore long-term, high-impact projects
• Research and engineering collaboration at AI2
• Major projects like Semantic Scholar

AI2's Scope and Accomplishments

• AI2 tackles projects companies won't do, and universities can't
• Examples of large-scale projects that require sustained, intensive resources
• Research and development in NLP, climate modeling, and more

Specific Projects

1. Semantic Scholar:

   • A free search engine for scientific content
   • 100 million users annually, 200 million papers in its corpus

2. Math and Science Reasoning:

   • AI systems solving math problems, taking educational tests
   • The infrastructure and resources required

3. Skylight - Anti-Illegal Fishing:

   • Computer vision and satellite data to fight illegal fishing
   • Recent competition win in analysis tools

4. Climate Modeling and Predictions:

   • Using deep learning for predicting weather changes, especially precipitation
   • Important for agriculture, infrastructure planning

5. Common Sense AI (Project Mosaic):

   • Building common sense into AI systems
   • Led by Yejin Choi at University of Washington/AI2
   • Addresses fundamental issues in AI like ethics and practical reasoning

Challenges and Ethical Considerations

• Importance of common sense in AI to avoid scenarios like overproduction without ethical consideration (e.g., paperclip maximizer)
• Balancing data-driven approaches with symbolic AI
• Navigating controversial and nuanced issues in ethical AI

AI2 Incubator

• Supports the creation of startups
• Over 20 companies, about $750 million in total valuation
• Examples of successes: Xnor (acquired by Apple)

Future of AI and Common Sense

• Collection of ethical norms and common sense knowledge
• Projects such as Delphi for ethical decision-making
• Addressing ambiguities and biases in human ethical reasoning within AI systems

Final Thoughts

• Recognition of rapid progress in AI but caution against overhyping capabilities
• Emphasis on realistic assessments and continual inquiry into AI’s role and impact on society

Conclusion

• Oren Etzioni emphasizes the blend of visionary ambition with realistic, grounded assessments of AI
• AI2 continues to drive meaningful, impactful research while navigating the ethical and practical complexities of AI development