Peter Norvig (Google Director of Research) – Town Hall on A.I., Machine Learning, and More (Jan 2017)


Chapters

00:00:08 Town Hall on AI, Machine Learning, and More
00:04:27 AI in the Modern World: Accessibility, Challenges, and Teaching
00:12:00 Addressing Practicality, Progress, and the Resurgence of AI
00:16:34 Expert Advice for Deep Learning Practitioners
00:23:27 Evolving Programming Challenges in an Uncertain World
00:32:35 Ethical Considerations in the Design of Artificial Intelligence Systems
00:35:15 AI Security, Fairness, and Data Challenges: Google's Perspective
00:39:58 AI: Combining Symbolic and Neural Approaches
00:47:40 Hybrid Research for Serving Google Users
00:50:45 Conversational AI Challenges and User Interface Solutions
00:56:02 Innovations in Online Education
00:59:07 AI's Overlooked Applications in Education, Environment, Health

Abstract

Lifelong Learning and the Future of AI: Insights from Peter Norvig and Beyond

Introduction: Embracing Lifelong Learning in Computing

The ACM Learning Webinar, as part of ACM’s commitment to lifelong learning, hosted a Town Hall with Peter Norvig on AI, machine learning, and related topics. The event was moderated by Rosemary Paradis, Principal Research Engineer and Secretary-Treasurer of ACM SIG AI. ACM, with over 100,000 computing professionals and student members, offers a wealth of educational and professional development resources. These resources are available at atthelearning.acm.org and include various educational and professional development opportunities.

Webinar Essentials: Ensuring a Smooth Experience

To ensure a seamless webinar experience, organizers addressed potential technical issues and encouraged interactive participation. Attendees experiencing technical difficulties were guided to press F5 (Windows), Command R (Mac), refresh their browser (mobile), or close and relaunch the presentation. The webinar provided additional widgets and resources on the bottom panel and right sidebar. Questions were submitted through the Q&A box. The session was recorded and archived for future access, with attendees receiving email notifications upon availability.

Speaker Spotlight: Peter Norvig’s AI Odyssey

Peter Norvig, serving as the Director of Research at Google, was the featured speaker. His impressive background includes leading Google’s Core Search Algorithms Group and the NASA Ames Computational Sciences Division, as well as being NASA’s Senior Computer Scientist. He was honored with the NASA Exceptional Achievement Award in 2001. Norvig’s journey into AI began in high school, where he learned basic programming and developed an interest in natural language processing through a linguistics class. This dual interest in linguistics and computer science in college paved his path towards AI.

Teaching AI: Bridging the Old and the New

Peter Norvig emphasized the need for an AI curriculum that balances foundational AI principles with advanced techniques like deep learning. This approach is designed to help students understand the evolution of AI and the various methodologies’ strengths and weaknesses.

Textbook Challenges: Balancing Priorities

The delay in releasing the new edition of “Artificial Intelligence: A Modern Approach” reflected the dynamic and rapidly evolving nature of AI, incorporating emerging concerns like AI safety and ethics. The new edition of the book, co-authored by Norvig, is expected to be completed during his co-author’s sabbatical next year.

Geoffrey Hinton’s Insights: The Power of Computing in Deep Learning

Geoffrey Hinton’s observations underscored the significance of increased computing power in enabling neural networks to converge and produce effective results, a feat previously limited by resource constraints. AI’s progress, particularly in tasks like speech recognition, image recognition, and machine translation, is attributed to more data availability, increased computing power, and new techniques such as deep learning.

Evolving AI Research Methods

AI research, while continually changing, maintains certain constant elements like representation, uncertainty management, planning, reasoning, and probability. Norvig highlighted the importance of objective evaluation in research, setting clear milestones and metrics for progress, especially when communicating with senior researchers. He predicted that while the latest AI methods might not last forever, it’s crucial to focus on fundamentals such as world representation, uncertainty handling, and reasoning, adapting lessons from the past to current trends.

Societal Impacts on AI Popularity

The rise in computer power and data availability has led to a more computerized society, where AI plays a significant role in tasks involving uncertainty, optimization, and personalization. Norvig advised following influential AI figures and exploring various resources for deeper AI insights. This societal shift, driven by interests in activities like reading, sharing pictures, and personalized recommendations, has increased the demand for AI solutions, focusing on optimization and uncertainty handling rather than just following definitive instructions.

Deep Learning Challenges and Meta-Analysis

Norvig pointed out challenges in deep learning, such as the lack of intuitive understanding and the necessity for meta-knowledge. He proposed a meta-analysis of deep learning applications to identify patterns and derive generalizable knowledge, aiding practitioners in their approaches.

The Right Language for AI Development

The choice of programming language is critical in AI development. Norvig recommended languages with strong community support and resources, like TensorFlow, and others including SciPy, NumPy, R, and MATLAB.

AI and Crowdsourced Knowledge

Integrating AI with structured data from sources like Wikipedia presents challenges, particularly in extracting information from text. Norvig emphasized the need for a portable way to integrate natural language understanding programs and make their outputs accessible through APIs.

Programming Paradigm Shift

The increasing role of AI in handling uncertainty could lead to a paradigm shift in programming. Traditional programmers might increasingly rely on APIs to access AI services, necessitating training in handling probabilistic outcomes.

AI Safety and Ethical Considerations

AI safety remains a significant challenge, with the focus being on risk management rather than absolute safety. Ethical considerations, including privacy, fairness, and security, are paramount, especially in areas like autonomous cars, where over-reliance could lead to safety issues.

Open Research Questions and Hybrid Approaches

Norvig pointed out unresolved issues in AI, such as the challenge of small data and the difficulty in combining specific and general knowledge. He advocated for a hybrid approach that merges learning and knowledge-based AI, combining research and application development for practical outcomes.

Conversational AI: Present and Future

The current state of conversational AI, while promising, faces challenges in understanding context and reasoning. The future lies in developing systems with clear boundaries and functionalities, akin to automated teller machines.

Educational Tools and Community Engagement in AI

Progress is being made in developing online resources and tools for AI education, with a focus on accessibility and engagement. Community engagement and personalized recommendations are key in fostering effective learning environments.

Over looked AI Applications

AI’s potential extends to civic, environmental, and health applications, offering solutions to societal challenges. Its ability to enhance public discourse, manage environmental issues, and advance personalized medicine underscores its transformative power.

The Future of Conversational AI

Challenges in Natural Conversation

Engaging in natural conversations with AI remains a significant challenge, particularly with the shift towards mobile-first and screenless devices, which underscores the need for conversational assistants. The goal is to create conversations that feel natural and achieve the user’s intended result, rather than just providing relevant information. It’s essential to strike a balance between making AI appear like a person and clarifying its limitations.

Chatbots and the Turing Test

Current chatbots can pass the Turing test in limited domains, offering human-like responses. However, these systems often struggle with complex conversations that require context, reasoning, and multiple levels of action.

User Interface and Usability

The user interface is crucial in creating usable AI systems. Systems that clearly define their capabilities and limitations are more helpful and practical than those aiming for perfect human-like interaction. Inspiration can be drawn from automated teller machines (ATMs) that perform specific tasks reliably within their defined scope.

Advances in Online Education: Insights from Peter Norvig

Tools for Course Creators and Learners

Tools for course authors and teachers have been developed to simplify the process of creating online courses, including video recording, editing, and course structuring. User interfaces in online learning systems have also been improved, making them more interactive and user-friendly.

Nonlinear Learning Paths

Linear learning paths are being replaced by nonlinear branching networks, allowing for personalized learning experiences. Students can choose their own paths based on their knowledge and interests.

Recommendations and Knowledge Graphs

Recommendation systems suggest relevant content and activities based on individual needs and progress. Knowledge graphs map relationships between concepts and skills, enabling the creation of personalized learning pathways.

Community and Motivation

The importance of motivation and community in online learning is recognized. Online platforms now include features that foster connections and peer-to-peer interactions, enhancing the learning experience and engagement.

Early Stages of Development

Despite progress, online education is still in its early stages. There is potential for further improvements in personalization, assessment, and technology integration. Ongoing research and innovation will continue to shape the future of online learning.

Overlooked Applications for AI and Opportunities for Improvement

Civic Education

AI can improve civic education, leading to a more informed and positive level of discourse.

Environmental Understanding

AI can analyze satellite data and other sources for a better understanding of the physical world, including weather patterns and climate change.

Health

AI is advancing in understanding diseases, drugs, and personalized medicine, offering significant healthcare improvement opportunities.

A Journey of Continuous Learning and Innovation

Peter Norvig’s webinar encapsulates AI’s journey – a blend of continuous learning, innovation, and practical application. From enhancing conversational AI to addressing ethical concerns, the field is ever-evolving, reflecting the dynamic interplay between research and real-world applications. This journey, marked by challenges and triumphs, underscores the importance of adapting and growing in a field that continually reshapes our world.


Notes by: Ain