Hal Varian (Google Chief Economist) – Hal Varian on Innovation (May 2018)
Chapters
00:00:07 Artificial Intelligence, Economics, and Industrial Organization
Speaker Introductions: Molly Yilbrock, head of communications and marketing at ESMT Berlin, welcomed the audience and introduced the speaker, Hal Varian. Varian is the chief economist at Google and an esteemed professor emeritus at the University of California, Berkeley. Yilbrock highlighted Varian’s contributions, including his renowned textbooks on economics and his co-authorship of the bestselling book, Information Rules.
ESMT Berlin Campus: ESMT Berlin was founded in 2002 and is known for its executive education programs, with 90% of the MBA class being non-German. The campus is located in the former state council building of the former GDR.
Open Lecture Series: The open lecture series at ESMT Berlin was inaugurated in 2009 and has hosted renowned speakers. Varian’s lecture focuses on the digital economy, the value of data, and competition in the digital age.
Moderator Introduction: Catalina Stefanescu-Kunze, professor of management science and Deutsche Post DHL chair at ESMT, was introduced as the moderator of the lecture.
About Hal Varian’s Lecture: Varian’s paper, titled “Artificial Intelligence, Economics, and Industrial Organization,” delves into the intersection of AI, economics, and industrial organization.
Introduction: In this presentation, Hal Varian delves into topics such as machine learning, artificial intelligence (AI), online competition, startups, and research and development (RD).
Machine Learning and AI: Machine learning and AI have seen significant advancements in recent years, particularly in image, voice recognition, and automatic translation. The combination of AI and machine learning techniques has led to breakthroughs in tasks once considered impossible. Factors contributing to the resurgence include better algorithms, hardware, data, and expertise.
Data and Kaggle: Data, the foundation of the information pyramid, can be refined into valuable information and knowledge. Unlike oil, data is not rivalrous and can be accessed simultaneously by multiple parties. Data ownership is a narrow concept; instead, rights, access, licensing, and contractual arrangements are more relevant. Kaggle competitions offer real-world examples of machine learning challenges with rewards for successful solutions. Kaggle participants represent a global community, fostering diversity and improvement in data utilization.
Data Portability: Google Takeout enables users to download their data easily, including photos, emails, and search history. Take-in, however, is rare, suggesting that the demand for data reuse is limited.
Sources of Data: The most common source of data is as a byproduct of operations. Free data sources include the web, web scrapers, and Common Crawl, providing petabytes of accessible data.
Data Acquisition Strategies: Faced with a lack of expertise and data in voice recognition, Google hired experts and created the Goog 411 information service. Leveraging this platform, Google trained its voice recognition system through user interactions, learning to recognize voices, accents, and directory information.
Data Sharing, Acquisition, and Cloud Services: Cloud providers like Microsoft, Google, and Amazon provide access to valuable data sources, including census, traffic, and weather data. The availability of such data repositories improves the quality of products and allows users to hit the ground running with data analysis.
Data Training Sets and Progress in Image and Video Recognition: Google’s Open Image Project released 9.5 million labeled images, aiding the development of image recognition systems. Video understanding, however, remains challenging, with machine systems lacking sufficient training in understanding video content and activities. Data from 8 million labeled YouTube videos is expected to contribute to advancing video understanding and resolving this problem.
Significance of the European Parliament Proceedings: The European Parliament’s production of multiple translations of its activities and speeches offers invaluable data for machine learning systems. This data has proven instrumental in the advancement of automated translation systems.
Importance of Data and the ImageNet Challenge: Data is essential for advancements in AI, but it’s not the sole factor. The ImageNet challenge, with its fixed training set, showed that error rates in image recognition significantly reduced through advancements in algorithms, hardware, and expertise.
00:18:36 AI Innovation in the Digital Realm: Competition, Commoditization, and Ease of
Introduction: * Hal Varian presents a comprehensive overview of the current state and impact of Artificial Intelligence. * Competition, technological advancements, and accessibility have revolutionized the field.
1. Image Recognition and Its Accessibility: * Image recognition has become highly accurate and accessible through services like Google Photos. * Examples demonstrate the ability to identify different dog breeds and provide detailed information. * Similar capabilities exist for plants, offering identification and care advice.
2. Extreme Competition in the Industry: * Intense competition exists between tech giants like Google, Amazon, and Microsoft in various sectors. * Pricing for services such as image recognition has dropped to a fraction of a cent per image. * This fierce competition has resulted in the commoditization of these services, making them accessible to all.
3. Evolution of Cloud Computing Services: * Cloud services provide access to algorithms, data sets, training, and assistance. * Ready availability of these services has reduced the barrier to entry for new businesses.
4. Ongoing Technological Advancements: * Investment in R&D, especially in autonomous vehicles, highlights the potential for technological progress. * Advances in image recognition, voice, and translation are evident, with features like smart speakers emerging.
5. Healthy Venture Capital Funding: * Investments in venture capital have reached record highs since 2000. * This funding surge is attributed to the increased ease of starting businesses with outsourced processes.
6. The “Construction Kit” for Starting a Business: * Services such as Kickstarter, LinkedIn, and cloud computing provide resources for new businesses. * Open-source software and legal assistance are also readily available. * Starting a business is now more efficient and accessible than ever before.
7. The Golden Age of Innovation: * Hal Varian asserts that we live in a transformative period for innovation in the digital context. * He expresses interest in the audience’s opinions, signaling a desire for further discussion.
Conclusion: * Hal Varian’s presentation highlights the significant impact of competition and technological advancements in the digital landscape. * The accessibility of AI-powered services, fierce competition, and ease of starting a business underscore the transformative nature of this era. * The expectation of continued innovation and the potential for further breakthroughs suggests an exciting future in the digital realm.
00:27:47 Uncharted Horizons: Exploring the Boundaries of Artificial Intelligence
AI’s Potential and Limitations: AI has made breakthroughs in fields like image recognition, voice recognition, and translation. Despite advancements, there are limits to AI’s usage and applicability. The field is still exploring the extent to which AI techniques can be pushed. Transfer learning allows AI to use knowledge from one area to solve problems in adjacent areas. Excitement surrounds AI’s potential, but the exact limits are unknown.
End-User Resistance and the Democratization of AI: Businesses and end-users will find AI services and capabilities helpful. AI will provide digital assistants with the same functionality as human assistants, democratizing access to these tools.
Causality in Big Data and Machine Learning: The holy grail of classical inference is determining causality. Google has built platforms for conducting experiments to understand causality. Quasi-causal data techniques, like difference in differences, can be used when designing experiments is not feasible.
Competition in the Online Marketplace: Online competition is more intense due to the ease of switching to alternative providers. This competition drives innovation and technological improvement in the online space. Low entry costs enable new entrants with innovative ideas to enter the market. The trend of intense competition in the online marketplace is expected to continue.
00:33:55 Artificial Intelligence and the Future of Online Business
Data Protection Regulations: The European Union’s General Data Protection Regulation (GDPR) will soon come into effect, impacting online activities and businesses. Large companies like Google, Facebook, Microsoft, and Amazon have prepared for compliance over the past 18-24 months. Concerns exist about the impact on small companies, who may struggle to meet the reporting and user request requirements, potentially affecting the vibrant online ecosystem.
Artificial Intelligence (AI) and Data Acquisition: AI requires expertise and data for effective application. Real-world data often comes from professionals with practical experience and psychological connections with consumers. Acquiring data from professionals can be challenging, as they may be hesitant to share it with enterprises that might replace them. Examples exist where AI has been successfully applied in real-world scenarios, such as improving the productivity of a family farm through machine learning. Kaggle datasets and YouTube images and videos provide accessible data sources for researchers. Data availability is generally not a problem, but finding the expertise to extract value from the data is the key challenge.
Information Economics and Hidden Information: Information economics remains relevant despite the advent of AI. Hidden action and hidden information continue to be important concepts. Computers offer better monitoring and compliance capabilities, aiding in the verification of contract performance. Online advertising has evolved with improved performance measurement, linking ad payments to actual clicks and visits.
Competition in Online Search: Competition in online search is intense, especially in commercial searches where revenue is generated. Google faces competition from numerous travel providers, shopping services, and comparison engines in these commercial areas. Only 6% of clicks are on ads in general purpose search, making competition fierce in this segment.
Content Filtering and Machine Learning: YouTube is actively using machine learning techniques to filter a substantial number of videos. Machine learning-based filtering of inappropriate content, such as football videos or politically sensitive material, is being developed and refined. Facebook currently relies on human moderators for content filtering, but machine learning is expected to play a greater role in the future.
00:43:32 Challenges and Opportunities in the Era of Artificial Intelligence
AI Challenges and Adversarial AI: Adversarial AI can fool AI systems with carefully designed images, creating optical illusions that differ from those that deceive humans. Developers must address the limitations of AI in recognizing and dealing with adversarial attacks.
Right to an Explanation: Some AI systems lack the ability to explain their decision-making process, raising concerns about transparency and accountability. The example of AI identifying gender from medical eye images highlights the need for understanding how AI systems reach conclusions. Efforts are ongoing to develop methods for explaining the decisions made by AI systems.
Regulation of Data Economy: Regulation of the data economy is expected in Europe, the US, and China, with differing approaches. The potential benefits of AI in sensitive areas like healthcare necessitate careful consideration before implementing regulations that could hinder innovation.
Opportunities in Medical Studies: Machine learning offers significant potential in medical studies and healthcare. The benefits of AI in these areas must be balanced with privacy concerns and the need for ethical data handling.
Chinese Internet Giants: The Chinese internet market is distinct, with highly entrepreneurial private enterprises and state-owned enterprises. Exporting the Chinese model to other countries may be challenging. Skepticism exists about the ability of Chinese internet giants to become global companies.
Abstract
The Transformation of Data, AI, and Online Competition: Insights from Hal Varian’s Lecture at ESMT Berlin
Introduction
Welcome to a thought-provoking summary of a lecture delivered by Hal Varian, Google’s Chief Economist, at ESMT Berlin. Varian, a distinguished figure in economics and technology, presented a comprehensive analysis of the interplay between Artificial Intelligence (AI), Economics, and Industrial Organization (AIEIU). This summary explores the lecture’s main themes, encompassing the advancements in AI and machine learning, the dynamics of online competition, and the implications of data handling and privacy regulations.
ESMT Berlin and Hal Varian: A Confluence of Expertise
Founded in 2002, ESMT Berlin is a leading business school known for its MBA, master’s in management, and executive education programs. Its open lecture series, initiated in 2009, has featured renowned speakers, including Hal Varian, Google’s chief economist since 2002. Varian, also a Professor Emeritus at UC Berkeley, is celebrated for his economic textbooks and the bestselling book “Information Rules.”
Lecture Overview: AIEIU in the Digital Age
Varian’s lecture encompassed a wide range of topics, including machine learning, AI, online competition, startups, and R&D. He highlighted recent breakthroughs in image, voice, and language recognition, attributing them to improved algorithms, hardware, data availability, and expertise. Varian emphasized that expertise is the scarcest resource, with platforms like Kaggle fostering collaboration and problem-solving.
Data: The New Oil in AI Development
Varian drew an interesting analogy between data and oil, noting that data, unlike oil, is non-rivalrous and can be simultaneously accessed by multiple parties. He discussed the nuances of data ownership, focusing on rights, access, licensing, and contractual arrangements. Services like Google Takeout facilitate data portability, but the concept of data intake, where others use your data for new purposes, is less prevalent. Varian pointed out that operational byproducts are a common data source, and vast amounts of free data are available online through methods like web scraping and Common Crawl.
Pioneering AI Applications: Voice Recognition and Beyond
Google’s Goog 411 service was a notable example Varian provided, illustrating how it gathered data to train a sophisticated voice recognition system. This initiative laid the groundwork for one of the world’s best voice recognition systems. Additionally, he highlighted Google’s release of extensive image and video datasets that have significantly advanced image recognition technologies and are contributing to solving video content identification challenges.
AI and Machine Learning: Key Insights and Challenges
Varian discussed the commoditization of AI services due to intense competition among tech giants, leading to affordable image recognition, transcription, voice, and translation services. He noted the easy entry for startups in the AI market, facilitated by cloud computing and open-source software. Despite advancements, AI still faces limits, particularly in areas like transfer learning, end-user resistance, and establishing causality in big data.
GDPR and Online Activity: A Balancing Act
The implementation of GDPR posed new challenges, particularly for small businesses grappling with compliance burdens. Varian expressed concerns about the potential impact of GDPR on the dynamic online ecosystem, given the increased friction in data handling.
The Economics of Information: A Persistent Relevance
In the digital age, economics remains relevant, especially in scenarios involving hidden actions and information. Computers have enhanced monitoring and compliance capabilities, which are critical in verifying performance in contracts. Online advertising and search have become highly competitive sectors, with Google maintaining a strong position in general-purpose searches while facing fierce competition in commercial searches.
AI in Content Filtering and the Challenges Ahead
YouTube’s use of machine learning for content filtering was another focal point. Varian addressed the challenges in AI, such as adversarial AI, where deliberately designed images can fool AI systems, and the ‘right to explanation,’ highlighting the difficulty in understanding AI decisions.
The Future of Data Regulation and the Global Internet Landscape
Varian touched on the trend toward stricter data privacy regulations, as seen in initiatives like California’s GDPR-like referendum. He advocated for a careful approach to regulation, balancing potential harms and benefits. The unique environment of Chinese internet giants and their uncertain global reach was also discussed, shedding light on the complex relationship between private enterprises and state entities in China.
Conclusion
Hal Varian’s lecture at ESMT Berlin offered profound insights into the evolving landscape of AI, machine learning, and online competition. His discussion spanned from the technical advancements in AI to the broader economic and regulatory implications, providing a comprehensive overview of the challenges and opportunities in the digital age. As AI continues to transform industries, understanding these dynamics becomes crucial for businesses, policymakers, and individuals navigating this rapidly changing world.
Supplemental Insights:
1. Image Recognition and Its Accessibility:
Image recognition has experienced remarkable advancements, becoming highly accurate and accessible. Services like Google Photos allow users to identify different dog breeds and provide detailed information. Similarly, apps can identify plants, offering identification and care advice.
2. Extreme Competition in the Industry:
The competition among tech giants like Google, Amazon, and Microsoft is intense across various sectors. As a result, the pricing for AI services has dropped significantly, making them accessible to all.
3. Evolution of Cloud Computing Services:
Cloud services provide access to algorithms, data sets, training, and assistance, reducing the barrier to entry for new businesses.
4. Ongoing Technological Advancements:
Investment in R&D, particularly in autonomous vehicles, highlights the potential for technological progress. Advancements in image recognition, voice, and translation are evident, with features like smart speakers emerging.
5. Healthy Venture Capital Funding:
Venture capital funding has reached record highs since 2000, attributed to the increased ease of starting businesses with outsourced processes.
6. The “Construction Kit” for Starting a Business:
Services like Kickstarter, LinkedIn, and cloud computing provide resources for new businesses. Open-source software and legal assistance are readily available, making starting a business more efficient and accessible.
7. The Golden Age of Innovation:
Hal Varian believes we live in a transformative period for innovation in the digital context. He expressed interest in the audience’s opinions, signaling a desire for further discussion.
8. Limits, Applications, and the Future of Artificial Intelligence:
AI has made breakthroughs in various fields, yet its limitations are still being explored. Transfer learning allows AI to use knowledge from one area to solve problems in related areas. End-user resistance and the democratization of AI are important considerations. Establishing causality in big data remains a challenge.
9. Data Protection Regulations:
The European Union’s GDPR will soon come into effect, impacting online activities and businesses. Large companies have prepared for compliance, but concerns exist about the impact on small companies.
10. Artificial Intelligence (AI) and Data Acquisition:
AI requires expertise and data for effective application. Acquiring data from professionals can be challenging, as they may be hesitant to share it with enterprises. Examples exist where AI has been successfully applied in real-world scenarios, such as improving farm productivity. Kaggle datasets and YouTube images and videos provide accessible data sources.
11. Information Economics and Hidden Information:
Information economics remains relevant despite the advent of AI. Hidden action and hidden information are still important concepts. Computers offer better monitoring and compliance capabilities, aiding in the verification of contract performance. Online advertising has evolved with improved performance measurement.
12. Competition in Online Search:
Competition in online search is intense, especially in commercial searches where revenue is generated. Google faces competition from travel providers, shopping services, and comparison engines.
13. Content Filtering and Machine Learning:
YouTube uses machine learning to filter a substantial number of videos. Machine learning-based filtering of inappropriate content is being developed and refined. Facebook currently relies on human moderators, but machine learning is expected to play a greater role in the future.
14. Insightful Points from the Transcript:
AI Challenges and Adversarial AI:
– Adversarial AI can fool AI systems with carefully designed images, creating optical illusions that differ from those that deceive humans.
– Developers must address the limitations of AI in recognizing and dealing with adversarial attacks.
Right to an Explanation:
– Some AI systems lack the ability to explain their decision-making process, raising concerns about transparency and accountability.
– The example of AI identifying gender from medical eye images highlights the need for understanding how AI systems reach conclusions.
– Efforts are ongoing to develop methods for explaining the decisions made by AI systems.
Regulation of Data Economy:
– Regulation of the data economy is expected in Europe, the US, and China, with differing approaches.
– The potential benefits of AI in sensitive areas like healthcare necessitate careful consideration before implementing regulations that could hinder innovation.
Opportunities in Medical Studies:
– Machine learning offers significant potential in medical studies and healthcare.
– The benefits of AI in these areas must be balanced with privacy concerns and the need for ethical data handling.
Chinese Internet Giants:
– The Chinese internet market is distinct, with highly entrepreneurial private enterprises and state-owned enterprises.
– Exporting the Chinese model to other countries may be challenging.
– Skepticism exists about the ability of Chinese internet giants to become global companies.
Hal Varian's journey illustrates the convergence of economics and technology, from his academic pursuits to his influential role at Google, shaping the field of information economics. Varian's insights on economic measurement, career advice, and the impact of globalization provide valuable perspectives in the digital age....
Hal Varian's work spans traditional economics to digital era complexities, addressing auction models, sales economics, and consumer behavior in the context of internet governance, privacy, and technological advancements. Varian's insights map the current state and future directions of economic theory and practice....
Hal Varian's contributions at Google highlight the transformative power of data analysis in shaping business strategies and decision-making, emphasizing the growing significance of statistical modeling and data-driven insights in the tech industry. Google's continued innovation in data analysis underscores the evolving nature of the field, with a growing demand for...
Google's digital tools, like Google Trends, revolutionize economics and forecasting by offering unprecedented insights into human behavior and economic activity. Google's tools enhance economic forecasting by combining private sector data with traditional government data....
Collaborative efforts and policy-making are crucial for a robust transatlantic digital economy, combining strengths for global innovation and growth. The future of work involves remote collaboration and apprenticeship models, emphasizing a balance between digital transformation and traditional skills....
Google's innovation and data analytics influence technological advancement, societal evolution, and ethical considerations in the digital age. Hal Varian's economic expertise guides Google's auction systems, ad revenue, and societal impact analysis....
Machine learning has seen exponential growth and deep learning has revolutionized various fields, from healthcare to robotics, by learning from raw data and handling diverse data types. AutoML and specialized accelerators like TPUs have accelerated machine learning advancements....