Michael I. Jordan: Machine Learning, Recommender Systems, and Future of AI | Lex Fridman Podcast #74

Introduction

In this section, the host introduces Michael I. Jordan, a professor at Berkeley and one of the most influential people in the history of machine learning statistics and artificial intelligence.

Introducing Michael I. Jordan

• Michael I. Jordan is a professor at Berkeley and one of the most influential people in the history of machine learning statistics and artificial intelligence.

• He has been cited over 170 thousand times and has mentored many world-class researchers defining the field.

The Michael Jordan of Statistics

In this section, the host talks about how Michael I. Jordan is often referred to as "the Michael Jordan of machine learning" but prefers to be called "the Miles Davis of machine learning."

Referring to Michael I. Jordan

• Although he is often referred to as "the Michael Jordan of machine learning," he prefers to be called "the Miles Davis of machine learning."

• John Le calls him "the Miles Davis of machine learning" because he reinvents himself periodically and sometimes leaves fans scratching their heads after changing direction.

Broadening the Scope of AI

In this section, the host talks about how Michael I. Jordan argues for broadening the scope or artificial intelligence field in many ways.

Broadening AI's Scope

• According to Michael I. Jordan, we need to broaden our understanding of what AI is beyond just engineering algorithms and robots.

• We need to understand and empower human beings at all levels from individual to civilization as a whole.

• This podcast shares that same spirit by seeing AI as a deeply human endeavor.

Cash App Ad

In this section, there is an ad for Cash App which allows users to send money, buy bitcoin, invest in the stock market, and trade fractional shares.

Cash App

• Cash App is a finance app that allows users to send money, buy bitcoin, invest in the stock market, and trade fractional shares.

• The order execution algorithm behind the scenes to create the abstraction of the fractional orders is an algorithmic marvel.

• If you use code "lex podcast" when getting Cash App from the app store or Google Play, you'll get $10 and Cash App will donate $10 to one of their favorite organizations helping to advance robotics and STEM education for young people around the world.

History of AI

In this section, Michael I. Jordan gives a brief history of computer science and AI as he saw it including four generations of AI successes.

Brief History of Computer Science and AI

• Michael I. Jordan compares the development of AI to chemical engineering from chemistry or electrical engineering from electromagnetism.

• In the 30s or 40s there wasn't yet chemical engineering; there was chemistry, fluid flow, mechanics but people viewed interesting goals like building factories that make chemicals products viably safely at scale.

• Chemical engineering developed in parallel with these goals.

• Similarly, what's happening now is not just AI as an intellectual aspiration but akin to chemical engineering or electrical engineering.

• There have been four generations of AI successes: rule-based systems (60s), knowledge-based systems (70s), machine learning (80s/90s), deep learning (2000s-present).

Introduction to Electrical Engineering

In this section, the speaker introduces electrical engineering as a field that involves building circuits, modules, and bringing electricity from one point to another safely. The speaker also discusses how statistics and computational ideas have led to the development of a proto form of engineering called machine learning.

Electrical Engineering

• Electrical engineering involves building circuits, modules, and bringing electricity from one point to another safely.

• Machine learning is a proto form of engineering based on statistical and computational ideas of previous generations.

Understanding the Human Brain

In this section, the speaker discusses our current understanding of the human brain. He explains that we have no clue how the brain does computation and that it is a problem for the next few centuries. The speaker also talks about metaphors used in neuroscience and how they are not real science per se.

Understanding the Human Brain

• We have no clue how the brain does computation.

• Real neuroscientists agree that understanding the human brain is a hundred-year task.

• Metaphors used in neuroscience are not real science per se.

• The problem of understanding the human brain is fantastic but we have our metaphors about it.

Brain Computer Interfaces

In this section, the speaker talks about brain computer interfaces (BCIs). He mentions Elon Musk's Neuralink company which is working on putting electrodes into the brain to read and send electrical signals. The speaker expresses hope for interesting things to happen from research but expects more progress in areas like Alzheimer's disease.

Brain Computer Interfaces

• Elon Musk's Neuralink company is working on putting electrodes into the brain to read and send electrical signals.

• The speaker hopes for interesting things to happen from research but expects more progress in areas like Alzheimer's disease.

Understanding Breakthroughs in Science and Engineering

In this section, the speaker discusses the difference between scientific understanding and engineering breakthroughs. He also talks about how some of the current technological advancements are not scientifically understood.

Breakthroughs in Science vs Engineering

• The speaker distinguishes between scientific understanding and engineering breakthroughs.

• He believes that some of the current technological advancements are not scientifically understood.

• The speaker acknowledges that there will be all kinds of breakthroughs but does not like to prize theoretical breakthroughs over practical ones.

• He points out that he would rather point to Edison's era than Kolmogorov's era as a source of inspiration for breakthrough ideas.

Elon Musk's Views on AI

• The speaker disagrees with Elon Musk's views on AI and thinks that it is not for our generation or even for this century.

• He criticizes Musk for saying things about AI that he knows very little about, which leads people astray when he talks about things he doesn't know anything about.

Deep Aspects of Intelligence

• The deeper aspects of intelligence are not going to happen now, such as programming a computer to understand natural language to be involved in a dialogue we're having right now.

• However, there have been breakthrough technologies such as Google, Amazon, and Uber that bring value to human beings at scale in new brand new ways based on data flows.

Introduction

In this section, the speaker discusses how AI is emerging and will be seen as a breakthrough in this era. He also talks about the history of AI and machine learning.

History of AI

• John McCarthy was a philosopher who wanted to put thought into a computer.

• He wrote down logical formulas and algorithms that would do that.

• However, what's happening now is not about building intelligence but making good decisions based on data.

• The goal is to build really good working systems at planetary scale.

Machine Learning vs. AI

In this section, the speaker explains the difference between machine learning and AI.

Machine Learning

• Machine learning is all about making large collections of decisions under uncertainty by large collections of entities.

• It involves systems that learn and make decisions based on pattern recognition.

• The decisions are consequential in the real world, such as medical operations or driving down the street.

• The goal is to use computers to help these things go forward.

Reclaiming the Word "AI"

In this section, the speaker discusses his dislike for using the term "AI" and suggests using a new word instead.

New Word for AI

• The speaker thinks that "AI" was a bad choice from the Dartmouth conference many decades ago.

• He wants a new word to describe what they're doing now: building really good working systems at planetary scale.

The Origins of AI Terminology

In this section, the speaker discusses the origins of AI terminology and how it has affected the field.

The Origins of AI Terminology

• McCarthy invented the term "artificial intelligence" to differentiate himself from others in the field.

• However, this led to unrealistic promises about what AI could achieve.

• The speaker believes that multiple voices are needed in the field, including more sober voices that emphasize fundamental principles and theories.

Progression of Science

In this section, the speaker discusses whether science progresses by personalities or by fundamental principles and theories.

Progression of Science

• The speaker believes that both personalities and fundamental principles are necessary for scientific progress.

• However, he notes that there is currently a personality type in ascendance that is too exuberant and makes unrealistic promises about what AI can achieve.

Disagreements in Machine Learning

In this section, the speaker talks about his relationship with Yann LeCun and any disagreements they may have had.

Disagreements in Machine Learning

• The speaker says he doesn't disagree with Yann LeCun on much.

• They both have a "let's build it" mentality but differ slightly on emphasis. Yann emphasizes pattern recognition while the speaker emphasizes prediction.

• However, perfect prediction is limited because we cannot predict novel situations or account for market forces or risks involved.

Introduction to Decision Making in AI

In this section, the speaker discusses the importance of decision making in AI and how it differs from pattern recognition.

Decision Making vs Pattern Recognition

• Decision making involves consequential decisions in the real world under uncertainty and messiness.

• Pattern recognition is more constrained to lab data sets.

• Decision making involves considering the consequences of decisions on others and the economy around them.

Prediction in Real World Events

In this section, the speaker talks about prediction in real-world events and how it cannot be done with just data sets.

The Importance of Context

• Prediction must be done within the context of strategic decision-making.

• Gathering data is not enough; reasoning processes around data are also important.

AI Systems for Decision Making at Scale

In this section, the speaker discusses AI systems that help make decisions at scale.

Example System: Music Market

• A music market does not currently exist for creators who make good music but are not famous.

• Record companies prop up a few superstars while many creators do not make money.

• People listen to a lot of music, but most of it is by famous people rather than lesser-known creators.

Overall, the transcript covers topics related to decision making in AI and its importance in real-world scenarios. The speaker emphasizes that decision making involves more than just gathering data and requires consideration of consequences on others and broader economic factors. Additionally, they discuss how prediction must be done within the context of strategic decision-making and provide an example system, a music market, to illustrate the challenges faced by creators who are not famous.

Creating a New Market with AI

In this section, the speaker discusses how AI can create new markets and jobs for music creators. He talks about how companies like Spotify make money off of subscriptions or advertising and offer bits and pieces of it to a few people. However, if you're a creator of music that people want to listen to, you should have access to data showing where your songs were listened to.

The Value of Transparency

• Creators should have access to transparent data showing where their songs were listened to.

• This data should be vettable so that people know it's real.

• With this information, creators can perform in places where they have a following and make money from live shows.

Creating New Markets

• AI creates new markets by connecting producers and consumers.

• Personal connections are important in the value of music.

• The internet should provide an easy way for fans to buy merchandise from their favorite musicians.

• Companies can take a small cut (e.g., 5%) from these transactions.

Challenges Ahead

• Creating new markets requires thinking through unwanted aspects of the market such as bad actors using data in wrong ways or failing to deliver value.

• Principles will need to be developed for creating two-way markets at scale.

How Can Companies Create Such Markets?

In this section, the speaker discusses what companies like Spotify or YouTube can do to create such markets. He argues that creating an ecosystem is key rather than just relying on top-down solutions from Silicon Valley.

The Importance of Culture

• Companies need to create an ecosystem in which creators are wanted.

• This means creating a culture that values and supports young creators.

Giving Back to USC Film School

In this section, the speaker talks about giving back to USC film school and how it is a common practice among rich white people. He also emphasizes the importance of respecting cultural differences.

Giving Back to USC Film School

• Giving back to USC film school is a common practice among rich white people.

• Companies cannot artificially create culture and must respect cultural differences.

• Technology should blend with cultural meaning.

The Role of Recommender Systems

In this section, the speaker discusses the role of recommender systems in connecting consumers with creators. He also shares his experience with Amazon's recommender system.

The Role of Recommender Systems

• Recommender systems play a significant role in connecting consumers with creators.

• A good recommender system is better than a bad one.

• Amazon's recommender system was historically one of the speaker's favorites.

• Recommendation systems are not meant to impose what we learn but rather find out what's in the data.

Politics and Recommendation Systems

In this section, the speaker talks about recommendation systems in politics and news. He also critiques Facebook for its monetization by advertising approach.

Politics and Recommendation Systems

• YouTube, Twitter, and Facebook have to deal with recommendation systems suggesting political content.

• Monetization by advertising is broken with some companies like Facebook not trying to connect producers and consumers economically.

• Advertising filled the gap for services that people don't want to pay for.

• Google and Facebook cornered the advertising market.

The Missed Opportunity of Direct Consumer-Producer Markets

In this section, the speaker discusses the missed opportunity of creating a direct market between producers and consumers. He argues that companies like Facebook and Google have not thought long and hard about this concept.

Direct Market Between Producers and Consumers

• There is an economic value to be liberated by connecting human producers and consumers directly.

• Companies like Facebook and Google should have created a direct market 20 years ago in parallel with the advertising ecosystem.

• Facebook is making huge amounts of money off of advertising, which is all connected to advertisers. They are incentivized to get more people to click on certain things.

• When problems arise, companies try to adjust with smart AI algorithms, but it gets into all the complexity in life.

• Instead of trying to fix individual problems, companies should fix the whole business model.

Example: Travel Recommendations

• A direct market could be used for travel recommendations. For example, someone going to Mumbai could broadcast their trip and receive personalized recommendations from someone on the other side of a market who knows them in some way.

• This person would pay for real value rather than being bombarded with advertisers. Micro payments or even larger payments could be made for these services.

• This creates a gig economy where people can make more videos or write papers if they are connected to a market.

Company Differences

• Amazon thinks about markets all the time because they were already out in the real world delivering packages. They worry about sellers and some things they do are great while others are not so great.

• Google hovers somewhere in between. They did not get it for a long time, but they see that YouTube is more pregnant with possibility than they might have thought.

• Silicon Valley has been dominated by the Google and Facebook mentality of subscription and advertising, which is the core problem.

Moving Away from the Advertising Model

In this section, the speaker discusses whether we should move away from the advertising model and have a direct connection between consumers and creators. The role of advertising in bringing products to customers is also explored.

The Role of Advertising

• Advertising is a signal that a company believes in its product enough to pay for it.

• Without advertising, new products may not be discovered by consumers.

• Advertising provides information flow in a good market.

Moving Away from Advertising

• The speaker believes we should slowly move away from the advertising model.

• Companies like Amazon are trying to create markets without relying on advertising.

• Society tends to move away from things that annoy people more than they provide information, so companies will need to reduce their reliance on advertising.

• A good company will realize that reducing advertising while increasing real value being delivered can cancel out lost revenue from poor quality ads.

The Role of Advertising in the Market

In this section, the speakers discuss the role of advertising in the market and whether it can be improved algorithmically. They also touch on the issue of privacy and trust when it comes to companies like Facebook.

The Value of Advertising

• Advertising is a way to get signals into a market that don't come any other way, especially algorithmically.

• Direct connection between consumer and producer is the best form of advertisement.

• It's possible to make advertisements better algorithmically to where it actually becomes a connection almost addressed.

Privacy Concerns with Facebook

• Facebook is an intermediary between consumers and producers, but they don't care about individuals in any real sense.

• Most people find it creepy that Facebook knows things about them and could exploit their information.

• There are companies like Microsoft that have more information about us than Facebook does, but they are trusted more because they have decided not to do creepy things with our data.

The Importance of Trust in Transactions

• When people connect as producers and consumers, they see each other and sense that if they transact some happiness will go up on both sides.

• Moments when people want to buy something quickly are not good times for advertising. People just want to go in and get what they need without being advertised at.

I apologize, but I cannot provide a summary of the transcript as there is no transcript provided in the prompt. Please provide me with the transcript so that I can create a comprehensive and informative markdown file.

The Role of AI in Personalization

In this section, the speaker discusses his concerns about companies using AI to collect and analyze personal data to personalize experiences for users.

Personalization vs. Privacy

• The speaker believes that while it may be possible to use AI to figure out one person, it is not feasible to do so for 500 million people.

• He argues that humans are complex and have unique quirks that cannot be predicted by algorithms.

• Instead of trying to anticipate users' preferences, the speaker suggests creating spaces where human creativity can flourish without being replaced by technology.

Recommender Systems

• The speaker believes that recommender systems are currently "creepy" because they rely on collecting traces of user behavior rather than direct interaction with users.

• While he acknowledges the potential value of natural language processing in helping users discover new things, he emphasizes the importance of transparency and control over personal data.

Alexa as a Research Platform

• The speaker sees potential in using Alexa as a research platform for reasoning tasks such as reminding him to turn off the water in his garden.

• However, he does not want Alexa or any other technology to intrude on his privacy or replace human interaction.

Control Over Privacy

In this section, the speaker discusses the importance of having control over privacy and how it should be an individual decision.

What Does Control Over Privacy Look Like?

• Privacy is not a zero-one legal thing, it's not just about which data is available or not.

• A couple of hundred years ago, everyone lived in villages where everybody knew everything about you. There was no privacy, but people helped each other because they knew everything about you.

• The answer to whether privacy is good or bad depends on the individual. Some people want their whole life out there while others don't.

• People should have some agency over who knows what about them. They shouldn't be adrift in a sea of technology where they have no idea.

Complexity of Privacy

In this section, the speaker talks about the complexity of privacy and how it requires new structures that we don't have right now.

Emerging Engineering Field

• The emerging engineering field will play a big part in solving the complexity of privacy.

• It's not just technology or legal scholars meeting technologists; there needs to be whole layers around it.

• For things as deeply interesting as privacy, which is at least essential as electricity, it's going to take decades to work out.

Importance of Sober Discussions

In this section, the speaker emphasizes the importance of sober discussions in public discourse regarding technology and privacy.

Resurgence of Podcasts

• There's a resurgence of podcasts because people are really hungry for conversation but their technology is not helping much.

• Comment sections on anything including YouTube are not helpful either.

Challenging Discussions

• The sober discussions in the middle, which are the challenging ones to have, are where we need to be having our conversations.

• There's not many forums for those discussions.

Worlds of Trust

In this section, the speaker talks about how technology can help create worlds of trust where people can have meaningful discussions without wasting their time.

Creating Worlds of Trust

• It's technically possible to create worlds of trust where people can enter and trust the people there.

• These worlds should have less anonymity and a little more locality.

• People should know what they're getting into when they enter these worlds.

Technology and Human Nature

In this section, the speaker discusses the impact of technology on human nature and whether humans are fundamentally good or evil.

Humans are Fundamentally Good but Limited

• The speaker believes that humans are fundamentally good but limited.

• Technology could open us up to more perspectives and understanding, which could help reduce ignorance and conflicts.

• Wars in human history often happen because of ignorance, lack of understanding, and different perspectives.

• People have grievances and grudges that cause them to do things they regret.

Optimization Problems

• Individual human life and society as a whole are complex phenomena that cannot be modeled as an optimization problem.

• Optimization is just one branch of mathematics that talks about finding a single point that is the optimum of a criterion function.

• Sampling is another mathematical paradigm that tries to find points with high density from a distribution or density surface.

Stochastic Worldview

• The world is highly stochastic, with massive uncertainty about what will happen in the next few minutes or hours.

• Probability distributions can be used to reason about uncertain events and make decisions based on rough estimates.

Optimization in Multi-Agent Systems

• Optimization makes sense for single agents trying to optimize some objective function, but it becomes more complicated when dealing with multi-agent systems where game theory concepts start popping up.

Game Theory and Equilibria

In this section, the speaker discusses game theory and equilibria. He explains that game theory is a way to understand what's happening in markets and other systems, and that there are certain principles that allow us to understand these systems.

Saddle Points in Game Theory

• A saddle point is a type of equilibrium in game theory.

• Nash equilibria have limitations and are not always explanatory.

• There are different types of equilibria, such as Stackelberg equilibrium.

Moving Towards or Away from Equilibria

• Mathematically, equilibria have certain topologies or shapes.

• Dynamically, it's important to know how to move towards or away from an equilibrium.

• These questions have been studied but some remain open problems.

Strategic Settings and Data Collection

• In strategic settings with uncertainty, data collection can be challenging.

• It may be necessary to push someone into a part of the space where you don't know much about them in order to collect data.

• Even games like poker can provide insights into strategic settings with uncertainty.

Optimization in Deep Learning

In this section, the speaker discusses the optimization problems in deep learning and how they can be solved.

Smoothness of Surface

• The surface of deep learning is pretty smooth from an optimization point of view.

• If it's over-parameterized, there are lots of paths down to reasonable optima.

Co-evolution Process

• Gradient descent algorithms are dominant in deep learning.

• There are ongoing research and development to figure out which algorithm is best suited for a particular situation.

• The co-design of the surface and algorithm will put pressure on the actual architecture.

Importance of Gradients

• Gradients are amazing mathematical objects that have a lot of properties that can be exploited.

• Even trivial operators like gradient descent are not trivial and exploiting their properties is still very important.

Stochasticity in Optimization

In this section, the speaker discusses the importance of stochasticity in optimization and how it can save you from certain features of surfaces that could hurt you if you were doing one thing deterministically.

Stochasticity and its Importance

• Stochasticity is important because it saves you from certain features of surfaces that could hurt you if you were doing one thing deterministically.

• Stochasticity means that by chance there's very little chance that you would get hurt.

• Stochasticity is especially appealing in high dimensions for a large number of reasons.

Nestorov's Work on Acceleration

In this section, the speaker talks about Nestorov's work on acceleration and how it achieves one over k squared.

Nestorov's Work on Acceleration

• Nestorov discovered a new algorithm that uses two gradients and puts those together in a certain kind of obscure way.

• The algorithm doesn't even move downhill all the time; sometimes it goes back uphill.

• It achieves one over k squared and has a mathematical structure.

Introduction to Statistics and Decision Making

This section provides an introduction to statistics and decision making. It explains how statistics is a set of principles that allow you to make inferences and decisions with some reason to believe you're not going to make errors.

The Origins of Statistics

• Statistics was originally called inverse probability, developed around 250 years ago as a formal discipline.

• It was used to explain gambling situations, where physicists started paying attention.

• Laplace analyzed data from a census of France for policy determination, leading him to call the field "statistics" because it's the study of data for the state.

Decision Theory in Statistics

• Decision theory is the starting point for most advanced statistical curricula.

• Statistics is all about decisions - what is the most beautiful, mysterious idea that you've come across?

James Stein Estimation

This section discusses James Stein estimation, which is a surprising concept that takes time to wrap your head around.

Understanding James Stein Estimation

• James Stein estimation is too technical but can be understood through Steven Stickler's paper on it.

• It helps view paradoxes and defeats attempts at understanding it.

Introduction to Bayesian and Frequentist Approaches

This section introduces the concepts of Bayesian and frequentist approaches to statistical inference.

Bayesian vs Frequentist Approaches

• The frequentist approach involves taking the data as random and averaging over the distribution, while the Bayesian approach involves taking the unknown parameter as random and averaging over it.

• The frequentist approach focuses on a single dataset, while the Bayesian approach conditions on a particular dataset.

• Empirical Bayes is a middle ground between these two approaches that starts with the Bayesian framework but plugs in estimates for uncertain quantities.

Frequency Guarantees

• Frequency guarantees are important when building software or making decisions based on statistical inference. They involve looking at how well a certain procedure will perform under all possible datasets.

• The frequentist approach is particularly appropriate in situations where you are working with scientists who have domain expertise.

False Discovery Rate

• False discovery rate is a criterion for evaluating multiple hypothesis tests. It involves looking at the fraction of false discoveries among all discoveries made.

False Discovery Rate and Human Intelligence

In this section, the speaker discusses false discovery rate and its Bayesian interpretation. They also touch on the topic of human intelligence and how it is studied.

False Discovery Rate

• The Bayesian approach goes from data to state of nature while classical frequency goes from state of nature to data.

• Empirical Bayes realizes that some priors can be estimated in a reasonable way.

• False discovery rate is a beautiful set of ideas that came out around 1960 by Robin's. Brad Efron has written about it in various papers and books.

Human Intelligence

• Defining what intelligence means is a difficult question for psychologists who aim to understand human intelligence.

• Psychologists study how babies understand things, how children learn words, and try to figure out things about human reasoning patterns.

• Humans' ability to reason abstractly, communicate effectively, and interact with their environment is staggeringly rich and complicated.

• Intelligent systems come in many forms such as markets which are decentralized sets of decisions that work at all scales.

The Future of Intelligence

In this section, the speaker discusses the concept of intelligence and how it relates to markets and computer systems.

Markets as a Form of Intelligence

• Markets are intelligent and have principles such as supply and demand curves, matching, and auctions.

• These principles lead to a form of intelligence that is not necessarily human intelligence but rather another kind of intelligence.

• To achieve general intelligence or understand intelligence in a deep sense, we need to move beyond just human intelligence.

Intelligent Infrastructure

• The speaker defines different kinds of intelligent infrastructure (III), which includes markets.

• When considering the perspective of markets and intelligent infrastructure, it's difficult to predict if the intelligence will increase linearly or exponentially.

Human-Level Superhuman Intelligence

• If we were to create a human-level superhuman level intelligent system, what would it take?

• This question is science fiction and not something that should be focused on too much in public fora.

• Science fiction can be intriguing for thinking about where thought exists.

Advice for Mentees

• The speaker has mentored some seminal figures in the field but does not provide any specific advice in this section.

Apprenticeship and Community in Science

In this section, the speaker discusses how success in science is not solely based on brilliance or great ideas but rather on apprenticeship and hard work. The graduate school is a wonderful phenomenon that emphasizes apprenticeship with an advisor and belonging to a community.

Importance of Apprenticeship

• Success in science is not solely based on brilliance or great ideas.

• Apprenticeship involves spending time, working hard, practicing basics, being humble about where you are, and realizing that you will never be an expert on everything.

• It takes years to enter any kind of creative community, including science.

• Human connections are critical to success in science.

Graduate School as a Community

• The graduate school emphasizes apprenticeship with an advisor and belonging to a community.

• It takes four or five years to start from nothing and gain expertise while having your own creativity start to flower.

• Science is more cooperative than competitive. People are always teaching each other something and willing to reveal their ignorance so others can teach them things.

• Science is very international because it sees no barriers. Nationalism is at odds with the way most scientists think about what they're doing here.

Learning French and Language Exploration

In this section, the speaker talks about his experience learning French as a late teen by reading books. He also emphasizes the importance of exploring different languages throughout life for personal growth.

Learning French

• The speaker learned French as a late teen by reading books.

• He found himself on trains in France next to older people who had lived their whole lives there and was able to communicate with them, which was special.

• The speaker embedded himself in French culture, which he found amazing.

Language Exploration

• The speaker encourages language exploration throughout life for personal growth.

• Exploring different languages allows you to express things differently and is one of the great fun things to do in life.

Learning Languages and the Richness of Human Experience

In this section, Michael I. Jordan talks about his interest in learning languages and how it has helped him appreciate the richness of human experience.

Learning Languages

• Michael I. Jordan learned Italian while at MIT because there were many Italians there.

• He goes to China, Asia, and Europe frequently and is amazed by the richness of human experience.

• He loves diversity and embedding himself in other people's experiences.

• Although he did not pursue natural language processing as a career, he admires the field and believes that understanding language is an interesting scientific challenge.

Words of Wisdom from Michael I. Jordan

In this section, Michael I. Jordan shares some words of wisdom from his blog post titled "Artificial Intelligence: The Revolution Hasn't Happened Yet."

Broadening the Scope of AI

• We are witnessing the creation of a new branch of engineering that can be human-centric.

• Let's broaden our scope, tone down the hype, and recognize the serious challenges ahead.

• We have a real opportunity to conceive something historically new - a human-centric engineering discipline.

Conclusion

In this conversation with Lex Friedman, Michael I. Jordan discusses his interest in learning languages and how it has helped him appreciate the richness of human experience. He also shares some words of wisdom from his blog post on artificial intelligence, calling for broadening the scope of AI to create a human-centric engineering discipline.