🚀 Faster, Please!: Week in Review+ #3
Talking to AI experts Melanie Mitchell and Erik Brynjolfsson, US recession odds, AI spreading through the economy, and much, much more ...
My free and paid Faster, Please! subscribers: Welcome to Week in Review+. Thank you all for your support! For my free subscribers, please become a paying subscriber today. (Expense a corporate subscription perhaps?)
It was another huge week, and I covered a wide range of subjects in the essays, Q&As, and micro reads on Monday and Thursday (as well as a special, paywall-free issue on Wednesday).
But this weekend issue isn’t just a rewarding recap for everyone. (Although it is that, too.) There’s also fantastic fresh content. I will soon be publishing a lengthy podcast chat I did with computer scientist and AI expert Melanie Mitchell of the Santa Fe Institute. She also wrote the 2020 book Artificial Intelligence: A Guide for Thinking Humans. But all my subscribers — and only my subscribers — get a special Five Quick Questions interview I did with her. (See below) Good stuff. Enjoy!
Melior Mundus
In This Issue
Weekend 5QQ: 5 Quick Questions for … computer scientist and AI expert Melanie Mitchell of the Santa Fe Institute
Essay Highlights: Boomflation, stagflation, or … recession — what are the odds?; Is AI finally ready to supercharge the US economy?
Best of 5QQ: A Culture of Growth author and Northwestern University economic historian Joel Mokyr; Stanford University economist and digital economy expert Erik Brynjolfsson
Best of Micro Reads
Weekend 5QQ
5 Quick Questions for … AI expert Melanie Mitchell of the Santa Fe Institute
1/ If I were reading a news report about AI in 1972, what would I have been reading about?
In 1972, you'd be reading about AI systems that can do mathematics, that have some ability to recognize handwritten letters. There'd be some robots that could do some navigation. They could move around a little bit. You could go see a demo of these programs, but it wasn't hard to cause them to break. They were what people in the field called "brittle," that any very small changes in what you were asking them to do could just stymie them. But that probably wouldn't be in the news report.
2/ And when I read about AI today, what am I reading about?
You're reading about some of the same kinds of things. We're reading about AI systems that can generate text that sounds like human text. People were doing that back in the '70s also. They had chatbots back then. They were much less sophisticated, but they still fooled people. You're reading about AI systems that can recognize faces now. In a crowd, you can pick out a face, although there are biases there. You're reading about progress in machine translation. You're probably reading a lot about self-driving cars today, and now there are a lot of questions about why they aren't working as well as they should be.
3/ Is the approach today just far more advanced or is it a completely different approach?
It's a completely different approach. Because back in the '70s, there were some of the same approaches as we're seeing now, but they were not mainstream AI. They were kind of on the fringe. Back then, people were really focusing on this symbol-processing type of AI, giving it explicit rules in a human-like language or logic-like language. Neural networks and machine learning were not big at all.
4/ We were supposed to have a million autonomous cars on the road two years ago. We don't. What should we learn about AI from that delay? It certainly seems like that technology is harder than some people thought.
There's a lot to learn. One is that whenever you're trying to deploy something in the real world — especially when it involves humans and human social systems, human culture — it's going to be a lot harder than you imagined. It's the fact that things are very unpredictable in the real world, that humans can do unpredictable things. And if you don't have the common sense that we bring to just acting in the world, you're going to have a hard time. You're going to run up against situations you can't deal with. It's the humans that make things hard.
Someone pointed out that there's an analogy with this recent deployment of vaccines. We think the hard part of was creating these mRNA vaccines. "That was the science. That was the hard part." But it turned out the much harder part was convincing people to be vaccinated and dealing with all the supply chain stuff. And all of that, more the human side, turns out to be the harder part of all of these technological advances.
5/ Is there a film, TV, or book depiction of AI that you like or one that you really don't like?
If you go way back to the original Star Trek, they had this onboard computer and they could ask it questions. And it wasn't a robot. It was like an Alexa or something, but it understood all the questions because it could process language and it understood all the things that your Alexa can't understand. I like that, and I think it's something that maybe could be possible to have that's much more likely than something like HAL in 2001. And people in Amazon and Apple and these companies have all said that watching Star Trek as a kid was kind of their inspiration for trying to build these smart assistants.
⭐ Bonus: You used to be a high school math teacher. Based on that experience, do you think that some people are math people and other people are not math people?
I think some people are more inclined towards math. Some people just naturally are better at it and like it, but I think anybody could train themselves to be much better at math than they already are. It's kind of like playing an instrument: There are some people who just naturally really pick it up very fast and are very good at it. And I don't know exactly what qualities that involves, but any of us could learn to play the piano and learn to play it pretty well. Only certain kinds of people can become professional mathematicians and be really brilliant at math. But people say, "Oh, I'm just bad at math." I don't think that's necessarily true. You haven't tried. You haven't practiced. It takes practice like everything else.
Essay Highlights
🎲 Boomflation, stagflation, or … recession — what are the odds?
The ‘70s flashback nature of current events — Afghanistan as Vietnam, Russia as the Soviet Union, inflation as, well, inflation — has encouraged talk about stagflation, even though the economy continues to expand and create gobs of jobs. So more boomflation for now. But Econ 101 suggests that when the price of a good or service rises, consumers tend to demand less of it. If consumers don't keep tolerating rising prices, they'll start spending less on all sorts of things. That means less revenue for businesses, who then start spending and hiring less. On top of that, add in geopolitical/uncertainty/oil shocks from Russia’s invasion of Ukraine and the start of a Federal Reserve tightening cycle. Here are two charts that I dig into:
⚡ Is AI finally ready to supercharge the US economy?
How do we get faster economic and productivity growth over the rest of this decade and beyond? The most obvious and likely way is that recent advances in artificial intelligence will broaden and deepen across the economy leaving almost no sector — biotech, energy, retailing, finance, manufacturing, among others — untouched and untransformed. In other words, AI will finally become an economically significant general-purpose technology in the 2020s, much as factory electrification did (finally) in the 1920s.
But are we making the complementary investments required for the deep diffusion of AI technologies? My search for an answer to that question took me to the new edition of the AI Index Report from the Stanford Institute for Human-Centered AI at Stanford University. It’s full of encouraging data and charts suggesting those investments are being made. Like this chart:
My optimistic take is that based on the sorts of data I found in the AI Index Report, another “AI winter” is not coming anytime soon. Instead, to keep up the Game of Thrones metaphor, the current AI spring will turn into a long, long summer of further technological progress and greater productivity gains. 🤞
Best of 5 Quick Questions
📈 In his fantastic 2016 book A Culture of Growth: The Origins of the Modern Economy, Joel Mokyr, an economic historian at Northwestern University, writes that “the fundamental belief that the human lot can be continuously improved by bettering our understanding of natural phenomena and regularities and the application of this understanding to production has been the cultural breakthrough that made what came after possible.” That Enlightenment way of thinking led to the Scientific Revolution, then the Industrial Revolution, and then created the modern world we know.
Pethokoukis: Have we become more risk averse as a society — lower rate of business formation, less relocation — or are there higher barriers to taking risks?
Mokyr: It does not matter if "society" is risk averse. All you need is a large enough group of people in the "upper tail" of risk-taking distribution who will undertake risky projects, what the rest do does not matter.
🌐 Erik Brynjolfsson is the Jerry Yang and Akiko Yamazaki Professor and Senior Fellow at the Stanford Institute for Human-Centered AI, and Director of the Stanford Digital Economy Lab. He’s also the author of several books, including Machine Platform Crowd: Harnessing our Digital Future (2017) and The Second Machine Age: Work, Progress and Prosperity in a Time of Brilliant Technologies (2014), both of which he co-authored with Andrew McAfee.
Pethokoukis: You engaged in a “long bet” with Robert Gordon about productivity growth. If you lose — if private nonfarm business productivity growth doesn’t “average over 1.8 percent per year from the first quarter (Q1) of 2020 to the last quarter of 2029 (Q4)” — what do you think probably went wrong?
Brynjolfsson: If I end up being wrong about productivity in 2029 it would probably be because I made the same mistake as I made about productivity in the last decade. The technology has progressed as fast or faster than I expected in most areas, but our organizations and skills have consistently been slower to adapt, even after I account for how slow they are to adapt.
Great technology is important, but it is a catalyst for the real changes needed to boost productivity: updating business processes and the skills of the workforce. I’ve been disappointed at how slow companies have been to make the necessary changes. I know it’s not easy, but I worry America just isn’t as dynamic and flexible as it needs to be to take full advantage of AI and the other amazing technologies that are increasingly available.
Best of Micro Reads
🚄 California’s Ambitious High-Speed Rail at a Crossroads - Jill Cowan, The New York Times | California's $105 billion (and still rising?) high-speed rail project, which would connect Los Angeles to San Francisco, has been in the works for over a decade. But what started as a symbol for the Golden State's big ideas has devolved into "an alarming vision of a nation that seems incapable of completing the transformative projects necessary to confront 21st century challenges." Proponents say the project will ease congestion and reduce carbon emissions while promoting California's economic strength, while detractors argue delays and ballooning costs signal the project has, well, gone off the rails. And even the pro-rail camp is splintered, with some calling for a shorter-term priority of running trains on a segment of the track within the Central Valley, while others insist on focusing on the ends of the line first.
🧠 AlphaFold, GPT-3 and How to Augment Intelligence with AI - Niko Grupen, a16z Future | With recent advances in artificial intelligence accompanied by fears of AI replacing human intellect, it's worth considering how machines can aid and augment our own thinking powers. Here are three possible AI applications from the piece: GPT-3, an AI language model, can boost search engine performance and assist with research by summarizing and categorizing text; a "GitHub-like service for protein design" would allow "experimentalists to store the diffs of protein candidates and any experimental results that exist for them;” and AI-powered writing tools could enable us to modify the style of a sentence the way we switch between fonts today.