Þis by Noah Smith May Well Be Better þan Anything I Read in All 2022, & BRIEFLY NOTED
CONDITION: Nearby “Bomb Cyclone”. I would say not “three magics”, ending with large-scale sparse regression & classification, but five—and maybe six: eye-hand-brain, language, writing, gift-exchange..
CONDITION: Nearby “Bomb Cyclone”:
Jill Tucker: Will the Bay Area get hit by the imminent bomb cyclone?: ‘“his one is going to bomb out over the open ocean”, said UCLA climate scientist Daniel Swain, adding he expects it to hit 500 to 800 miles offshore. But the intense system will generate a strong atmospheric river and cold front that will hit the Bay Area and other parts of Northern California…
FOCUS: Þis by Noah Smith May Well Be Better þan Anything I Read in All 2022:
I would say not “three magics” but five—and maybe six:
Noah Smith: The third magic: A meditation on history, science, and AI: More profound and fundamental meta-innovations… these are ways of learning about the world:
The first magic… history… knowledge… recorded in language…. Animals make tools, but they don’t collectively remember…. History… is what allows tinkering to stick….
Our second magic trick… was science… figuring out… principles about how the world works…. Controlled experiments…. It’s pretty incredible that the world actually works that way. If… in the year 1500… [you] told… [someome] one kooky hobbyist rolling little balls down ramps could be right about how the physical world works, when the cumulated experience of millions of human beings around him was wrong… they would have thought you were crazy. They did think that was crazy. And yet, it… worked…. But… complex phenomena have so far defied the approach…. Language, cognition, society, economics, complex ecologies these things so far don’t have any equivalent of Newton’s Laws, and it’s not clear they ever will….
The Third Magic…. “Statistical Modeling: The Two Cultures”… [a] split between… parsimonious models… [and] predictive accuracy…. Control…. Generalize… without finding any simple “law” to intermediate…. Halevy… Norvig, and… Pereira… “The Unreasonable Effectiveness of Data”…. “Represent all the data with a nonparametric model… with very large data sources, the data holds a lot of detail…. Trust… language…. See how far you can go by tying together the words that are already there…. Now go out and gather some data, and see what it can do”…. Underlying regularities that are difficult to summarize but which are still possible to generalize…. Black-box prediction…. We’re always in danger of overfitting and edge cases…. The “third magic” may be more like actual magic than the previous two…. But even wild, occasionally-uncontrollable power is real power….
A number of subfields of economics… have largely resisted the natural experiment approach…. Might we apply AI tools?… Khachiyan et al. argues… “yes”: “Deep learning… daytime satellite[s]… predict changes in income and population… grid cells with lateral dimensions of 1.2km and 2.4km (where the average US county has dimension of 55.6km)…. Model predictions… exceed the accuracy of existing models… are 3-4 times larger than for… nighttime lights…. In-period accuracy… 0.90 to 0.94 for levels predictions and 0.49 to 0.51 for time differences…. Predicting out-of-sample… an R2 of 0.20 for the change in log population over 2010 to 2020…. However, the income model is unable to outperform the true mean (i.e., R2 = 0) when forecasting income changes over 2007 to 2017…. R2 values are 0.50 for the 2000-2020 population change and 0.42 for the 2000-2017 income change (with initial conditions), which are similar to results for the 2000-2010 sample period…”. Being able to predict the economic growth of a few city blocks 10 years into the future with even 30% or 40% accuracy…. And this is just a first-pass attempt…
The first magic is at the individual level: eyes, hands, and brains that allow us, but just to use and make but to plan for tool-use, and so the East African Plains Ape gains fangs, claws, fur, levers, and much else. The second magic is at the small group-collective level: ears and mouths that allow us to talk using language, and so the band of fifty of so East African Plains Apes becomes an anthology intelligence: what one knows, soon all know.
Then comes the third magic, which is Noah’s first: the artificial memory and communication device that is writing, and all of a sudden it is not just the band of fifty that is an anthology intelligence, but potentially all humanity in the present and the past.
The fourth magic is a magic of coordination: How do you get us all in harness, pulling in the same direction, but each doing our (relatively specialized) part? The fourth magic is gift-exchange, that in its hypertrophied form turns into the market economy. Doing favors for one another, in a social context in which there must br approximate balance, is what allows us to, collectively, not just know things but do things.
After that comes, as Noah correctly notes, the fifth magic of science: the experimental method, and the exaltation of ideas not because they solidify the band or help some élite run a force-and-fraud exploitation-and-domination game, but because the ideas are true and are a univeral remote control—they enable us to understand and hence control the universe.
Those five magics have brought us where we are today.
Now how likely is it that we are now at the cusp of a sixth magic: predictive accuracy, generalization, and control without any simple intermediating laws, abstractions, or encapsulations, based simply on the fact that we have a huge amount of data, and can thus classify situations very finely by looking at what the situations’ nearby neighbors are?
On the one hand, I feel that this must be true—I find it very hard to imagine what our brains, or, say, a dogs’ brains, are doing in wetware other than this. On the other hand, our computers are still a lot less sophisticated than our brains, and it is unclear if Moore’s Law will get us to computers of sufficient complexity, and unclear if we can do as good a job of programming our computers as evolution has done at programming us.
But you really do need to subscribe to Noah Smith…
MUST-READ:
I confess I was disappointed that nobody noticed this little microblog I did. Maybe SubStack Chat is not the right venue?:
There is an interesting debate between David Glasner <https://uneasymoney.com/2023/01/01/you-say-potato-i-say-potahto-you-say-tomato-i-say-tomahto-you-say-distribution-i-say-expectation/> and Olivier Blanchard going on:
I think the easiest way to conceptualize what I think of as the major point is to set up a model in which:
The central bank has a target rate of inflation.
The rate of inflation is a constant markdown applied to the rate of nominal wage increase.
The rate of increase of nominal wages that workers are able to demand, and enforce, is a declining function of the unemployment rate and of the real wage.
In this model, there is a warranted rate of nominal wage increase: the central bank’s inflation target, plus the wedge between price inflation and nominal wage increase. In this model, the natural rate of unemployment is the rate at which the actual rate of nominal wage increase is equal to the warranted rate, and thus depends on the level of the real wage—with a lower real wage sustainable only with a higher natural rate of unemployment.
You can understand this as a distributional conflict: total demands for production greater than 100% of supply, and so attaining a feasible equilibrium requires one of:
diminishing other groups’ claims on income and so pushing real wages up enough that nominal wage increases demanded are at the nominal warranted rate of wage increase.
raising the unemployment rate enough that nominal wage increases demanded are at the nominal warranted rate of wage increase.
accepting an inflation rate higher than the central bank’s target, and thus achieving feasibility via the exploitation of workers’ short-run money illusion.
If you take this approach, I think it becomes clear that Blanchard and Glasner are talking past each other, and are not really disagreeing…
BRIEFLY NOTED:
ONE IMAGE: Slouching…:
I continue to try to get stable diffusion to generate a truly striking and apposite Slouching Towards Utopia <bit.ly/3pP3Krk> illustration. But it keeps going horribly wrong, and the things it produces that I like are vastly too dystopian:
ONE VIDEO: Warren Buffett Thinks You Should Not Bet on Meme Stocks:
Very Briefly Noted:
Mary Weitzman: Prices vs. Quantities…
Charlie Sykes: Kevin McCarthy's Hat-Trick of Humiliation: ‘“You're supposed to throw the grenade after you pull the pin. They pulled the pin and passed the grenade around!" —James Carville…
Matt Levine: Private Markets Don’t Like to Go Down: ‘One cynical way to understand private investing generally is that private investment firms — venture capital, private equity, private real estate, etc. — charge their customers high fees for the service of avoiding the visible volatility of public markets…
Patricia Marx: Hell on Two Wheels, Until the E-Bike’s Battery Runs Out: ‘In 2020, Americans bought more than twice as many electric bikes as electric cars. I test-drove a fleet of them and lived to tell the tale—and make recommendations…
Noah Millman: What Do The Republican Renegades Want?: ‘McCarthy…. I can’t see what he could give them the insurgents that would be of comparable value to his scalp…. What they want to prove is that they can veto a choice they don’t approve of, [so] what can he possibly give them?…
Robert Minchin: ‘My hope for 2023 is that this will be the year when people start following the advice of the National Institute of Standards and Technology that: “12 a.m. and 12 p.m. are ambiguous and should not be used”. Everytime you break this rule, a kitten cries…
Alexandra Petri: ‘Little-known fact: if the speaker vote goes to a seventh ballot John Quincy Adams rises from the sea astride a seahorse of enormous dimensions and the winner has to fight him…
Jake Sherman: ‘Interesting moment—Kat Cammack said Democrats have been drinking during the speaker vote. Dems ask to “take down her words”—a mechanism to formally erase her words from the record. There are no rules, so there's no mechanism to do this…
¶s:
Jacob Levy: ‘Cammack's speech is pretty entertaining, and not primarily for her “the House is not in order so I’m allowed to stay stuff I would be forbidden to say under normal rules of decorum”. There’‘ standard pablum about unity and working together— which only applies intra-party, so it alternates sentence by sentence with “and we are united in hating THEM”. And there are appeals to the authority of Ronald Reagan, as if it were 1993 or 2003 or 2013. “Are we Reagan Republicans?” doesn't work in 2023…
Mark Gongloff: Kevin McCarthy falls victim to the GOP rebellion about nothing: ‘The American voting public, in its infinite wisdom, handed the Republican Party the gavel to the House of Representatives. The party took the gavel this week and promptly began smashing itself in the face with it, again and again. Sometimes, when the party’s smashing arm got tired, from all the smashing, it chewed on the gavel. At one point the House sergeant at arms had to pry the gavel’s handle from one of the party’s nostrils. Everything’s going great, in other words…. McCarthy has already done almost everything to appease the extremists short of renaming both of his children “Donald Trump.” Nothing has worked…. Like Robert Redford in “The Candidate,” after it won the House, the GOP more or less said, “So what do we do now?” It campaigned without a plan for governance and forgot to develop one between November and January, Bloomberg’s editorial board writes. Its Rebellion About Nothing is par for the course, really…
Josh Barro: When Does the House Need a Speaker?: ‘Republicans have good reasons to want to get this speaker election mess over with…. But they’re not excellent reasons, and they’re not urgent…. If you think waiting will get you leadership and rules that are more to your liking, and especially if you’re someone like Matt Gaetz whose whole purpose in office is to gain attention, which becomes easier rather than harder with this mess going on. And that’s why I have trouble seeing how this thing ends
Timothy Snyder: Life as a Lie: Trump, Santos, and Putin: ‘After he lost, Trump was lying to extend his political life. It wasn't that he labored under a misapprehension about the election. He knew that he had lost. But he was lying not so much to deny the truth to invite people into an alternative reality…. Trump knew that he had lost the election, and also knew that his also knew that his specific claims of fraud were untrue. And that is all made abundantly clear. Yet there is a deeper point to be made about the nature of politics, which is that it can be transformed by big lies issued from positions of authority.… Big Lies demand violence, since they command the faith of some, but cannot overcome the common sense or lived experience of others. The smaller lies within the Big Lie, by generating distrust of institutions, create a sense that only violence can restore the righteous order of things…. The striving for an all-embracing fiction explains the deep affinity between Trump and Putin…. Putin also tells big lies, for example that Ukraine does not exist, that there is no Ukrainian society, no Ukrainian nation…. Once factual truth is no defense in politics, all that remains is spectacle and force…
m
Henry Farrell, Hugo Mercier, & Melissa Schwartzberg: Analytical Democratic Theory: A Microfoundational Approach: ‘A… literature challenges the quality of democratic decision making, drawing on… specific claims about the ubiquity of cognitive bias…. A competing literature… defends the wisdom of crowds…. We… demonstrate that the former literature is based on outdated and erroneous claims and that the latter is overly sanguine…. “Interactionist” scholarship shows how individual-level biases are not devastating for group problem solving, given appropriate conditions. This provides possible microfoundations for… investigating how different group structures are associated with both success and failure in democratic decision making. This agenda would have implications for both democratic theory and democratic practice…
Marcin Piatkowski: Béla Tomka’s “Austerities and Aspirations: A Comparative History of Growth, Consumption, and Quality of Life in East Central Europe since 1945”: ‘While communist countries lagged behind Western Europe in practically every facet of development, they nonetheless scored much higher on education outcomes, mortality, gender equality and many other key elements of development and well-being than what would be suggested by their levels of income.... By now all four countries... Poland, Czechia, Slovakia and Hungary... [and]] the rest of the CEE region from Estonia down to Bulgaria—are all living through their Golden Age, defined as the shortest distance in history between their levels of incomes and quality of life and those in Western Europe... This is surprising in view of the book’s generally useful coverage of the
Matt Levine (Feb 10, 2021): Now... things are valuable not based on their cash flows but on their proximity to Elon Musk: ‘It was a joke?... You might think that “did Elon Musk tweet about a thing” would be a simpler valuation metric than, like, “estimate its cash flows in perpetuity and apply an appropriate discount rate,” but I don’t know, there’s a lot going on.... Here is a website called “Elon Stocks,” which promises to send you a text “when Elon mentions a stock in a tweet.” (I cannot advise you either on investing strategy, or on the wisdom of typing your phone number into a random website, but I mention it here purely for entertainment purposes.) I assume they are working on a premium high-speed direct feed to alert high-frequency traders to Musk’s tweets microseconds before everyone else. I hope Twitter Inc. is doing that, actually.... You don’t need economics anymore, you need memes, I’m so sorry...
Maciej Ceglowski: Why Not Mars: ‘Going to Mars made sense, back when astronauts were a cheap and lightweight alternative to costly machinery…. No one had been in space long enough to discover the degenerative effects of freefall.... But fifty years of progress in miniaturization and software changed the balance between robots and humans… space probes improved by something like six orders of magnitude, while the technologies of long-duration spaceflight did not.... As for that space station, the jewel of human spaceflight, it exists in a state of nearly perfect teological closure, its only purpose being to teach its creators how to build future spacecraft like it…. The idea of sending something like it on a three year journey to Mars does not get engineers’ hearts racing, at least not in the good way.... When you hold on to a belief so strongly that neither facts nor reason can change it, what you are doing is no longer science, but religion. So I’ve come to believe the best way to look at our Mars program is as a faith-based initiative...
Re Blanchard: In FRED, I graph nonfarm payroll # with core PCE and the only time I see core inflation fall out of trend with employment count is in 2008, when there was a particularly hard recession. In fact the relationship is strongly inverse in 1970's and early 80's. What am I missing?
Because this doesn't look like the relationship to bet the economy on.
> Now how likely is it that we are now at the cusp of a sixth magic: predictive accuracy, generalization, and control without any simple intermediating laws, abstractions, or encapsulations, based simply on the fact that we have a huge amount of data, and can thus classify situations very finely by looking at what the situations’ nearby neighbors are?
I think this is factually untrue and actually looking less likely by the month. Leaving aside things like tricking humans about images and language, having a huge amount of data and classifying situations according to nearby neighbors has shown to be very very *bad* at generalization and control, hype aside [see e.g. https://www.science.org/content/blog-post/alphafold-excitement ]. If nothing else, the plausibility achievements of chatGPT should highlight its absolute lack of concern about its truth value.
The near-future of AI in science IMHO looks more like this: https://arxiv.org/abs/2212.13254 Using ML to augment existing knowledge in areas so complex that all the data we have ever had is probably not enough to even get a good handle on things without a lot of previous theory.
For a Sixth Magic [I like the labels], I'd propose what's currently called causal discovery and automatic experiment design. Basically, for the first time ever we have a _calculus of understandings of the world_ --- we can infer, study, and figure out the next optimal experiment for scientific models far beyond the complexity of what we can currently handle. At the moment no entity in the planet can really put together and take advantage of our current set of models, theories, and data about, say, biochemistry, but we are in the process of building them. But I bet they will look more like "Excel for arbitrary complex theories" than the Azathoths that grab headlines today.