Astral Codex Ten

UC Berkeley

14 min read

UC Berkeley

Article

UC Berkeley is a recurring organization in the Astral Codex Ten archive, appearing 6 times across 6 issues between August 27, 2021 and October 13, 2022. The archive places it in contexts such as “near the UC Berkeley parking lot”; “a couple of researchers from UC Berkeley”; “So do Stanford, Cambridge, UC Berkeley, etc”. It most often appears alongside AGI, Astralcodexten Com, Bay Area.

Metadata

  • Category: Organizations
  • Mention count: 6
  • Issue count: 6
  • First seen: August 27, 2021
  • Last seen: October 13, 2022

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Berkeley Meetup This Saturday
August 27, 2021 · Original source
Where: deflection.jump.puppy, aka the lawn where West Circle meets Free Speech Bikeway near the UC Berkeley parking lot.
Inline links: deflection.jump.puppy
On Hreha On Behavioral Economics
August 30, 2021 · Original source
Somewhere in this process, they did an experiment where they gave participants a quarter minted in Denver and asked them if they wanted to exchange it for a quarter minted in Philadelphia. 60% of people very reasonably didn’t care, but another 35% had grown attached to their Denver quarter, with only 5% actively seeking the novelty of Philadelphia. Psychology is weird. I understand why some people would summarize this paper as “loss aversion doesn’t exist”. But it’s very different from “power posing doesn’t exist” or “stereotype threat doesn’t exist”, where it was found that the effect people were trying to study just didn’t happen, and all the studies saying it did were because of p-hacking or publication bias or something. People are very often averse to losses. This paper just argues that this isn’t caused by a specific “loss aversion” force. It’s caused by other forces which are not exactly loss aversion. We could compare it to centrifugal force in physics: real, but not fundamental. Also, you can’t use this paper to argue that “behavioral economics is dead”. At best, the paper proves that loss aversion is better explained by other behavioral economic concepts. But you can’t get rid of behavioral econ entirely! The stuff you have to explain is still there! It’s just a question of which parts of behavioral econ you use to explain it. Complicating this even further is Mrkva et al, Loss Aversion Has Moderators, But Reports Of Its Death Are Greatly Exaggerated (h/t Alex Imas, who has a great Twitter thread about this). This is an even newer paper, 2019, which argues that Gal and Rucker are wrong, and loss aversion does have an independent existence as a real force. There are many things to like about this paper. Previous criticisms of loss aversion argue that most experiments are performed on undergrads, who are so poor that even small amounts of money might have unusual emotional meaning. Mrkva collects a sample of thousands of millionaires (!) and demonstrates that they show loss aversion for sums of money as small as $20. On the other hand, I’m not sure they’re quite as careful as G&R at ruling out every other possible bias (although I don’t have a great understanding of where the borders between biases are and I can’t say this for sure). The main point I want to make is that all the scientists in this debate seem smart, thoughtful, and impressive. This isn’t like social priming experiments where one person says a crazy thing, nobody ever replicates it at scale, and as soon as someone tries the whole thing collapses. These have been replicated hundreds of times, with the remaining arguments being complicated semantic and philosophical ones about how to distinguish one theory from a very slightly different theory. If that takes replicating your result on a sample of thousands of millionaires, people will gather a sample of thousands of millionaires and get busy on the replication. Just overall really impressive work. I don’t feel qualified to take a side in the G&R vs. Mkrva debate, but both teams make me really happy that there are smart and careful people considering these questions. And this is just a drop in the bucket. Alex Imas also links Replicating patterns of prospect theory for decision under risk, which says: Though substantial evidence supports prospect theory, many presumed canonical theories have drawn scrutiny for recent replication failures. In response, we directly test the original methods in a multinational study (n = 4,098 participants, 19 countries, 13 languages), adjusting only for current and local currencies while requiring all participants to respond to all items. The results replicated for 94% of items, with some attenuation. Twelve of 13 theoretical contrasts replicated, with 100% replication in some countries. Heterogeneity between countries and intra-individual variation highlight meaningful avenues for future theorizing and applications. We conclude that the empirical foundations for prospect theory replicate beyond any reasonable thresholds. Beyond any reasonable thresholds! IV. Do Nudges Work? or, How Small Is Small? Continuing through the Hreha article: For a number of years, I've been beating the anti-nudge drum. Since 2011, I've been running behavioral experiments in the wild, and have always been struck by how weak nudges tend to be. In my experience, nudges usually fail to have *any* recognizable impact at all. This is supported by a paper that was recently published by a couple of researchers from UC Berkeley. They looked at the results of 126 randomized controlled trials run by two "nudge units" here in the United States. I want you to guess how large of an impact these nudges had on average... 30%? 20%? 10%? 5%? 3%? 1.5%? 1%? 0%? If you said 1.5%, you'd be right (the actual number is 1.4%, but if I had written that out you would have chosen it because of its specificity). According to the academic papers these nudges were based upon, these nudges should have had an average impact of 8.7%. But, as you probably understand by now, behavioral economics is not a particularly trustworthy field. I actually emailed the authors of this paper, and they thought the ~1% effect size of these interventions was something to be applauded—especially if the intervention was cheap & easy. Unfortunately, no intervention is truly cheap or easy. Every single intervention requires, at the very minimum, administrative overhead. If you're going to do something, you need someone (or some system) to implement and keep track of it. If an intervention is only going to get you a 1% improvement, it's probably not even worth it. Uber infamously had a team of behavioral economists working on its product, trying to “nudge” people in the right direction. Relatedly, Uber makes $10 billion in yearly revenue. If they can “nudge” people to spend 1% more, that’s $100 million. That’s not much relative to revenue, but it’s a lot in absolute terms. In particular, it pays the salary of a lot of behavioral economists. If you can hire 10 behavioral economists for $100,000 a year and make $100 million, that’s $99 million in profit. Or what if you’re a government agency, trying to nudge people to do prosocial things? There are about 90 million eligible Americans who haven’t gotten their COVID vaccine, and although some of them are hard-core conspiracy theorists, others are just lazy or nervous or feel safe already. (source) Whoever decided on that grocery gift card scheme was nudging, whether or not they have an economics degree - and apparently they were pretty good at it. If some sort of behavioral econ campaign can convince 1.5% of those 90 million Americans to get their vaccines, that’s 1.4 million more vaccinations and, under reasonable assumptions, maybe a few thousand lives saved. Hreha says that: Every single intervention requires, at the very minimum, administrative overhead. If you're going to do something, you need someone (or some system) to implement and keep track of it. If an intervention is only going to get you a 1% improvement, it's probably not even worth it. This depends on scale! 1% of a small number isn’t worth it! 1% of a big number is very worth it, especially if that big number is a number of lives! A few caveats. First, a small number only matters if it’s real. It’s very easy to get spurious small effects, so much so that any time you see a small effect you should wonder if it’s real. I’m ready to be forgiving here because behavioral economics is so well-replicated and common-sensically true, but I wouldn’t blame anyone who steers clear. Second, Hreha says: To be honest, you can probably use your creativity to brainstorm an idea that will get you a 3-4% minimum gain, no behavioral economics "science" required. Which leads me to the final point I'd like to make: rules and generalizations are overrated. The reason that fields like behavioral economics are so seductive is because they promise people easy, cookie-cutter solutions to complicated problems. Figuring out how to increase sales of your product is hard. You need to figure out which variables are responsible for the lackluster interest. Is the price the issue? Is the product too hard to use? Is the design tacky? Is the sales organization incompetent? Is the refund/return policy lacking? etc. Exploring these questions can take months (or years) of hard work, and there's no guarantee that you'll succeed. If, however, a behavioral economist tells you that there are nudges that will increase your sales by 10%, 20%, or 30% without much effort on your part... Whoa. That's pretty cool. It's salvation. Thus, it's no surprise that governments and companies have spent hundreds of millions of dollars on behavioral "nudge" units. Unfortunately, as we've seen, these nudges are woefully ineffective. Specific problems require specific solutions. They don't require boilerplate solutions based on general principles that someone discovered by studying a bunch of 19 year old college students. However, the social sciences have done a good job of convincing people that general principles are better solutions for problems than creative, situation-specific solutions. In my experience, creative solutions that are tailor-made for the situation at hand *always* perform better than generic solutions based on one study or another. Hreha is a professional in this field, so presumably he’s right. Still, compare to medicine. A thoughtful doctor who tailors treatment to a particular patient sounds better (and is better) than one who says “Depression? Take this one all-purpose depression treatment which is the first thing I saw when I typed ‘depression’ into UpToDate”. But you still need medical journals. Having some idea of general-purpose laws is what gives the people making creative solutions something to build upon. (also, at some point your customers might want to check your creative solution to see whether it actually gives a “3-4% minimum gain, no behavioral economics required”, and that would be at least vaguely study-shaped.) Third, everyone who said nudging had vast effects is still bad and wrong. Many of them were bad and wrong and making fortunes consulting for companies about how to implement the policies they were claiming were super-powerful. This is suspicious and we should lower our opinion of them accordingly. In a previous discussion of growth mindset, I wrote: Imagine I claimed our next-door neighbor was a billionaire oil sheik who kept thousands of boxes of gold and diamonds hidden in his basement. Later we meet the neighbor, and he is the manager of a small bookstore and has a salary 10% above the US average... Should we describe this as “we have confirmed the Wealthy Neighbor Hypothesis, though the effect size was smaller than expected”? Or as “I made up a completely crazy story, and in unrelated news there was an irrelevant deviation from literally-zero in the same space”? All the people talking about oil sheiks deserve to get asked some really uncomfortable questions. And a lot of these will be the most famous researchers - the Dan Arielys of the world - because of course the people who successfully hyped their results a lot are the ones the public knows about. Still, the neighbor seems like a neat guy, and maybe he’ll give you a job at his bookstore. V. Conclusion: Musings On The Identifiable Victim Effect I actually skipped the very beginning of Hreha’s article. I want to come back to it now. It begins: The last few years have been particularly bad for behavioral economics. A number of frequently cited findings have failed to replicate. Here are a couple of high profile examples: The Identifiable Victim Effect (featured in the workbooks I wrote with Dan Ariely and Kristen Berman in 2014)
Inline links: Loss Aversion Has Moderators, But Reports Of Its Death Are Greatly Exaggerated, Alex Imas, Replicating patterns of prospect theory for decision under risk, They looked at the results of 126 randomized controlled trials, others, https://substackcdn.com/image/fetch/$s_!1gY_!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2F5b127d7c-045a-4266-b515-a0e9201ce651_571x382.png, a previous discussion of growth mindset, the workbooks I wrote with Dan Ariely and Kristen Berman in 2014
Practically-A-Book Review: Yudkowsky Contra Ngo On Agents
January 19, 2022 · Original source
The story thus far: AI safety, which started as the hobbyhorse of a few weird transhumanists in the early 2000s, has grown into a medium-sized respectable field. OpenAI, the people responsible for GPT-3 and other marvels, have a safety team. So do DeepMind, the people responsible for AlphaGo, AlphaFold, and AlphaWorldConquest (last one as yet unreleased). So do Stanford, Cambridge, UC Berkeley, etc, etc. Thanks to donations from people like Elon Musk and Dustin Moskowitz, everyone involved is contentedly flush with cash. They all report making slow but encouraging progress.
ACX Grants ++: The Second Half
February 10, 2022 · Original source
#83: Detect And Fight Healthcare Fraud Our company is using data to detect fraud against the government. Access to quality healthcare is dwindling in the United States. There is an estimated hundred billion dollars in fraud every year leading to lower standards of care and making healthcare unaffordable. We’re seeking a hundred thousand dollars to buy data from the Centers for Medicare and Medicaid services. This will allow us to find fraud and file lawsuits on behalf of the government. The Department of Justice signaled a new level of support for independent companies using data methods to identify fraud in June of last year when they picked up a case brought by Integra Med Analytics. For the past twelve months we’ve been working with attorneys specializing in this area (qui tam). We’ve been consolidating data returned from broad FOIA requests and begun assisting law firms with data science. Our team combines broad technical expertise (Google, NASA, LANL, NIST, UC Berkeley) with business acumen and investigative experience. The three of us have been working together on projects with positive externalities for five years. Previous successful projects include providing flexible housing, and a micro-targeting methods for political action. [Contact erbahr@gmail.com if you can help]
Universe-Hopping Through Substack
September 29, 2022 · Original source
Both designers who graduated from the prestigious Rhode Island School of Design, the pair created fictitious cereals called “Obama O's, the Cereal of Change,” and “Cap’n McCain's, a Maverick in Every Box.” They designed the box artwork themselves and convinced a student at UC Berkeley to print 500 units of each box on the cheap. The boxes were delivered as flat rectangles that had to be cut and assembled by hand.
Highlights From The Comments On The Central Valley
October 13, 2022 · Original source
"Severe drought... partly [from] California diverting water to hydrate growing coastal cities." Coastal urban water demand has almost nothing to do with agricultural water shortages. Going to self-cite on this one (laziness; PhD in water resources engineering; married to a UC Berkeley water resources economist). Lack of winter rain/snow and associated land fallowing in the Central Valley almost definitely impacts agricultural labor demand.