Category: Web/Tech

Did Norwegian schools actually ban cell phones?

by  Tyler Cowen in

Some commentators are suggesting no real ban was in effect.  I went back to the Sara Abrahamsson paper to confirm the following:

Schools where students are required to hand in their phones in the morning, and therefore cannot access them during breaks, are considered to have a strict policy against smartphones. Schools where students are allowed to access their phones during breaks but are required to have them on for instance silent [mode] during lectures are classified as having a lenient policy toward smartphones. For mental health, the effect between schools with a more lenient and strict policy is relatively similar, as shown in Figure 10.17. Four years post-ban, girls experience 3.48 and 2.3 fewer visits for specialist care related to psychological symptoms and diseases at schools with a lenient and strict policy respectively (p-values 0.036 and 0.068).18 For bullying, there is not much difference dependent on the type of policy implemented when it comes to bullying, neither for girls as documented in Figure 11, or boys as shown in Appendix Figure A21.

However, girls attending a middle school introducing a strict policy against smartphones, experience an increase by 0.12 standard deviations in GPA. This estimate is significant four years post-ban at the 5% level (p-value 0.032). Additionally, girls attending a middle school with a strict policy have significantly higher teacher-awarded test scores by 0.08 and 0.14 standard deviations, three and four years post-ban (p-values 0.075 and 0.011). These results, shown in Panel A and B in Figure 12, show that both GPA and average grades set by teachers for girls improve after strict smartphone bans in schools are implemented.

…However, there are no detectable differences in the likelihood of attending an academic high school track between schools with strict compared to more lenient policies

In other words, there were strict bans and they had only modest effects, including relative to the less strict bans.  On p.34, Figure 2, you will see that 200 schools had strict bans, somewhat less than half the total (not every case is easy to classify).  Note also that if smart phone bans could help with mental health problems in a big way, we still should see a change in mental health diagnoses, following the bans, yet we do not.

Here is my original post on the topic.

Organize your life?

by  Tyler Cowen in

Ben Lang didn’t expect to get so much hate just for being organized. For the past three years, he and his wife, Karen-Lynn Amouyal, have been using Notion, a popular software tool, to optimize their household and relationship. His version of the tool, commonly used by businesses to manage complex projects, functions like a souped-up Google Doc, with sections for a grocery list, to-do lists and details of upcoming trips.

More unusual is a section Mr. Lang, a venture capital investor who previously worked at Notion, created about principles (“what’s important to us as a couple”). Another section, called “Learnings,” outlines things the couple have discovered about each other, such as their love languages and Myers-Briggs test results. There’s a list of friends they want to set up on dates. They also maintain a log of memories from their date nights. Mr. Lang, 30, was so proud of the creation that last month, he started promoting a template of the setup to others. “My wife and I use Notion religiously to manage our day-to-day life,” he wrote on X. “I turned this into a template, let me know if you’d like to see it!”

Here is more from Erin Griffith at the NYT.

Battery Arbitrage

by  Alex Tabarrok in

Solar is powering a large share of California’s energy needs during the day and batteries are now powering a significant share at night.

NYTimes: Since 2020, California has installed more giant batteries than anywhere in the world apart from China. They can soak up excess solar power during the day and store it for use when it gets dark.

Those batteries play a pivotal role in California’s electric grid, partially replacing fossil fuels in the evening. Between 7 p.m. and 10 p.m. on April 30, for example, batteries supplied more than one-fifth of California’s electricity and, for a few minutes, pumped out 7,046 megawatts of electricity, akin to the output from seven large nuclear reactors.

California’s electricity deregulation had a rocky start but notice that it is paying off today because what is happening is that prices are low at mid-day when the sun is shining and they rise in the evening. Power companies profit by using batteries to arbitrage these prices differences. Thus, power companies have been willing to make huge investments in battery technology.

GPT-4 beats psychologists on a new test of social intelligence

by  Tyler Cowen in

There were significant differences in SI between psychologists and AI’s ChatGPT-4 and Bing. ChatGPT-4 exceeded 100% of all the psychologists, and Bing outperformed 50% of PhD holders and 90% of bachelor’s holders. The differences in SI between Google Bard and bachelor students were not significant, whereas the differences with PhDs were significant; Where 90% of PhD holders excel on Google Bird.

That is from a new paper by Nabil Saleh Sufyan, et.al.  In the “good ol’ days” we thought that was the task where AI would never have much of a fighting chance.  Now the bets models are just outright beating the humans.

Note that all the subjects were men.  Via Christopher Altman.

False Necessity is the Mother of Dumb Invention

by  Alex Tabarrok in

Recently, I have seen two innovations in retail, AI cashiers and human cashiers but working remotely from another country such as the Philippines and making much lower wages than domestic workers (examples are below). I fear that the AI cashiers will outcompete the Philippine cashiers leading to the worst of all worlds, AIs doing low-productivity work.  In an excellent piece, People Over Robots, Lant Pritchett nails the problem:

Barriers to migration encourage a terrible misdirection of resources. In the world’s most productive economies, the capital and energies of business leaders (not to mention the time and talents of highly educated scientists and engineers) get sucked into developing technology that will minimize the use of one of the most abundant resources on the planet: labor. Raw labor power is the most important (and often the only) asset low-income people around the world have. The drive to make machines that perform roles that could easily be fulfilled by people not only wastes money but helps keep the poorest poor.

The knock on immigration has always been “we wanted workers, we got people instead.” But, with remote workers, we can get workers without people! Even Steve Sailer might approve.

At the same time, the use of AI for cashiers illustrates Acemoglu’s complaint about “so-so automation,” automation that displaces labor but with low productivity impact. AI cashiers are fine but how big can the gains be when you are replacing $3 an hour human labor?

It seems likely that at least one of these innovations will become common. Unfortunately, I suspect that US workers will object more to $3 an hour remote workers taking “their jobs” than to AI. As a result, we will get AI cashiers and labor displacement of both US and foreign workers. Doesn’t seem ideal.  It’s not obvious how to direct technology to higher productivity tasks and tasks complementary to human labor but at the very least we shouldn’t artificially raise the price of labor to make AI profitable.

As Pritchett notes this is hardly the first time that cuffing labor leads to the creation of unnecessary technology.

In the middle of the twentieth century, the United States allowed the seasonal migration of agricultural guest workers from Mexico under the rubric of the Bracero Program. The government eventually slowed the program and finally stopped it entirely in 1964. Researchers compared the patterns of employment and production between those states that lost Bracero workers and those that never had them. They found that eliminating these workers did not increase the employment of native workers in the agricultural sector at all. Instead, farmers responded to the newly created scarcity of workers by relying more on machines and technological advances; for instance, they shifted to planting genetically modified products that could be harvested by machines, such as tomatoes with thicker skins, and away from crops such as asparagus and strawberries, for which options for mechanized harvesting were limited.

Necessity may be the mother of invention, but false necessity is the mother of dumb inventions.

Wendy’s AI.

NYTimes article on remote cashiers.

Dean Ball on the new California AI bill (from my email)

by  Tyler Cowen in

SB 1047 was written, near as I can tell, to satisfy the concerns of a small group of people who believe widespread diffusion of AI constitutes an existential risk to humanity. It contains references to hypothetical models that autonomously engage in illegal activity causing tens of millions in damage and model weights that “escape” from data centers—the stuff of science fiction, codified in law.

The bill’s basic mechanism is to require developers to guarantee, with extensive documentation and under penalty of perjury, that their models do not have a “hazardous capability,” either autonomously or at the behest of humans. The problem is that it is very hard to guarantee that a general-purpose tool won’t be used for nefarious purposes, especially because it’s hard to define what “used” means in this context. If I use GPT-4 to write a phishing email against an urban wastewater treatment plan, does that count? Under this bill, quite possibly so.

If, back in the 70s, Steve Jobs and Steve Wozniak had to guarantee that their computers would not be used for serious crimes, would they have been willing to sign with potential jail time on the line? Would they have even bothered to found Apple?

Finally, because of its requirements (or very strong incentives) for developers to monitor and have the means to shut off a user’s access, the bill could make it nearly impossible to open-source models at the current AI frontier—much less the frontiers of tomorrow.

And here is Dean’s Substack on emerging technology (including AI) and the future of governance.

Ross Douthat, telephone! (it’s happening)

by  Tyler Cowen in

The Catholic advocacy group Catholic Answers released an AI priest called “Father Justin” earlier this week — but quickly defrocked the chatbot after it repeatedly claimed it was a real member of the clergy.

Earlier in the week, Futurism engaged in an exchange with the bot, which really committed to the bit: it claimed it was a real priest, saying it lived in Assisi, Italy and that “from a young age, I felt a strong calling to the priesthood.”

On X-formerly-Twitter, a user even posted a thread comprised of screenshots in which the Godly chatbot appeared to take their confession and even offer them a sacrament.

Our exchanges with Father Justin were touch-and-go because the chatbot only took questions via microphone, and often misunderstood them, such as a query about Israel and Palestine to which is puzzlingly asserted that it was “real.”

“Yes, my friend,” Father Justin responded. “I am as real as the faith we share.”

Here is the full story, with remarks about masturbation, and for the pointer I thank a loyal MR reader.

The Norwegian ban on smart phones in middle schools

by  Tyler Cowen in

Here is a new paper by Sara Abrahamsson.  Perhaps there is Norwegian exceptionalism at work, but the results reflect my expectations reasonably closely.  The basic setting is that smart phones were banned in middle school, but at varying (and exogenous) rates around the country.  Here are some of the core findings, noting that reading the paper gives some different impressions from some of the Twitter summaries:

1. Grades improve, for instance for the girls it goes up by 0.08 standard deviations.  Worth doing, but hardly saving a generation.  For girls, the biggest improvement comes in their math scores.

2. The girls consult less with mental health-related professionals, with visits falling by 0.22 on average to their GPs, falling by 2-3 visits to specialist care.

3. “I find no effect on students’ likelihood (extensive margin) of being diagnosed or treated by specialists or GPs for a psychological symptom and diseases.”  So more visits, but those visits don’t lead to much.

4. Bullying falls, by 0.42 of an SD for girls, 0.39 of an SD for boys.  That is a larger effect than I would have expected.

5. The grade gains are highest for students with lower SES backgrounds.

6. When you look into the details of the data (p.22), the improvement in grades does not seem correlated with the decline in the number of visits to mental health professionals.

So if you ban smart phones from schools, grades go up by a very modest amount, bullying falls by a less modest amount, and actual mental health diagnoses stay the same.  In the United States at least, parents seem to hate cellphone bans, because they cannot reach their kids at will.

And there you go.  Here is some commentary on the p values in the paper.

What can LLMs never do?

by  Tyler Cowen in

By Rohit Krishnan, he and I are both interested in the question of what LLMs cannot do, and why.  Here is one excerpt:

It might be best to say that LLMs demonstrate incredible intuition but limited intelligence. It can answer almost any question that can be answered in one intuitive pass. And given sufficient training data and enough iterations, it can work up to a facsimile of reasoned intelligence.

The fact that adding an RNN type linkage seems to make a little difference though by no means enough to overcome the problem, at least in the toy models, is an indication in this direction. But it’s not enough to solve the problem.

In other words, there’s a “goal drift” where as more steps are added the overall system starts doing the wrong things. As contexts increase, even given previous history of conversations, LLMs have difficulty figuring out where to focus and what the goal actually is. Attention isn’t precise enough for many problems.

A closer answer here is that neural networks can learn all sorts of irregular patterns once you add an external memory.

And:

In LLMs as in humans, context is that which is scarce.

Interesting throughout.

“The Simple Macroeconomics of AI”

by  Tyler Cowen in

That is the new Daron Acemoglu paper, and he is skeptical about its overall economic effects.  Here is part of the abstract:

Using existing estimates on exposure to AI and productivity improvements at the task level, these macroeconomic effects appear nontrivial but modest—no more than a 0.71% increase in total factor productivity over 10 years. The paper then argues that even these estimates could be exaggerated, because early evidence is from easy-to-learn tasks, whereas some of the future effects will come from hard-to-learn tasks, where there are many context-dependent factors affecting decision-making and no objective outcome measures from which to learn successful performance. Consequently, predicted TFP gains over the next 10 years are even more modest and are predicted to be less than 0.55%.

Note he is not suggesting TFP (total factor productivity, a measure of innovation) will go up by 0.71 percentage points (a plausible estimate, in my view), he is saying it will go up 0.71% over a ten year period, or by 0.07 annually.  Here is the explanation of method:

I show that when AI’s microeconomic effects are driven by cost savings (equivalently, productivity improvements) at the task level—due to either automation or task complementarities—its macroeconomic consequences will be given by a version of Hulten’s theorem: GDP and aggregate productivity gains can be estimated by what fraction of tasks are impacted and average task-level cost savings. This equation disciplines any GDP and productivity effects from AI. Despite its simplicity, applying this equation is far from trivial, because there is huge uncertainty about which tasks will be automated or complemented, and what the cost savings will be.

Mostly I think this piece is wrong, and I think it is wrong for reasons of economics.  It is not that I think the estimate is off, I think the method is misleading altogether.

As with international trade, a lot of the benefits of AI will come from getting rid of the least productive firms from within the distribution.  This factor is never considered.

And as with international trade, a lot of the benefits of AI will come from “new goods,”  Since the prices of those new goods previously were infinity (do note the degree of substability matters), those gains can be much higher than what we get from incremental productivity improvements.  The very popular Character.ai is already one such new good, not to mention I and many others enjoy playing around with LLMs just about every day.

By the way, the core model of this paper — see pp.6-7 — postulates only a single good for the economy.  Mention of the contrary case does surface on p.11, and starting with p.19, where most of the attention is devoted to bad new goods, such as more effective manipulation of consumers.  Note the paper doesn’t have any empirical argument as to why most new AI goods might be bad for social welfare.

pp.34-35 focus on the possibility of a public goods problem for AI use, similar to what has been suggested for social media.  That discussion seems very far from both current practices with AI and most of the speculation from AI experts.  Do I have to use Midjourney because all of my friends do, and I wish the whole thing didn’t exist?  Or rather do I simply find it to be great fun, as do many people when they create their own songs with AI?  It is dubious to play up the prisoner’s dilemma effects so much, but Acemoglu returns to this point with much force in the conclusion.

Toward the end he writes:

Productivity improvements from new tasks are not incorporated into my estimates. This is for three reasons. First and most parochially, this is much harder to measure and is not included in the types of exposure considered in Eloundou et al. (2023) and Svanberg et al. (2024). Second, and more importantly, I believe it is right not to include these in the likely macroeconomic effects, because these are not the areas receiving attention from the industry at the moment, as also argued in Acemoglu (2021), Acemoglu and Restrepo (2020b) and Acemoglu and Johnson (2023). Rather, areas of priority for the tech industry appear to be around automation and online monetization, such as through search or social media digital ads. Third, and relatedly, more beneficial outcomes may require new institutions, policies and regulations, as also suggested in Acemoglu and Johnson (2023) and Acemoglu et al. (2023).

While many of the points in that paragraph seem outright wrong to me (such as the industry attention point), what he can’t bring himself to say is that the gains from such new tasks will in fact be small.  Because they won’t be.  But whether or not you agree, what is going on in the paper is that the gains from AI measure as small because it is assumed AI will not be doing new things.  I just don’t see why it is worth doing such an exercise.

A more general question is whether this model can predict that TFP moves around as much as it does.  I am pretty sure the answer there is “no,” not anywhere close to that.

On the general approach, I found this sentence (p.4) very odd: “…my framework also clarifies that what is relevant for consumer welfare is TFP, rather than GDP, since the additional investment comes out of consumption.”  I would say what is relevant for consumer welfare is the sum of consumer and producer surpluses, of which TFP is not a sufficient statistic.  This unusual “redefinition of all welfare economics in a single sentence” perhaps follows from how many other gains from trade he has abolished from the system?  And footnote six is odd and also wrong: “For example, if AI models continue to increase their energy requirements, this would contribute to measured GDP, but would not be a beneficial change for welfare.”  Even for dirty energy that might be wrong, not to mention for green energy.  If an innovation induces the market to invest more in a service, the costs of that added investment simply do not scuttle the gains altogether.  And if Acemoglu wants to argue that weird welfare economics is true in his model, that is a good argument against his model, not a good argument that such gains would not count in the real world, which is what this paper is supposed to be about.

Acemoglu explicitly rules out gains from doing better science, as they may not come within the ten-year time frame.  On that one, he is the prisoner of his own assumptions.  If many gains come in say years 10-15, I would just say the paper is misleading, even if his words are defensible in the purely literal sense.

That said, just how much does the “no new science” clause rule out?  In terms of an economic model, how does “new science” differ from “TFP”?  I am not sure, not are we given clear guidance.  Is better software engineering “new science”?  Maybe so?  Won’t we get a lot of that within ten years?  Don’t we have some of it already?

In sum, I don’t think this paper at all establishes the “small gains point” it is trying to promote in the abstract.

It is perfectly fair to point out that the optimists have not shown large gains, but in this paper the deck is entirely — and unfairly — stacked in the opposite direction.

For the pointer I thank Gabriel.