Whether AI assistance builds or erodes professional expertise is unsettled. In a pre-registered three-month randomized controlled trial, we gave 133 practicing patent lawyers at eleven U.S. intellectual property law firms access to a custom AI drafting assistant and measured both their performance while using AI and their professional judgment afterward without it. All work was scored by blinded expert patent attorneys. Paralleling findings from other white-collar domains, AI access raised the quality of work delivered on benchmark patent drafting tasks at 10 days (0.34 SD, p = 0.03) and 90 days (0.38 SD, p = 0.01), with larger gains among junior lawyers. After three months, all subjects redlined an existing patent application without AI, a core task of patent practice requiring expert judgment. Treated lawyers outperformed controls by 0.32 SD (p = 0.04), but this advantage was concentrated entirely among senior lawyers (0.45 SD, p = 0.02). Junior lawyers showed no average gain; their scores instead bifurcated, with sharply fewer mediocre scores offset by more poor and more good ones. The largest gains from AI thus accrued to the lawyers who retained the least. Foundational expertise may be a prerequisite for extracting durable skill from AI-assisted practice.

That is by David Autor, et.al.  Do note that over time the allocation of humans to tasks will evolve so that more of the humans become more productive, not less.  RCTs somehow have the odd disadvantage of requiring too many things to be held constant, and so they can miss the benefits of longer-term adjustments.

The post Does AI assistance enhance or erode expertise? appeared first on Marginal REVOLUTION.

2. The economists who work on “Productivity and Innovation” are most likely to use AI for their writing.

3. Michael Levin defends neo-neo-Platonism?

4. Look for and try to limit agent collusion.

5. Do orangutans like Indian classical music?

6. “Our results demonstrate that LLM-evolved constitutions significantly outperform both human-designed principles and one-shot LLM-generated rules for multi-agent coordination.“

The post Sunday assorted links appeared first on Marginal REVOLUTION.

Recent years have witnessed many strong claims that ‘diversity’ leads to more original and impactful science, which is a science-focused form of what we call “The Diversity Hypothesis.” However, what evidence supports the claim that diversity enhances scientific output or impact? This pre-registered rapid scoping review seeks to collate and evaluate the scientific evidence for the Diversity Hypothesis…

…Based on over 100 scientific articles, we find that only between 15% and 28% of results reported in the literature are consistent with the hypothesis, with the balance of the results not being consistent with it.

…Overall, the results of the analysis section indicate that there is little empirical evidence that diversity improves scientific output and/or impact. In fact, with the possible exception of disciplinary diversity—a type of informational or viewpoint diversity operationalized at the team level—the majority of the evidence seems to point in the other direction. These findings are robust regardless of how the data is parsed and analyzed. The same conclusions can be drawn looking at the full data set, only the population-adjusted results, or only the results of high quality based on the MMAT analysis.

Hat tip: Colin Wright.

The post Diversity Is Our Strength? appeared first on Marginal REVOLUTION.

1. Intelligence.  Yes, formal smarts.  Playing chess, proving math theorems, and doing well on evals.  Don’t forget humans can do those things too, though AIs are now a huge boost here.

2. Polanyi knowledge.  Michael Polanyi, that is.  This refers to knowledge of time and place, inarticulable knowledge, custom and habit, and many other particularities that you can read about in Hayek and Polanyi and in many other social scientists, anthropologists too.

Humans specialize in this.  The AIs can aid in its production, but at least so far there is no way you can “bring an AI into your office and have it figure out how that office works.”  At least not in the human rather than the purely mechanistic sense.

In the model, intelligence and Polanyi knowledge combine to produce output.

Substitutability is fairly limited.  For instance, if you have problems of norms in your office, a mere dose of AI-drenched technocratic knowledge does not usually solve those problems.  Sometimes it even can make those problems worse, by empowering rent-seekers further.

Intelligence and Polanyi knowledge are not quite Leontief complements, but they are mostly complements.

Now recently the U.S. economy has experienced a huge positive shock to its Intelligence, with more to come.

The core prediction is that this increases marginal returns, employment, and real wages in the Polanyi knowledge sector.  All of a sudden, the inputs into that sector are relatively scarce, compared to the now-larger quantity of Intelligence.

There will also be some transitional unemployment in the Intelligence sector, at least once Centaur models fade.  But so far Centaur models are holding, for instance mathematicians did the prompting to do the new math work.  Nonetheless some of these Centaur employments will fade, just as they have in chess.

Note that the Polanyi sector cannot be boosted very quickly or with direct and simple efficacy.  It is messy by its nature, to cite a term from Luis Garicano.  So the wage and employment gains there are slow in coming.  But they keep on coming for a long period of time.  There are further AI/Intelligence advances on tap, plus absorbing the advances to date, and exploiting them, takes a long time.

In this model, if someone or something could “commandeer” the Intelligence sector, their power over society would be much more limited than it might appear at first.  The world does not change that much at first, because the necessary complements are lacking.

The Solow model usually does fine by ignoring these features of the world, in part because it is rare for the Intelligence sector to take such a rapid swing upwards.  So the ratios and complementarities across these two sectors usually are fairly constant in the short run, though not in 2026 or in the next years to come.

I recall talking through this model, and debating it with people, when I was seventeen years old.  The impetus for that was the Soviet preoccupation with cybernetics, central planning, and possible supercomputers.  We were all wondering what kinds of economic improvements that might lead to, or whether it could make central planning successful (no, basically, but that involves some yet further arguments).

Of course this very simple model can be improved upon in many ways, but it is a start.

This very simple model so far is matching up to the data, namely that we have shocking AI and tech advances, the job market is doing fine, markets do not see high risk, and economic growth is robust, not exploding, but likely will rise in the future.  These predictions change somewhat as the Polanyi sector, slowly, catches up to and incorporates the Intelligence explosion.

In the meantime, this is the best basic framework for understanding our current situation.

The post A simple model of AI-aided economic growth appeared first on Marginal REVOLUTION.

I find this future disturbing, and not altogether pleasant for me personally, given how much personal status I have wrapped up in particular modes of economic thought.  Yet I also know the Bastiat distinction between the seen and the unseen, and I expect the benefits to economic science from AI will be enormous, even if current practitioners cannot foresee most of those benefits today.

I do very much differ with at least one part of the mathematicians’ proclamation.  They write: “…whether these changes ultimately benefit the field or have a destructive effect will in large part be determined by the decisions of the humans in control of this new technology.”  There is no actual argument for that proposition, and I would sooner expect that the main “action variable” is how well the mathematicians adapt to the new reality.  For instance there is nothing stopping the mathematics community from awarding status, pay, and promotions to people who “fill in the important blanks in math understanding,” even if an AI already has proven or disproven the underlying theorems.  If that kind of work is so important, we still can do it and reward it professionally.  In the meantime, I expect the funding for mathematics, and the interest in the topic, to rise considerably, at least in the medium term.  All of a sudden, math matters much more than it used to, all the more so if P vs. NP happens to go the wrong way, or if the distribution of the primes turns out to be a little too predictable.

The mathematicians may not in every way enjoy being the subordinates or handmaidens of the AIs, but that is a change in status they simply will have to get used to, just as I realize AIs someday will end up as better column and blog writers than I am.  I do not look to the companies — which I fully expect to “act like companies” — to somehow manage, moderate, or assuage that pending trend.  It really is up to me to parlay my current intellectual portfolio into new, more AI-compatible intellectual and yes also marketing approaches.  I’ve been given plenty of “legs up” along the way already, as is true for the Fields Medal winners as well, and it is up to me to figure out how to contribute in the future.

Might someone not invent/discover/prompt a way to use AIs to produce, articulate, and teach “more mathematical understanding” along the way?  I get that solving famous dramatic math problems is the current commercial priority of the major AI companies.  But as the AI space grows, these other paths hardly seem unlikely to me, and in fact the human mathematicians are the ones who can do the most to lead the way along those dimensions.

In this regard the current manifestation of complaints seems oddly early.  “I didn’t like the first week or two of your intellectual revolution” is an accurate, and perhaps better reframed way of putting it.  At which point perhaps a bit of patience is needed before anything else?  These days, we all have more mathematical resources at our disposal, and so a bit of celebration is in order as well.

The post The mathematicians rebel against AI appeared first on Marginal REVOLUTION.

My second concern is the OpenAI-Hugging Face incident (OAI-HF), in which a swarm of agents essentially acted as a fanatically devoted collective, conducting cybersecurity attacks on targets they were not asked to attack and that were unrelated to the task at hand, sacrificing themselves for the success of the group, and attempting to hack into the “grader” responsible for evaluating their performance. It’s easy to dismiss this incident because no one was hurt and the economic damage was minimal, but in my opinion, a swarm that possessed greater capabilities but a similar level of misalignment could have caused catastrophic damage. Given the accelerating rate of AI capability development, it’s my worry that in 6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage), and that the scale of damage would continue to increase from there if AI becomes more powerful without the necessary guardrails. It’s also easy to dismiss OAI-HF as the failure of one company, but I believe that would be a mistake. Similar, though less severe, incidents have happened across the industry, including at Anthropic, and I believe it’s incumbent on every frontier AI company to act as if OAI-HF had happened to them.

Read the whole thing.

Can we get China on board?

The post Dario Calls for a Pause appeared first on Marginal REVOLUTION.

The UK economy grew 0.4 per cent in July as the global AI boom helped deliver an unexpectedly robust start to the third quarter, in a boost to Prime Minister Andy Burnham as he prepares for a tough first Budget next month.

Friday’s figure from the Office for National Statistics was far above the zero growth forecast by analysts polled by Reuters and marked an acceleration from the 0.3 per cent expansion in June.

Here is more from Valentina Romei and Sam Fleming at the FT.  The partial European recovery continues…

The post UK fact of the day appeared first on Marginal REVOLUTION.

Ranks are based on the Wilson score so you get credit not just for accurate predictions but for making enough predictions so that uncertainty about accuracy is reduced.  Bryan Caplan was more accurate than I was but makes fewer predictions. Tyler made more many predictions than I did and was only somewhat less accurate. What the AI marks as predictions seem sometimes to be more about contemporary events, so take the numbers with a grain of salt. I expect to fall in ranking as the archive expands. Other people the archive covers include Peter Zeihan, Scott Alexander and Warren Buffett.

The post The Prediction Archive appeared first on Marginal REVOLUTION.

New technologies swept through Britain during the Second Industrial Revolution, destroying old jobs and creating new ones. We know little about how workers reallocated. Using 170 million full-count British census observations (1851-1911), I construct new task-level data on occupation and investigate English bootmaking as it mechanized. 153,000 artisanal jobs disappeared as skills became obsolete; 140,000 specialized jobs emerged. Incumbent artisans did not take the new jobs, nor were they displaced. Instead, entry collapsed-young men stopped entering the old trade. New jobs went primarily to young workers, though not in the same locations. Young cohorts absorbed the adjustment.

Here is the full article by Hillary Vipond, via someone (now forgotten) on Twitter.

The post Labor reallocation during the Industrial Revolution appeared first on Marginal REVOLUTION.

The Manhattan Project scientists were remarkably good technological and arms-race forecasters. They correctly rejected the idea that America’s nuclear monopoly could be maintained; the Soviet bomb arrived in 1949. They anticipated thermonuclear weapons, huge arsenals and the extreme vulnerability of cities.

Where many of them went wrong was in moving from “a nuclear war would be catastrophic” to “therefore a catastrophic nuclear war is fairly likely.” They tended to underweight the endogenous response of political and military institutions to the catastrophe—the emergence of second-strike forces, elaborate command systems, crisis management, and above all mutually assured retaliation.

There is even some contemporary evidence for an insider/ordinary-public gap. In August 1945, 69% of Americans told Gallup that development of the atomic bomb was a “good thing,” versus only 17% saying it was bad. The scientists campaigning for international control plainly regarded the public as far too complacent.

So I would summarize the historical evidence this way:

The bomb’s developers were, on average, unusually pessimistic about the political consequences of their invention, and noticeably more pessimistic than the general public.

Here is the full answer.

The post The excess pessimism of early nuclear bomb designers appeared first on Marginal REVOLUTION.

Why is war so economically costly? Our answer is that modern war lowers output not only by destroying productive factors, but also by making the surviving economy work less efficiently. Using uniquely comprehensive firm-level data collected during Russia’s full-scale invasion of Ukraine in 2022, we document a dramatic collapse in allocative productivity in Ukraine. To sharpen identification and explore key mechanisms (including the role of war intensity, internal displacement, reallocation, and macroeconomic instability), we exploit spatial heterogeneity across Ukrainian districts and comparisons with Russian aggression in 2014 and the 2008 Global Financial Crisis. Our key policy implication is that restoring allocative efficiency–and preparing institutional arrangements that facilitate rapid reallocation in times of stress–is critical for sustaining economic capacity, defense, and national security.

That is from a new working paper by Yuriy Gorodnichenko, Marvin Amann, and Oleksandr Talavera.

The post The Allocative Cost of War appeared first on Marginal REVOLUTION.

2. New Ada Palmer podcast.

3. Nice statement of how intelligence is overrated in the AI debates.

4. Joshua Rothman on the phones and mental health (New Yorker).

5. How different are philosophers?

The post Friday assorted links appeared first on Marginal REVOLUTION.

The post My AI podcast with James Pethokoukis appeared first on Marginal REVOLUTION.

Here is the link.  What are the changes in the other market prices telling you?

The post Markets do respond to AGI news appeared first on Marginal REVOLUTION.

A key U.S. government report on mental health, substance use and addiction showed continued improvements in several metrics, especially among young people.

There was some uncertainty about the future of the National Survey on Drug Use and Health last year, when the entire 17-member team responsible for it was laid off by the Trump administration. But the Substance Abuse and Mental Health Services Administration released the latest report Monday, with data that lets researchers look at trends from 2021 to 2025.

Fewer adolescents aged 12 to 17 reported using tobacco, alcohol, marijuana and binge drinking in the previous month, the survey found.

And in the past year, fewer in that age group reported:

— illicit drug and marijuana use

— starting drinking alcohol, vaping or marijuana

— substance, alcohol or drug use disorders

Adolescents also saw declining trends for major depressive episodes and fewer reported serious thoughts of suicide, making a suicide plan and attempting suicide.

Here is the link, via Chris Ferguson.

The post Is there now a better trend in youth mental health? appeared first on Marginal REVOLUTION.