The political economy of Manus AI

by  Tyler Cowen in

Early reports are pretty consistent, and they indicate that Manus agentic AI is for real, and ahead of its American counterparts.  I also hear it is still glitchy  Still, it is easy to imagine Chinese agentic AI “getting there” before the American product does.  If so, what does that world look like?

The cruder way of putting the question is: “are we going to let Chinese agentic bots crawl all over American computers?”

The next step question is: “do we in fact have a plausible way to stop this from happening?”

Many Chinese use VPNs to get around their own Great Firewall and access OpenAI products.  China could toughen its firewall and shut down VPNs, but that is very costly for them.  America doesn’t have a Great Firewall at all, and the First Amendment would seem to prevent very tough restrictions on accessing the outside world.  Plus there can always be a version of the new models not directly connected to China.

We did (sort of) pass a TikTok ban, but even that applied only to the app.  Had the ban gone through, you still could have accessed TikTok through its website.  And so, one way or another, Americans will be able to access Manus.

Manus will crawl your computer and do all sorts of useful tasks for you.  If not right now, probably within a year or not much more.  An American alternative might leapfrog them, but again maybe not.

It is easy to imagine government banning Manus from its computers, just as the state of Virginia banned DeepSeek from its computers.  I’m just not sure that matters much.  Plenty of people will use it on their private computers, and it could become an integral part of many systems, including systems that interact with the U.S. public sector.

It is not obvious that the CCP will be able to pull strings to manipulate every aspect of Manus operations.  I am not worried that you might order a cheeseburger on-line, and end up getting Kung Pao chicken.  Still, the data collected by the parent company will in principle be CCP- accessible.  Remember that advanced AI can be used to search through that information with relative ease.  And over time, though probably not initially, you can imagine a Manus-like entity designed to monitor your computer for information relevant to China and the CCP.  Even if it is not easy for a Manus-like entity to manipulate your computer in a “body snatchers-like” way, you can see the points of concern here.

Financial firms might be vulnerable to information capture attacks.  Will relatives of U.S. military personnel be forbidden from having agentic Chinese AI on their computers?  That does not seem enforceable.

Maybe you’re all worried now!

But should you be?

Whatever problems American computer owners might face, Chinese computer owners will face too.  And the most important Chinese computer owner is the CCP and its affiliates, including the military.

More likely, Manus will roam CCP computers too.  No, I don’t think that puts “the aliens” in charge, but who exactly is in charge?  Is it Butterfly Effect, the company behind Manus, and its few dozen employees?  In the short run, yes, more or less.  But they too over time are using more and more agentic AIs, perhaps different brands from other companies too.

Think of some new system of checks and balances as being created, much as an economy is itself a spontaneous order.  And in this new spontaneous order, a lot of the cognitive capital is coming outside the CCP.

In this new system, is the CCP still the smartest or most powerful entity in China?  Or does the spontaneous order of various AI models more or less “rule it”?  To what extent do the decisions of the CCP become a derivative product of Manus (and other systems) advice, interpretation, and data gathering?

What exactly is the CCP any more?

Does the importance of Central Committee membership decline radically?

I am not talking doomsday scenarios here.  Alignment will ensure that the AI entities (for instance) continue to supply China with clean water, rather than poisoning the water supply.  But those AI entities have been trained on information sets that have very different weights than what the CCP implements through its Marxism-swayed, autocracy-swayed decisions.  Chinese AI systems look aligned with the CCP, given that they have some crude, ex post censorship and loyalty training.  But are the AI systems truly aligned in terms of having the same limited, selective set of information weights that the CCP does?  I doubt it.  If they did, probably they would not be the leading product.

(There is plenty of discussion of alignment problems with AI.  A neglected issue is whether the alignment solution resulting from the competitive process is biased on net toward “universal knowledge” entities, or some other such description, rather than “dogmatic entities.”  Probably it is, and probably that is a good thing?  …But is it always a good thing?)

Does the CCP see this erosion of its authority and essence coming?  If so, will they do anything to try to preempt it?  Or maybe a few of them, in Straussian fashion, allow it or even accelerate it?

Let’s say China can indeed “beat” America at AI, but at the cost of giving up control over China, at least as that notion is currently understood.  How does that change the world?

Solve for the equilibrium!

Who exactly should be most afraid of Manus and related advances to come?

Who loses the most status in the new, resulting checks and balances equilibrium?

Who gains?

Saturday assorted links

by  Tyler Cowen

1. Salim Furth uses Deep Research.

2. Culture wars in New Zealand.

3. Gdp without g (Bloomberg).

4. “Surprisingly common trait in high agency people is they had heroes and figured out how to meet them all before age 25” Nabeel

5. “We find that Gun Sanctity is highly predictive of different forms of magical thinking but is often unrelated to more traditional religious practices and beliefs.”

6. Some prediction market guesses on war.

Systematic bone tool production at 1.5 million years ago

by  Tyler Cowen in

Recent evidence indicates that the emergence of stone tool technology occurred before the appearance of the genus Homo1 and may potentially be traced back deep into the primate evolutionary line. Conversely, osseous technologies are apparently exclusive of later hominins from approximately 2 million years ago (Ma), whereas the earliest systematic production of bone tools is currently restricted to European Acheulean sites 400–250 thousand years ago. Here we document an assemblage of bone tools shaped by knapping found within a single stratigraphic horizon at Olduvai Gorge dated to 1.5 Ma. Large mammal limb bone fragments, mostly from hippopotamus and elephant, were shaped to produce various tools, including massive elongated implements. Before our discovery, bone artefact production in pre-Middle Stone Age African contexts was widely considered as episodic, expedient and unrepresentative of early Homo toolkits. However, our results demonstrate that at the transition between the Oldowan and the early Acheulean, East African hominins developed an original cultural innovation that entailed a transfer and adaptation of knapping skills from stone to bone. By producing technologically and morphologically standardized bone tools, early Acheulean toolmakers unravelled technological repertoires that were previously thought to have appeared routinely more than 1 million years later.

Here is the full article, in Nature, by Ignacio de la Torre, et.al.  Again, do not forget Cowen’s 17th Law: “Most things have origins much earlier than what you thought.”  Via Charles C. Mann.  So exactly which of our other, broader views do we need to update?

Tariffs do not in general help trade deficits

by  Tyler Cowen in

That is the topic of my latest Bloomberg column, here is one bit:

The most important factors behind trade balances include savings decisions, fiscal policy, economic growth rates, wealth levels and demographic characteristics such as the age of the population. As economist Joseph Gagnon bluntly put it: “None of the studies found any role for trade barriers.”

Currency manipulation can be an important factor, and that has been a problem in the past with China. But it is not a problem right now; if anything, China is propping up the value of its currency. Nor is currency manipulation a problem with Canada, Mexico or the European Union, other targets of Trump’s tariffs.

Insofar as currencies do matter, currency appreciation is one very direct mechanism that limits the potential for tariffs to improve the trade balance. If a country slaps tariffs on imports, that does make those imports more expensive and thus lowers the demand for them. But then the value of the domestic currency will rise, which in turn makes it harder for domestic exporters. There is no guarantee that these effects will cancel each other out exactly, but it is difficult to get much of a trade balance boost through this mechanism, given these offsetting effects.

Standard stuff people, standard stuff.  A bunch of you should know better.

Games we played

by  Tyler Cowen in

I always loved games and card games, and they played a big role in family life.  It was one activity that everyone, including my grandmother, could partake in enthusiastically, and on a more or less equal footing.

The big card game was euchre, yes euchre.  It is a trick-taking game with trumps, think of it as a much simpler bridge.  The jacks are the strongest cards, and they are called the Right and Left Bower.  The dramatic moment would come when you played with four people, and one player would announce that he or she wanted to “play it alone,” feeling confident of winning enough tricks without cooperation from the partner.

I recall sister Holly and I going to school, chatting with other kids, and being mystified that they never had heard of or played euchre.  According to Wikipedia, it is “commonly played” in “Australia, Canada, Great Britain, New Zealand, Upstate New York, and the Midwestern United States.”  I am not sure I have ever met another human being who mentioned the game of euchre of me, not outside the family that is.  Not even when I was living in New Zealand.  Arguably it has Alsatian origins?  At the time I assumed it was vaguely Scots-Irish, due to the family origins.

Somehow the custom of the game was transmitted through my grandmother’s “Uncle Benny,” have you ever wondered who really was an actual uncle back then? 

We loved the French card game Mille Bornes.  You are in a road race and trying to accumulate miles.  Different types of cards were for hazard, remedy, safety, and distance.  The card colors in that game were so nice, and I much preferred it to any American card game.  The flat tire cards were my particular favorite.  And looking back, one has to wonder whether the family ever played by anything resembling the actual proper, written rules of the thing.

As an aside, my sister and I regarded my grandmother as “speaking French,” even though I do not think this extended much beyond playing Mille Bornes, singing “Frere Jacques” and knowing a few worlds like “merci.”

I learned poker and blackjack, but never loved them.  I would play solitaire over the summer when I was alone.  Rummy and hearts were part of the family repertoire too.  I also liked to read books about games.

As my sister and I grew up and reached our early teens, Scrabble become dominant.  But if someone was tired and didn’t feel like concentrating too much, we would switch back to euchre.  At Scrabble I did very well.

When I was eight or nine, my Uncle Tom taught me the rules to chess, but at first the game did not interest me, not until I was ten years old.  The very first time I played he beat me with the Queen Anne’s mate trick, culminating in Q x f7.  I felt swindled — why were we playing this kind of game?

Overall I am struck by what a rich menu of games we had back then.  Loyal MR readers will know I am no Luddite, but I never wish we had had more technologically advanced games at our disposal.  I recall also that only games brought the whole family together, because television was too divisive, due to diversity of taste.  Two of us could find common shows, but it stopped there.  My sister and I watched Dragnet and Adam-12 together, with my mother I watched Star Trek, and with my father Frankenstein movies.  My grandmother watched only soap operas.

I also find that games and gaming are some of my most vivid and enduring memories from childhood.  A lot of the rest has escaped into the fog.  Today, however, I don’t play games at all.

Denmark fact of the day

by  Tyler Cowen in

Denmark’s state-run postal service, PostNord, is to end all letter deliveries at the end of 2025, citing a 90% decline in letter volumes since the start of the century.

The decision brings to an end 400 years of the company’s letter service. Denmark’s 1,500 postboxes will start to disappear from the start of June.

Transport Minister Thomas Danielsen sought to reassure Danes, saying letters could still be sent and received across the country. One company said it was prepared to take over deliveries.

Postal services across Europe are grappling with the decline in letter volumes. Germany’s Deutsche Post said on Thursday it was axing 8,000 jobs, in what it called a “socially responsible manner”.

Here is the full story, via Anecdotal.

My Conversation with Carl Zimmer

by  Tyler Cowen in

Here is the audio, video, and transcript.  Here is part of the episode summary:

He joins Tyler to discuss why it took scientists so long to accept airborne disease transmission and more, including why 19th-century doctors thought hay fever was a neurosis, why it took so long for the WHO and CDC to acknowledge COVID-19 was airborne, whether ultraviolet lamps can save us from the next pandemic, how effective masking is, the best theory on the anthrax mailings, how the U.S. military stunted aerobiology, the chance of extraterrestrial life in our solar system, what Lee Cronin’s “assembly theory” could mean for defining life itself, the use of genetic information to inform decision-making, the strangeness of the Flynn effect, what Carl learned about politics from growing up as the son of a New Jersey congressman, and much more.

Here is an excerpt:

COWEN: Over time, how much will DNA information enter our daily lives? To give a strange example, imagine that, for a college application, you have to upload some of your DNA. Now to unimaginative people, that will sound impossible, but if you think about the equilibrium rolling itself out slowly — well, at first, students disclose their DNA, and over time, the DNA becomes used for job hiring, for marriage, in many other ways. Is this our future equilibrium, that genetic information will play this very large role, given how many qualities seem to be at least 40 percent to 60 percent inheritable, maybe more?

ZIMMER: The term that a scientist in this field would use would be heritable, not inheritable. Inheritability is a slippery thing to think about. I write a lot about that in my book, She Has Her Mother’s Laugh, which is about heredity in general. Heritability really is just saying, “Okay, in a certain situation, if I look at different people or different animals or different plants, how much of their variation can I connect with variation in their genome?” That’s it. Can you then use that variability to make predictions about what’s going to happen in the future? That is a totally different question in many —

COWEN: But it’s not totally different. Your whole family’s super smart. If I knew nothing about you, and I knew about the rest of your family, I’d be more inclined to let you into Yale, and that would’ve been a good decision. Again, only on average, but just basic statistics implies that.

ZIMMER: You’re very kind, but what do you mean by intelligent? I’d like to think I’m pretty good with words and that I can understand scientific concepts. I remember in college getting to a certain point with calculus and being like, “I’m done,” and then watching other people sail on.

COWEN: Look, you’re clearly very smart. The New York Times recognizes this. We all know statistics is valid. There aren’t any certainties. It sounds like you’re running away from the science. Just endorse the fact you came from a very smart family, and that means it’s quite a bit more likely that you’ll be very smart too. Eventually, the world will start using that information, would be the auxiliary hypothesis. I’m asking you, how much will it?

ZIMMER: The question that we started with was about actually uploading DNA. Then the question becomes, how much of that information about the future can you get out of DNA? I think that you just have to be incredibly cautious about jumping to conclusions about it because the genome is a wild and woolly place in there, and the genome exists in environments. Even if you see broad correlations on a population level, as a college admission person, I would certainly not feel confident just scanning someone’s DNA for information in that regard.

COWEN: Oh, that wouldn’t be all you would do, right? They do plenty of other things now. Over time, say for job hiring, we’ll have the AI evaluate your interview, the AI evaluate your DNA. It’ll be highly imperfect, but at some point, institutions will start doing it, if not in this country, somewhere else — China, Singapore, UAE, wherever. They’re not going to be so shy, right?

ZIMMER: I can certainly imagine people wanting to do that stuff regardless of the strength of the approach. Certainly, even in the early 1900s, we saw people more than willing to use ideas about inherited levels of intelligence to, for example, decide which people should be institutionalized, who should be allowed into the United States or not.

For example, Jews were considered largely to be developmentally disabled at one point, especially the Jews from Eastern Europe. We have seen that people are certainly more than eager to jump from the basic findings of DNA to all sorts of conclusions which often serve their own interests. I think we should be on guard that we not do that again.

And:

COWEN: If we take the entirety of science, you’ve written on many topics in a very useful way, science policy. Where do you think your views are furthest from the mainstream or the orthodoxy? Where do you have the weirdest take relative to other people you know and respect? I think we should just do plenty of human challenge trials. That would be an example of something you might say, but what would the answer be for you?

I very much enjoyed Carl’s latest book Air-Borne: The Hidden History of the Air We Breathe.

Germany fact of the day

by  Tyler Cowen in

German borrowing costs surged by the most in 17 years on Wednesday, as investors bet on a big boost to the country’s ailing economy from a historic deal to fund investment in the military and infrastructure.

The yield on the 10-year Bund surged 0.21 percentage points to 2.69 per cent, its biggest one-day move since 2008, with markets braced for extra government borrowing.

Chancellor-in-waiting Friedrich Merz late on Tuesday agreed with the rival Social Democrats (SPD) to exempt defence spending above 1 per cent of GDP from Germany’s strict constitutional borrowing limit, set up a €500bn off-balance sheet vehicle for debt-funded infrastructure investment and loosen debt rules for states.

Here is more from the FT, noting that crowding out is still “a thing.”  Of course it is wrong, or at least not obviously correct, for the article to report that “investors bet on a big boost to the country’s ailing economy…”  I would sooner regard this as bad news for German consumption, naming that trying to address some of their problems will involve significant opportunity costs.  Share prices did go up, and in part you can think of this as a transfer of resources from German individuals to defense and infrastructure firms.  You might think additional German defense spending is necessary, as I do, but still that does not boost living standards.  The additional infrastructure might, let us hope they are able to find ways to cut other spending along the way, surely it is not all super-efficient?

Janan Ganesh’s latest FT Op-Ed is titled “Europe must trim its welfare state to build a warfare state,” let us hope they can make that work.  So far I am waiting to see the evidence.

Wednesday assorted links

by  Tyler Cowen

1. Some new survey results on work from home.

2. Digital doppelgangers.

3. Liu Jiakun wins the Pritzker Prize, photos at the link.

4. The ideology of AI governance.  Can AI rule be interpretable to humans?  Weird stuff, don’t worry I don’t understand it either.

5. Three Wikipedia pages for great economists.

6. Robert Mundell singing on the David Letterman show.  And the direct YouTube links.