MAMLM as a General Purpose Technology: The Ghost in the GDP Machine
Silicon Promises, estimated statistics, sunnier realities: the slow diffusion of “artificial Intelligence”. rom steam to silicon, history’s most transformative inventions rarely deliver on schedule...
Silicon Promises, estimated statistics, sunnier realities: the slow diffusion of “artificial Intelligence”. rom steam to silicon, history’s most transformative inventions rarely deliver on schedule—or as advertised—in the short- and the medium-run. The true transformation of work, wealth, and society will be slower, messier, and far more uneven than Silicon Valley’s PowerPoint prophets would have you believe. If history is any guide, the productivity revolution will come not with a bang, but with a series of small, cumulative changes—provided we have the patience, and the institutional imagination, to see it through…
Torsten Slok: Productivity Gains Are Coming <https://www.apolloacademy.com/productivity-gains-are-coming/>: ‘The Census conducts a biweekly survey of 1.2 million firms, and one question is whether a business has used AI tools such as machine learning, natural language processing, virtual agents, or voice recognition to help produce goods or services in the past two weeks, see chart below. Nine percent of firms reported using AI, and the rising trend in AI adoption increases the likelihood of a rise in productivity over the coming quarters:
First, even though every CEO on Earth is now “investing in AI”, only 1% of firms describe themselves as “mature” in AI deployment: that is, having integrated AI deeply enough that it is actually altering workflows, decision-making, and value creationat scale. The bottleneck, it turns out, is not the technology itself. The algorithms are, if anything, overabundant; the cloud-compute capacity, vast; the open-source models, multiplying like rabbits—if unreliable, hallucinating rabbits. The constraint is organizational: the willingness—and, I would add, the managerial imagination—to tear up old workflows and rebuild them around the new tools. To borrow a line from Paul David’s classic account of electrification, installing the dynamo is easy; reorganizing the factory to exploit it is hard. The same holds for AI. It is not enough to bolt a chatbot onto your customer support page or automate a few invoices.
Second, somebody ought to make it their thing to push the Silicon Law of Attention Conservation. If an improvement in information technology means that you and others can write three times as fast, that also means that you have three times as much to read and think about. If, initially, you spent more than 3/4 of your time writing, you come out ahead. If, initially, you spent less than 3/4 of your time writing, you come out behind.
Consumer surplus is different. The coming of natural-language interfaces is absolutely wonderful. The flood of cultural and intellectual production means that with even a slightly effective filter the quality of the experiences you obtain goes way up. You come out ahead as long as you are master of your own attention, rather than the victim of those who hack your brain to make your attention their slave for their own profit. And I am enough of a cyber-optimist to believe that even in the medium run we can use rather than be used by our technologies for our benefit, no matter what the counterexamples of FaceBook and Twitter and Mark Zuckerberg and Elon Musk might suggest.
So if you want to argue—as you should—that user surplus matters, and that our measures of economic performance should be human-welfare rather than production-quantity based, should be the result of assessments of willingness-to-pay rather than of factor cost, well, then, yes, AI—modern advanced machine-learning models, with their extraordinary powers of very big-data, very high-dimension, very flexible-function classification, estimation, and prediction, are going to be a very substantial boon. But GDP? Profits? In order to sell you a product, right now Amazon needs a website, a warehouse, and a truck plus web programmers, warehouse workers, and truck drivers. In the age of AI, Amazon will also need a GPT LLM model, huge honking silicon service farms, and AI-programmers. To sell the same product. How is this not a measured decline in productivity in the GDP accounts? Same product out; more required resources in.
This productivity decline is not real—a better, human welfare-based measure would include the value to the user of interacting with Amazon via the natural-language interface. But it will be reported. If there is a productivity increase measured in the GDP accounts, it will come from:
Reducing the number of workers needed to do the job once the workers have AI-assistants…
Not then turning around and having to hire those workers back to process all the AI-slop masquerading as information flowing into the system…
Finding new things that MAMLMs enable that people will pay for…
I am enough of a cyber-optimist to, again, think that these will emerge. But I do think it will take quite a while, at least as financial market horizons measure time.
Artificial Intelligence (AI), we are told, is the new steam engine, the new electrification, the new computer: a “general-purpose technology” (GPT) that promises profound, economy-wide transformation. But what, precisely, does that mean? General-purpose technologies are those rare innovations—think James Watt’s steam engine in the late 18th century, Edison’s electrification in the 19th, or Turing’s computer in the 20th—that fundamentally alter the production possibilities of entire economies. They are not mere gadgets or sectoral upgrades; rather, they are platforms upon which countless other innovations are built.
The steam engine did not just power textile mills; it made possible the railroad, the ocean liner, the mechanized mine. Electricity did not simply replace gaslight; it reorganized factories, enabled skyscrapers, and redefined the rhythms of urban life. Computers, for their part, have insinuated themselves into every corner of economic activity, from the humble spreadsheet to the global supply chain. Yet, if one consults the historical record, the story is not one of instant, universal uplift. The productivity impacts of GPTs are real, but they are also stubbornly diffuse, slow to appear in the official statistics, and highly uneven across sectors.
Take the classic example of electrification: Paul David’s now-famous study of American manufacturing shows that, although the electric dynamo was invented in the 1870s, it took nearly half a century for factories to reorganize themselves to take full advantage of it. Productivity growth in manufacturing lagged, and only after complementary investments—in new factory layouts, new skills, new business models—did the “electric age” really take off. The same pattern recurred with computers: Robert Solow quipped in 1987 that “you can see the computer age everywhere but in the productivity statistics.” It was not until the late 1990s that the IT revolution delivered its long-promised productivity surge.
AI, I think, is likely to follow this well-trodden path. The hype cycle—fueled by breathless headlines, venture capital exuberance, and the fever dreams of Silicon Valley—inevitably runs ahead of the slow, plodding march of the productivity statistics. Early adopters trumpet dramatic gains; skeptics point to the absence of aggregate effects. Sectors with the right mix of complementary assets—data, digital infrastructure, skilled labor—leap ahead, while others dawdle or even fall behind. And all the while, the measurement of productivity itself becomes more fraught, as new forms of value (think: better recommendations, faster search, personalized content) slip through the fingers of GDP accountants.
In sum: AI is the latest GPT to promise the moon. If history is any guide, we should expect the benefits to be real but delayed, broad but uneven, and—at least for a while—more visible in the stories we tell than in the numbers we collect. The challenge, as always, is to turn potential into reality, and to ensure that the gains are widely shared.
And as for AI’s “superagency” or Mark Zuckerberg’s building ASI—Artificial Super Intelligence? It is, right now, truly nowhere. While the phrase “AI superagency” is now a favorite in the armory of Silicon Valley’s self-mythologizing rhetoricians, and the idea is seductive, there has been next to no true advance in turning every knowledge worker into a sort of intellectual superhero, equipped with digital exoskeletons for mind and memory. In the real world—by which I mean the world of quarterly earnings and harried middle managers—AI is most commonly found and will be found crunching invoices, triaging customer service tickets, or generating boilerplate marketing copy. The much-vaunted “copilots” and “assistants” are, for now, glorified autocomplete engines.
Thus I believe that Daron Acemoglu is right with his central estimate: the impact of current-generation AI on U.S. GDP growth is likely to be “nontrivial, but modest”—perhaps a 1% cumulative boost over a decade. This is not nothing, but it is a far cry from the “fourth industrial revolution” rhetoric that animates so many consulting reports and TED talks. Why the apparent mismatch between the fevered expectation and the measured reality? The answer, I think, lies in the nature of work and the limits of current technology. While AI excels at automating tasks that are routine, repetitive, and codifiable—think invoice processing, spam filtering, or basic customer service—the vast majority of economic activity resists such neat encapsulation. Most jobs, even those in ostensibly “automatable” sectors, consist of a patchwork of tasks: some ripe for automation, others stubbornly dependent on human judgment, dexterity, or social intelligence.
I do recommend Acemoglu’s “The Simple Macroeconomics of AI”. Acemoglu’s back-of-the-envelope calculations suggest that perhaps 5% of all tasks in the U.S. economy are currently profitable to automate with existing AI systems. The rest—well, they remain the province of fallible, expensive, but surprisingly adaptable Homo sapiens. This is not to say that AI’s impact will be invisible. Even a 1% boost to GDP is, in historical terms, significant—roughly equivalent to the entire economic contribution of the internet in its first decade. But it is also worth recalling that the internet, for all its transformative promise, took decades to percolate through the economy, and its benefits were highly unevenly distributed. The same, I suspect, will be true for AI.
The truly transformative scenarios—where AI systems not only automate routine tasks but also create new industries, new forms of organization, and new sources of value—remain, for now, speculative. They may arrive, in time.
Autor, David. “Why Are There Still So Many Jobs? The History and Future of Workplace Automation.” Journal of Economic Perspectives, vol. 29, no. 3, 2015, pp. 3–30. https://pubs.aeaweb.org/doi/pdfplus/10.1257/jep.29.3.3
David, Paul A. “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox.” American Economic Review, vol. 80, no. 2, 1990, pp. 355–361. https://www.jstor.org/stable/20066677
If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…
"Second, somebody ought to make it their thing to push the Silicon Law of Attention Conservation. If an improvement in information technology means that you and others can write three times as fast, that also means that you have three times as much to read and think about. If, initially, you spent more than 3/4 of your time writing, you come out ahead. If, initially, you spent less than 3/4 of your time writing, you come out behind."
It seems to me that there is yet another aspect when one considers economic productivity.
Assume that there is some generally available information technology that "means that you and others can write three times as fast". If this technology is generally available, then "others" will include all (or most) of your competitors, as well. And this means that, though all of you may be more "productive" in some sense, there will be little or not competitive advantage and thus little room for anyone to profit from that greater productivity. Which in turn means that any increase in GDP may turn out to be minimal.
A simple test. Download the text from a recent fiction book that you have recently read and ask an LLM to summarize it. I've found that the summary will make basic plot errors while bs'ing its way through with vapid insights.
Again, chiming in very late, but...
"Second, somebody ought to make it their thing to push the Silicon Law of Attention Conservation. If an improvement in information technology means that you and others can write three times as fast, that also means that you have three times as much to read and think about. If, initially, you spent more than 3/4 of your time writing, you come out ahead. If, initially, you spent less than 3/4 of your time writing, you come out behind."
It seems to me that there is yet another aspect when one considers economic productivity.
Assume that there is some generally available information technology that "means that you and others can write three times as fast". If this technology is generally available, then "others" will include all (or most) of your competitors, as well. And this means that, though all of you may be more "productive" in some sense, there will be little or not competitive advantage and thus little room for anyone to profit from that greater productivity. Which in turn means that any increase in GDP may turn out to be minimal.
A simple test. Download the text from a recent fiction book that you have recently read and ask an LLM to summarize it. I've found that the summary will make basic plot errors while bs'ing its way through with vapid insights.