‘You cannot make a machine to think for you.’ This is a commonplace that is usually accepted without question. It will be the purpose of this paper to question it.

Turing was fascinated by machine capabilities throughout his career. In 1936 he made a fundamental contribution to the question of what it means for something to be “computable.” Kurt Gödel and Alonzo Church had recently tackled the question of how to define the set of computable functions, meaning the set of functions whose values can be calculated by an algorithm. Turing developed the most powerful way of thinking about this question. He imagined an abstract computer, now called a Turing machine, that can carry out computations according to the inputs specified on a possibly infinite tape—for example, instructions to implement basic mathematical operations. He then defined a function to be computable if such a machine could compute its values. A machine is said to be a universal Turing machine if it can compute any number that can be calculated by any Turing machine. Notably, if the human mind is in essence a very sophisticated computer and the tasks that it performs are within the class of computable functions, then a universal Turing machine could replicate all human capabilities. Before World War II, however, Turing did not venture into the question of whether machines could really think and how far they could go in performing human tasks. During the war, Turing joined the top-secret Bletchley Park research facility, where mathematicians and other experts worked to understand encrypted German radio messages. He devised a clever algorithm—and designed a machine—to speed up the breaking of enemy ciphers. This then helped British intelligence to quickly decipher encrypted communications that the Germans had presumed to be unbreakable. After Bletchley, Turing took the next step in his prewar work on computation. In 1947 he declared to a meeting of the London Mathematical Society that machines could be intelligent. Undeterred by the hostile reactions of participants, Turing continued to work on the problem. In 1951 he wrote: “‘You cannot make a machine think for you.’ This is a commonplace that is usually accepted without question. It will be the purpose of this paper to question it.” His seminal 1950 paper, “Computing Machinery and Intelligence,” defines one notion of what it means for a machine to be intelligent. Turing imagined an “imitation game” (now called a Turing test) in which an evaluator engages in a conversation with two entities, one human and one machine. By asking a series of questions communicated via a computer keyboard and screen, the evaluator attempts to tell which one is which. A machine is intelligent if it can evade detection. No machine is currently intelligent according to this definition, but one could turn it into a less categorical ranking of machine intelligence. The better a machine can imitate humans, the more intelligent it is. To make this operational, one can define the notion of “human parity” at a task, which would be achieved if a machine can perform that task at least as well as humans. Then, the more tasks a machine can reach human parity in, the more intelligent it is. Turing’s own thoughts on this subject were subtler. He understood that passing this test might not mean true thinking capacity: “I do not wish to give the impression that I think there is no mystery about consciousness. There is, for instance, something of a paradox connected with any attempt to localise it.” Despite this reservation, the modern field of AI followed in Turing’s footsteps and focused on artificial intelligence, defined as machines acting autonomously, reaching human parity, and subsequently outperforming humans.

I propose to consider the question, “Can machines think?” This should begin with definitions of the meaning of the terms “machine” and “think.” The definitions might be framed so as to reflect so far as possible the normal use of the words, but this attitude is dangerous, If the meaning of the words “machine” and “think” are to be found by examining how they are commonly used it is difficult to escape the conclusion that the meaning and the answer to the question, “Can machines think?” is to be sought in a

statistical survey such as a Gallup poll. But this is absurd. Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words.

that the use of AI can set up self-reinforcing dynamics in which those who benefit economically can gain political influence and power at the expense of wider democratic participation.

“We need these discussions,” Johnson says. “There’s nothing inherent in technology. It’s within our control. Even if you think we can’t say no to new technology, you can channel it, and get better outcomes from it, if you talk about it.”

With an eye to identifying these groups, the economic literature has concentrated on the computation of “task exposures,” the share of tasks that AI might carry out autonomously in each job, finding that white-collar high-skilled workers are most exposed. Part of the literature adopts a view of AI as an automation technology—a technology that replaces workers, like industrial robots. In this view, occupations with high task exposure will face higher displacement, and exposed workers will see reduced employment opportunities and wages. Another strand of the literature instead tries to assess the actual consequences of such exposure, finding that AI may augment workers in exposed occupations. We close this section with a description of policy options to cope with potential worker displacement that research has proposed.

https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024199-print-pdf.pdf

The rapid progress over the past few years has compressed the timelines for achieving AGI. In the 2010s, the median estimate of AI researchers for when AGI would be reached was in the second half of the 21st century (Grace et al., 2018). More recently, a growing number of AI researchers and industry leaders have offered much shorter timelines. Geoffrey Hinton, one of the three “godfathers of deep learning” who won the Turing award for their contributions to the field, proclaimed that “[he had] suddenly switched [his] views on whether these things are going to be more intelligent than us” and expects AGI to be reached in “5 to 20 years but without much confidence [since] we live in very uncertain times” (Hinton, 2023). Likewise, Sam Altman, CEO of OpenAI, stated that “AGI will be a reality in 5 years, give or take” in early 2024. These predictions are also in line with the expectations of the general public (Pauketat et al., 2023). AGI would be transformative for our economy and society, creating both opportunities and risks of unprecedented scope. This has led to a growing sense of alarm among technology experts who see potentially radical changes occurring within a short period. They see the exponential growth of AI and associated scaling laws and wonder why the world is not more alarmed. Geoffrey Hinton’s warnings are akin to Albert Einstein’s letter to US President Franklin Delano Roosevelt, in which he warned of the possibility that nuclear power could be harnessed for atomic weapons, giving rise to the nuclear age. There was significant uncertainty about how realistic the predictions of Einstein and other experts were at the time, and how soon their predictions would be realized. In 1933, Ernest Rutherford called the notion that nuclear power could ever be harnessed “moonshine.” We are facing similar uncertainty today, but the scaling of AI is nonetheless proceeding relentlessly.

“We believe that the primary economic challenge posed by the proliferation of AI will be one of income distribution. We economists set ourselves too easy a goal if we just say that technological progress can make everybody better off – we also have to say how we can make this happen. The following paper is an attempt to do so by discussing some of the key economic research issues that this brings up.”

Keynes famously foresaw the steady increase in per capita income during the twentieth century from the introduction of new technologies, but incorrectly predicted that this would create widespread technological unemployment as machines replaced human labor (Keynes 1930).

Geoff Hinton:

Nobody—I think 20 years in the future—nobody’s got a clue what the impact of this is going to be. The impact on society in particular, because it’s clear that a lot of jobs are going to disappear. It’s not clear that it’s going to create jobs to replace them. This isn’t AI’s problem. This is our political systems’ problem. If you get a massive increase in productivity, how does that wealth get shared around? And we don’t—at least in the United States—you don’t at present have the ideal government for that.

Jeff Dean:

Yeah. I mean, I actually got together with a bunch of great co-authors last year and looked at what might the impact of AI be on a bunch of different areas—some of which it clearly could have amazing impact, like healthcare, education, being able to create new kinds of media. But we also looked at the potential impact on employment, misinformation, geopolitical issues. And I think there’s a whole balance of these things.

One of the areas I’m most excited about is how could we make scientific breakthroughs happen faster by making connections between disparate fields that in any one person’s head may not really necessarily realize this other thing is important in some completely different field, or automating the loop for scientific discovery for some fields where that’s possible.

Geoff Hinton:

Let me take off on that. One thing these big models have is they’re compressing a huge amount of knowledge into not many connections—only a trillion. And we know that to compress lots of knowledge, you have to find what’s common to apparently different bits of knowledge. So I believe that when you’re training these big models, they’re already finding commonalities between things where people haven’t ever seen the commonalities. They know much more than any one person. They’re probably finding commonalities between things in Greek literature and things in quantum [mechanics]…

Across all these domains, the possibilities are boundless, but the goal remains constant: AI that augments human expertise, accelerates human discovery, and amplifies human care—not replacing the judgment, creativity, and empathy that are central for being humans.

Supposedly, we will all be the beneficiaries of these spectacular new capabilities. The current CEOs of Amazon, Facebook, Google, and Microsoft have all claimed that AI will beneficially transform technology in the next decades. As Kai-Fu Lee, former president of Google China, puts it, “And like most technologies, AI will eventually produce more positive than negative impacts on our society.”

The two potential risks from advanced AI chosen as the most important by both the US and UK public from a list of seven were increasing unemployment (US: 44%, UK: 49%) and the creation of more dangerous military robots (US: 34%, UK: 39%). Significantly increasing electricity consumption (US: 15%, UK: 16%) and increasing economic inequality (US: 22%, UK: 25%) were chosen the least often (Public First, 2023a, 2023b).

Unemployment across society People are generally worried that AI will increase unemployment but this does not yet appear to be an overwhelming concern for the public.

In January 2023, 73% of US adults surveyed felt that machines with the ability to think for themselves would hurt jobs and the economy, largely unchanged since April 2015 (72%) (Monmouth University Poll, 2023).

In the same poll, the increase in unemployment was the top risk chosen by 49% of UK adults when asked to choose the greatest risk from advanced AI from a selection of seven (ahead of worries about more dangerous military robots, chosen by 39%). A similar pattern was observed in the US, with 44% of adults choosing the risk of increasing unemployment, again the top concern ahead of military robots (34%) (Public First, 2023a, 2023b).

Personal automation concern There is a non-negligible concerned minority in most countries that worry about their own jobs being replaced by AI, but the majority of people in North America and Europe do not think AI will replace them in the next years.

For example, in May 2023, 64% of UK adults thought that more jobs will be lost to automation by robotics/AI than will be created. Only 7% thought more jobs would be created, and 12% thought it would be about the same. Despite this, only 14% were very/fairly worried about the impact that robotics/AI would have on their current job, and only 22% about its impact on their future career, perhaps because a majority (59%) also thought that their job would primarily still be done by humans in the next 30 years (YouGov, 2023a).

The percentage who agree that AI products and services have more benefits than drawbacks is substantially higher in China (78%) and India (71%) than in the United States (35%) and Great Britain (38%). There are substantial other cross-country differences across the questions asked on the survey (Ipsos, 2022b).

Artificial Intelligence (AI), like any transformative technology, has the potential to be a double-edged sword, leading either toward significant advancements or detrimental outcomes for society as a whole. As is often the case when it comes to widely-used technologies in market economies (e.g., cars and semiconductor chips), commercial interest tends to be the predominant guiding factor. The AI community is at risk of becoming polarized to either take a laissez-faire attitude toward AI development, or to call for government overregulation.

Our view is that we are still in the early days of practical AI, and that focused efforts by practitioners, policymakers, and other stakeholders can still maximize the upsides of AI and minimize its downsides.

We think the best bet going forward is to assume AI progress will continue or speed up, and not slow down. AI’s impact on society will be profound.

AI Milestone: AI Scientific Breakthroughs for the UN Sustainable Development Goals (SDGs). To encourage the greater use of AI systems in science and to help acknowledge the potential of AI systems to benefit humanity, this milestone is for an important scientific breakthrough using AI that helps with one or more of the UN SDGs. AI Milestone: Scientist’s AI Aide/Collaborator. John Jumper pointed out that one way to accelerate the pace of science is to improve the productivity of scientists. It’s easy to imagine a GenAI aide that helps with grant writing and progress reports, which can be tedious. Another task would be to identify important new publications of interest to the scientist, ideally customized to the individual to summarize what is new compared to what the scientist already knew. A more powerful AI system like the one Amodei envisions would go beyond what a Scientist’s Aide could do. To keep the human involved to ensure safety and affordability, we pose this advance as a “scientist’s AI collaborator.”

From an employment perspective, we believe that education is elastic, as there is a huge demand for improving the effectiveness and efficiency of learning. Indeed, the U.S. and many other highincome nations face a shortage of K-12 teachers, as well as STEM graduates who could teach those topics in K-12 schools.

Today’s AI tutors such as CK-12 and Khanmigo likely already help some students. A major educational challenge in the U.S. is that K-12 students in high-poverty schools do much worse on standardized tests compared to students in other countries or to U.S. students from low-poverty schools. Selective use of AI tools might track socioeconomic status, which inadvertently could expand the educational gap between students at high-poverty schools versus low-poverty schools.

The next topic is healthcare, responsible for 16% of the U.S. GDP [Vankar]. Like education, many believe that society should offer high-quality healthcare regardless of the wealth of individuals [Einav and Finkelstein].

We believe that healthcare is also elastic: demand for healthcare will increase more than proportionately as the cost and quality of provided healthcare improves [Baumol]. Indeed, the U.S. and many other countries are facing a shortage of healthcare professionals. Beyond improving the employment prospects of healthcare workers via productivity gains, these tools must also keep healthcare professionals in the decision path for actual recommendations for patient therapy, as AI systems are not guaranteed to make the best recommendation 100% of the time. Since people and AI systems tend to make different mistakes, the collaboration of experts with AI systems might help the quality of healthcare.

More generally, technological progress is by definition always a step into the unknown. The more fundamental an innovation, the more unknowns there will be in practice, and the more difficult it will be to apply the proposed policies. Nonetheless, for a great deal of innovative activity, we do have a sense of which factors will benefit and which factors will be hurt. Even if policymakers cannot ascertain this, innovators might be able to. And it may also be possible to guide innovation by committing to implement some of the proposed policies with ex-post measures that are taken once the impact of an innovation is clear.

“what are the economic prospects for most existing humans in such a world? New technologies often bring labor market shocks, and in the past humans have always recovered from them, but I am concerned that this is because these previous shocks affected only a small fraction of the full possible range of human abilities, leaving room for humans to expand to new tasks. AI will have effects that are much broader and occur much faster, and therefore I worry it will be much more challenging to make things work out well.”

Taking the population of the world at any number, a thousand millions, for instance, the human species would increase in the ratio of—1, 2, 4, 8, 16, 32, 64, 128, 256, 512, &c. and subsistence as—1, 2, 3, 4, 5, 6, 7, 8, 9, 10, &c. In two centuries and a quarter, the population would be to the means of subsistence as 512 to 10: in three centuries as 4096 to 13; and in two thousand years the difference would be almost incalculabel, though the produce in that time would have increased to an immense extent.

First, That food is necessary to the existence of man. Secondly, That the passion between the sexes is necessary and will remain nearly in its present state.

Population, when unchecked, increases in a geometrical ratio. Subsistence increases only in an arithmetical ratio. A slight acquaintance with numbers will shew the immensity of the first power in comparison of the second. By that law of our nature which makes food necessary to the life of man, the effects of these two unequal powers must be kept equal.

“Thomas Malthus’ Essay on the Principle of Population (1798) has an oft-quoted statement that neatly summarizes his theory: “Population, when unchecked, increases in a geometrical ratio. Subsistence increases only in an arithmetical ratio.”

The basic idea is that population grows exponentially, but the growth of a society’s means of subsistence is only linear. Linear growth in food supply cannot make up for the skyrocketing needs of an exponentially growing population. Conversely, exponential growth of population outpaces the linear increase of subsistence.

As population grows faster than subsistence, food intake per capita shrinks inexorably. At some point, population growth runs against the hard limit posed by minimum food intake per capita—with tragic results.

“Malthus was aware that population can temporarily exceed its long-term limits. For example, there can be good harvests for twenty or thirty years. In that case, population levels may exceed the longterm subsistence base. Malthus cautioned, however, that this can happen only for so long. The longer such overshoot lasts, the greater the overpopulation. With growing overpopulation, the famine bound to occur after a bad harvest will then be even more devastating. If the famine is not sufficient to decimate population to a viable level, then other calamities such as war and pandemics may do so. A combination of famine, war, pandemics, and deviant behaviour (“vice and misery”) will decimate population to a level consistent with the means of subsistence, until the cycle starts anew.”

“Logically speaking, the solution to the Malthusian problem is closing the gap between exponential growth in population and linear growth in subsistence. This can happen either through “vice and misery” or through a managed solution. One way of achieving a managed solution is to constrain”

Another way is exponential growth in the means of subsistence.

Malthus called for the former and discarded the latter, but modernity has shown that it is possible, at least temporarily, to increase agricultural yields at a pace exceeding population growth.

As it seems, however, most of history has neither followed classical Malthusian nor modern cornucopian patterns. Instead, the most common way of dealing with the problem of overpopulation has been to shift the problem to subalterns and/or outsiders.

By moral restraint, Malthus intended the control of reproductive behaviour in ways acceptable to him as a Protestant Christian.

Married couples should engage in sexual moderation to curtail the number of their offspring, and non-married people should renounce sexual intercourse altogether.

Neo-Malthusian traps are problems that have the same logical structure as the classical Malthusian trap. One function (f1) outpaces and strains another function (f2). The system has some inertia built into it, enabling temporary overshoot. Given the inexorable decline of the second function, however, overshoot leads to significant systemic disruptions, ultimately constraining the first function, and thus bringing the system back to equilibrium—but only after a devastating crisis

“In the classical version of the Malthusian trap, discussed in Section 2, population growth outpaces and strains the means of subsistence. Demographic inertia leads to overpopulation. Given the declining means of subsistence, overpopulation leads to what Malthus called “vice and misery,” constraining and disrupting population growth and thus bringing the system back to equilibrium, but only after a dreadful period of famine, war, pestilence, moral decay, and so on.”

Environmental neo-Malthusianism. According to ecological footprint analysis, overshoot results from society’s environmental impact outpacing and straining ecosystem services, leading to environmental degradation and thus undermining the Earth’s regenerative capacity.37 Unless society finds a way to reduce its impact deliberately to whatever ecosystem services are able to replenish, this is bound to lead to Malthusian calamities that are difficult to fathom and will force humanity’s environmental impact back to sustainable levels.38 • Climate-based neo-Malthusianism. When greenhouse gas emissions outpace the ability of the climate system to absorb them, climate change may cause serious trouble to human civilization, which appears ill-prepared for more than 2 °C of global warming. Especially when accompanied by positive reinforcements after possible tipping points,39 human misery may ultimately force a reduction of climate stresses but the climatic perturbation may last for centuries and the social and political consequences are bound to be catastrophic.40 • Energy-based neo-Malthusianism. Given the heavy dependence of industrial society on energy, fossil fuel depletion may lead to a decline, or even reversal, of economic growth. Insofar as industrial civilization needs economic growth to be sustainable, some authors argue that fossil fuel depletion may trigger the demise of prevailing social and economic models, forcing a new equilibrium at lower levels of energy throughput. It may be possible to reach such an equilibrium in the long term, but only after a long emergency.41

“The reinforcing interaction between technological progress and the size and composition of the population operated persistently throughout history, gradually gaining momentum. Nevertheless, for most of human history, this interaction had a largely negligible long-term impact on income per capita. Eventually, however, as the pace of technological advancement accelerated beyond a critical threshold, education became essential for navigating the rapidly evolving technological landscape, prompting parents to allocate some of their limited resources to their children’s education. Although technological progress increased parental income, enabling a surge in fertility and population growth,8 the allocation of resources toward children’s human capital caused population growth to lag behind technological progress, fueling the onset of economic growth.”

Throughout most of human history, the development process was dominated by Malthusian dynamics. Technological progress and land expansion fueled rising birth rates and declining mortality rates, ultimately leading to proportional increases in both population and resources. Variations in technological progress and land productivity across societies contributed to differences in population densities but had only short-term effects on living standards.6 Yet, larger populations were more likely to give rise to ingenious individuals capable of developing new tools, goods, and practices (Simon 1977; Kremer,1993; Galor 2022), while also generating a greater demand for these innovations (Boserup 1965). Moreover, sizable societies benefited from increased specialization and cross-fertilization, facilitating the rapid dissemination of new technologies. This self-reinforcing feedback loop between population and technology emerged at the dawn of humanity and persisted over most of human history. Technological advances enabled larger populations to be sustained, while increasing population size spurred faster innovations. Nevertheless, during the Malthusian epoch, technological progress was ultimately counterbalanced by population growth, leaving income per capita hovering near the subsistence level in the long run. Just a few centuries ago, human life was often “nasty, brutish, and short” (Hobbes 1651). One in four newborns succumbed to cold, hunger, or illness before reaching their first birthday. Women frequently died during childbirth, and life expectancy rarely surpassed forty years. Most people subsisted on meager, monotonous diets and lived in a state of widespread illiteracy. Amid these grim realities, perhaps even more startlingly, economic crises did not merely demand austeritythey often triggered widespread starvation and societal collapse. Remarkably, most people endured conditions more akin to those of their distant ancestors millennia earlier than to those of their modern descendants. The living conditions of a fifteenthcentury English farmer were strikingly similar to those of a medieval Chinese serf or a Mayan peasant. Similarly, they closely resembled those of an ancient Greek herder, an Egyptian farmer, or even a shepherd in Jericho at the dawn of the Neolithic Revolution.

“The introduction of artificial intelligence (AI) is the continuation of a long process of automation. Advances in mechanization in the late‐nineteenth and early‐twentieth century automated much of the physical labor performed by humans. Advances in information technology in the mid‐ to late‐twentieth century automated much of the standardized data processing that used to be performed by humans. However, each of these past episodes of automation left large areas of work that could only be performed by humans.”

“If AI significantly improves the practice of science and invention and/or creates new high-productivity tasks, such changes may occur in the future. But within a 10-year horizon, it is reasonable to assume that they are constant.”

“Some propose that advances in AI are merely the latest wave in this long process of automation (see e.g. Gordon, 2016). Others, by contrast, emphasize that AI critically differs from past inventions: as artificial intelligence draws closer and closer to human general intelligence, much of human labor runs the risk of becoming obsolete and being replaced by AI in all domains. In this view, progress in artificial intelligence is not only a continuation but the culmination of technological progress – it could lead to a course of history that is markedly different from the implications of previous waves of innovation, and may even represent what James Barrat (2013) has termed “Our Final Invention.””

The second category of technological unemployment is as a transition phenomenon, i.e. when technological change makes workers redundant at a faster pace than they can find new jobs or that new jobs are created. This phenomenon was already observed by Keynes (1930). It is well understood that there is always a certain “natural” or “equilibrium” level of unemployment as a result of churning in the labor market. In benchmark models of search and matching to characterize this equilibrium level of unemployment (see Mortensen and Pissarides, 1994 and 1998), employment relationships are separated at random, and workers and employers need to search for new matches to replace them. The random shocks in this framework can be viewed as capturing, in reduced form, phenomena such as lifecycle transitions but also technological progress in individual firms. In this view, an increase in the pace of technological progress corresponds to a higher job separation rate and results in a higher equilibrium level of unemployment. The transition may be especially prolonged if technology implies that the old skills of workers become obsolete and they need to acquire new skills and/or find out what new jobs match their skills (see e.g. Restrepo, 2017). Even if in the long run, workers adjusted to AI, the transition may be difficult. AI will impact some sectors more than others, and there will be significant job dislocation. As a general lesson, markets on their own are not good at structural transformation. Often, the pace of job destruction is greater than the pace of job creation, especially as a result of imperfections in capital markets, inhibiting the ability of entrepreneurs to exploit quickly new opportunities as they are opened up, which has arguably been the case for globalization in many developing countries.

the difference between human labor and horses is that humans have a comparative advantage in new and more complex tasks. Horses did not. If this comparative advantage is significant and the creation of new tasks continues, employment and the labor share can remain stable in the long run even in the face of rapid automation.

“These exercises suggest that a further wave of automation similar to that of 1980-2016 would increase inequality by even more, because this change would have a bigger impact at the bottom of the wage distribution, given the patterns of comparative advantage implied by 1980-2016 data. AI is predicted to increase inequality by even more, because it expands the set of automated tasks significantly, but does not induce as much “deepening of automation”—meaning productivity improvements in already-automated tasks—as a uniform fall in the price of capital, which tends to increase productivity, benefiting labor of all types. Consequently, low-skill workers are harmed even more by artificial intelligence, while the highest-skill workers continue to benefit, because there is an increase in task services that are complementary to their skills. A sizable minimum wage like $16 an hour, on the other hand, compresses the wage distribution considerably, raising wages at the bottom and reducing them at the top. The wage declines at the top are due to the fact that the minimum wage reduces overall employment and output, which then depresses demand for all tasks.”

The impact of productivity gains on jobs depends on whether the demand for goods produced by that work is elastic or inelastic. If demand is inelastic, productivity gains means jobs will be lost [Bessen]. For example, agriculture is inelastic in the U.S., so gains meant dramatic declines in absolute numbers (fourfold) and its portion of the workforce (from 40% in 1900 to 20% in 1940, 4% in 1970, and 2% today) [Daly]. If product demand is sufficiently elastic, productivity-enhancing technology will increase industry employment [Bessen].

Another perspective on employment is the split between nonphysical tasks and physical tasks. In our view, the main impact of near-term AI systems will be on nonphysical tasks. We think robots will eventually have a large effect on the way in which physical tasks are performed in the world beyond manufacturing, but this may be five or more years behind the use of AI systems for purely digital or knowledge tasks [National Academies].

Here’s how Amodei and others fear the white-collar bloodbath is unfolding:

OpenAI, Google, Anthropic and other large AI companies keep vastly improving the capabilities of their large language models (LLMs) to meet and beat human performance with more and more tasks. This is happening and accelerating.

The U.S. government, worried about losing ground to China or spooking workers with preemptive warnings, says little. The administration and Congress neither regulate AI nor caution the American public. This is happening and showing no signs of changing.

Most Americans, unaware of the growing power of AI and its threat to their jobs, pay little attention. This is happening, too.

And then, almost overnight, business leaders see the savings of replacing humans with AI — and do this en masse. They stop opening up new jobs, stop backfilling existing ones, and then replace human workers with agents or related automated alternatives.

The public only realizes it when it’s too late.

“What will be the effect of this infusion of incredible “human” capital on the economy? Clearly, the most obvious effect will be to greatly increase economic growth. The pace of advances in scientific research, biomedical innovation, manufacturing, supply chains, the efficiency of the financial system, and much more are almost guaranteed to lead to a much faster rate of economic growth. In Machines of Loving Grace, I suggest that a 10–20% sustained annual GDP growth rate may be possible.”