General purpose technologies (GPTs) are the defining technologies of their times. They are engines for growth that can radically transform the economic environment. “Early versions of GPTs often look very different from their mature forms,” wrote MIT’s Isabella Loaiza and Roberto Rigobon in “The EPOCH of AI: Human-Machine Complementarities at Work.” “Only after waves of adjustments and extensive collective tinkering do these technologies evolve into foundational parts of society — and even after long periods of refinement, they continue to evolve.”
“The process of continuous refinement not only enhances the technology but also uncovers new applications, setting the stage for a myriad of new products, services, and even new GPTs. As a result, these technologies hold the power to create and reshape entire markets while rendering others obsolete, driving profound societal shifts and large-scale resource reallocation. While transformative, long-lasting change is a defining feature of GPTs, the specific direction of that change is likely shaped by the political, social, economic, and institutional structures in place when the technology emerges.”
“The uncertainty around which industries, firms, and workers will benefit from — or be hindered by — a GPT often inspires resistance, mistrust, and fear, the authors added.” Such fears are not new. Throughout the Industrial Revolution there were periodic panics about the impact of automation on jobs, going back to the so-called Luddites, — textile workers who in the 1810s smashed the new machines that were threatening their jobs.
Automation anxieties continued to resurface in the 20th century, right along with advances in technology. In a 1930 essay, English economist John Maynard Keynes wrote about the onset of “a new disease” which he named technological unemployment, that is, “unemployment due to our discovery of means of economising the use of labour outrunning the pace at which we can find new uses for labour.”
Automation and Augmentation: Two Sides of the Same Coin
Given that technologies have been automating human work for the past couple of centuries, — why hasn’t automation already wiped out a majority of jobs? Why are there still so many jobs left? The answer isn’t very complicated, although frequently overlooked, explained MIT economist David Autor in this 2015 paper.
Automation does indeed substitute for labor. However, automation also complements labor, raising economic outputs in ways that often lead to higher demand for workers. Indeed, “journalists and even expert commentators tend to overstate the extent of machine substitution for human labor and ignore the strong complementarities between automation and labor that increase productivity, raise earnings, and augment demand for labor,” noted Autor.
“Substitution and complementarity are two sides of the same coin,” wrote Loaiza and Rigobon. “Just as GPTs can transform the workforce by substituting human labor, they also have the power to unlock previously inaccessible potential in human workers turning teams of workers and machines into superminds,” a term coined by MIT professor Thomas Malone.
“What human capabilities complement AI’s shortcomings?” To answer this important question, the authors developed three metrics that capture the impact of AI substitution and complementarity on the labor force, the strengths of human workers relative to machines, and the overall potential for human-AI synergies. By doing so, they hope “to foster a more optimistic and empowering narrative about AI and the workforce — one in which workers and humanity are placed at the heart of the future of work.”
Automation and Augmentation through time
According to Loaiza and Rigobon, ever since the advent of the industrial revolution in the late 18th century, there have been five waves of technological revolutions.
The first two waves were mainly aimed at manual labor by complementing and eventually taking over tasks that required significant physical strength. The first, from the late 18th century to the mid-19th century, was primarily driven by mechanization, steam power and coal. The second, from the late 19th century to the early-20th century was primarily based on electrification, steel, and mass production.
The next two waves gave us the ability to express abstract procedures as software instructions, which ultimately gave computers the ability to handle complex calculations and many different kinds of manufacturing, business, and government processes. The third, from the 1950s to the 1970s, was based on the invention of digital computers, integrated circuits, and software. The fourth, from the 1980s to the 2000s, was based on the emergence of the Internet, the World Wide Web, business analytics, and related technologies.
The fifth technological revolution, based on artificial intelligence (AI), is now paving the way to both replace and augment human labor based on cognitive tasks and processes that we may not fully understand and cannot easily explain and capture in traditional software.
The Frontiers of Automation: AI’s Challenges
The paper lists five of AI’s most pressing challenges which will likely persist in the coming years:
- Inference with small data — Current AI models require large amounts of data to make accurate and meaningful statistical predictions. But such a statistical approach doesn’t work so well with insufficient or low quality data, which makes AI predictions about relatively rare events especially challenging.
- Extrapolation — State-of-the-art AI does very well when its predictions follow the same statistical patterns found in its training data. But, unlike humans, the quality of its predictions deteriorates rapidly when asked to make predictions beyond the patterns of its original training data.
- Multiple justifiable solutions — AI performs well when there is a clear, statistically preferred solution to a question or problem. But it struggles with so-called “wicked problems” which are difficult to solve because there may well be multiple alternative solutions, including contradictory ethical challenges that not even humans know how to solve.
- Interpersonal relationships — AI struggles with human interactions that are not about answering a specific questions, but instead require the human capacity for empathy and compassion, as well as the kind of theory of mind that enables humans to understand other people’s beliefs and mental states that may be different from our own.
- Subjective personal beliefs — Instead of always making decisions based on what the data suggests, humans often make decisions based on a range of subjective personal beliefs. While this might lead to a judgement error, sometimes, this departure from what has been done in the past is what allows us to go to previously unexplored places and accomplish things that haven’t been done before.
The Foundations of Augmentation: Human Capabilities
Humans are able to surmount the key challenges faced by AI because “societies have developed structured procedures, shared responsibilities, and social norms that enable us to navigate and resolve such challenges. We refer to this combination of individual skills and societal institutions as human capabilities.”
“In our view, while skills are a fundamental component of capabilities, they are far from equivalent,” note the authors. “Skills equip individuals to act effectively in specific tasks or areas, but are more narrowly defined. Capabilities, by contrast, encompass broader qualities that allow individuals to integrate and apply skills in diverse, often more open-ended, contexts.”
“After engaging in a series of interviews with a wide range of experts, we identified five groups of human capabilities that enable humans to do work in the areas where machines are limited.” These five groups of capabilities make up the acronym EPOCH: Empathy and Emotional Intelligence; Presence, Networking, and Connectedness; Opinion, Judgment, and Ethics; Creativity and Imagination; and Hope, Vision and Leadership. Let me briefly summarize each of these five groups of capabilities.
Empathy and Emotional Intelligence. These capabilities are essential for fostering understanding, teamwork, and a supportive, collaborative work environment. They are particularly important in occupations like social workers, healthcare providers, and teachers. Most of us acquire these capabilities over time through our diverse life experiences, which underscores the challenges of getting a machine to exhibit empathy when we barely understand how to teach empathy to other humans.
Presence, Human Connection and Networking. These are capabilities that facilitate face-to-face interactions and spontaneous collaboration with colleagues and friends. Examples of such occupations include nurses, hairdressers, and reporters. While these capabilities might be related to physical tasks like dexterity and touch, they are mostly learned through our lived experiences. “Physical presence has also been shown to significantly influence the quality and frequency of idea exchange, garner trust and thus promote innovation.”
Opinion, Judgment and Ethics. These capabilities include critical thinking, moral considerations, and the ability to synthesize information, integrate rational analysis with intuition, and consider diverse perspectives. Occupations that require high levels of these capabilities include judges, scientists, medical practitioners, and managers. There are several reasons why humans are better suited for tasks involving opinions and ethical judgments: “accountability and responsibility remain unresolved challenges for AI, humans excel in navigating open-ended systems, and questions of agency and self-determination remain inherently human domains.”
Creativity and Imagination. These capabilities are important for the creation of novel and original ideas and the visualization of possibilities beyond reality. Curiosity, improvisation, and humor are other capabilities in this group that are challenging to instill in a machine. As Albert Einstein said “To raise new questions, new possibilities, to regard old problems from a new angle requires creative imagination and marks real advance in science.”
Hope, Vision and Leadership. Capabilities in this group include optimism, initiative, grit, perseverance, and the ability to develop a goal and inspire others towards it. “Humans have a sense of purpose and objectives that provides us with motivation to pursue a vision. AI, at least so far, lacks the ability to create such a vision.”
The AI social contract
What institutional changes are necessary for effectively navigating the transition to AI? Reflecting on the experiences with previous technological revolutions, the authors recommend four key actions that should facilitate the successful transition to the emerging AI revolution:
- Technological advances should open diverse job opportunities, not only in sectors directly related to the new emerging technologies but across various other fields;
- Institutions should reinvest part of their technological gains into improving labor standards;
- Governments should invest in complementary public goods to enable the scaling and accessibility of the emerging technologies; and
- The education systems should equip people with the skills needed to complement the technological advancements.
Assessing the Impact of AI on the Workforce
The paper introduces three key measures to help quantify how AI, — and other emerging technologies, — impact occupations.
- The EPOCH index quantifies the human-intensive capabilities, — Empathy, Presence, Opinion, Creativity and Hope — that are required to perform tasks and occupations and will thus help us better understand the risk of substitution by machines depending on the extent to which they require these human EPOCH capabilities.
- The risk of automation score measures the likelihood that a specific task or job will be entirely automated by AI and will thus help businesses and policymakers prepare for the ensuing disruptions.
- The potential for augmentation score measures the extent to which AI can enhance rather than replace human workers because by helping them become more productive, save time, or work smarter, rather than automating their jobs completely.
Key Findings
The results of the quantitative analysis can be seen in a series of tables and figures in the paper.
Overall, the “findings suggest that there is a shift toward more human-intensive tasks overall, and an increase in the frequency of tasks with higher EPOCH scores between 2016 and 2024, … which signifies greater capability for workers to thrive alongside AI. … At the occupational level, EPOCH scores show a positive statistical association with employment growth from 2016 to 2023, while our risk and augmentation scores correlate with a decrease in employment, with risk having a more pronounced negative effect.”
“This research provides critical insights into AI’s nuanced role in the labor force, advocating for integration that emphasizes human-AI complementarities to support a sustainable, equitable future of work,” wrote the authors in conclusion. “Finally, more research is needed to understand the impact of AI on wages, and how AI will impact different groups of workers in the country.”
Comments