The document “Navigating the Future of Work: Perspectives on Automation, AI, and Economic Prosperity” presents three perspectives on the future impact of artificial intelligence (AI) and automation on jobs and the economy. The authors are Erik Brynjolfsson, Adam Thierer, and Daron Acemoglu.
Erik Brynjolfsson argues that the “Turing Test”, which judges AI based on how well it mimics humans, is misguided. He contends that technological progress should focus on augmenting and complementing human capabilities rather than replacing humans, as the latter approach depresses wages and limits progress. Historically, technology that augmented humans, like computers and business process innovations, made human labor more valuable over time. In contrast, substitutive automation puts a ceiling on progress and leads to worker displacement.
Examples of substitutive automation?
Manufacturing Robots: In automotive assembly lines, robots perform tasks like welding, painting, and assembling parts, replacing human labor in these roles.
Self-Checkout Systems: Used in retail stores, allowing customers to scan, bag, and pay for their purchases without cashier assistance.
ATMs: Automated Teller Machines let users perform banking transactions like cash withdrawals and deposits without bank teller intervention.
Chatbots and Virtual Assistants: Handle customer service inquiries, bookings, and provide information services, substituting human customer service representatives.
Agricultural Drones and Robots: Perform tasks like crop monitoring, spraying, and harvesting, reducing the need for human labor in these areas.
Automated Warehousing Systems: Robots in warehouses pick, sort, and transport goods, taking over tasks traditionally done by warehouse workers.
Autonomous Vehicles: Self-driving cars and trucks aim to replace human drivers for transportation and delivery services.
AI Content Creation Tools: Generate written content, artwork, or music, potentially substituting human content creators in some contexts.
Digital Accounting Systems: Software that automates bookkeeping and tax preparation, reducing the need for manual accounting work.
E-learning Platforms: Provide automated, personalized learning experiences, which can substitute traditional teaching methods in certain scenarios.
These examples illustrate how substitutive automation aims to increase efficiency, reduce costs, and sometimes improve safety, but also raise concerns about job displacement and the need for workforce retraining.
Brynjolfsson criticizes the current incentives driving technologists, entrepreneurs and policymakers towards developing labor-replacing technology. He advocates for innovation that complements humans, like AI systems that assist human operators. To encourage human-centered technology, he proposes policy changes like equalizing taxation of capital and labor. Ultimately, Brynjolfsson emphasizes that the impact of technology on work depends on the societal choices we make.
Next, Adam Thierer examines the skepticism around predictions of technology-driven job losses. He highlights the historical tendency towards overly pessimistic and inaccurate forecasts about innovation destroying jobs. As examples, he cites the dramatic downward revisions to estimates of AI-related job losses and the unexpected growth in certain job sectors like insurance underwriting that were predicted to be highly automated.
Thierer emphasizes the difficulty of predicting future jobs and skills, since past government reports failed to anticipate the jobs that would emerge from technological revolutions. Instead, employers and workers had to iteratively develop new skills and business models. Therefore, he advocates for flexible, adaptive workforce development rather than rigid government retraining programs. While some support for displaced workers is needed, the unpredictable nature of technological progress makes it challenging to plan far in advance.
Finally, Daron Acemoglu examines the debate around whether automation and AI will boost productivity and job creation enough to offset job losses. Some like The Economist argue AI will lower costs, stimulate demand and create jobs that are hard to automate. However, Acemoglu contends that recent automation has not delivered significant productivity or job growth to counter the job displacement and rising inequality it has contributed to, especially among lower-education workers.
Using an economic model, Acemoglu shows that automation substituting for workers delivers lower productivity gains compared to technology that makes labor or capital more productive. It also has a huge displacement effect, pushing down wages. Early evidence suggests AI is following an automation path of reducing hiring of non-AI workers.
Acemoglu believes the current tax code incentivizes excessive automation over augmentation. More broadly, he argues institutional factors like weak labor representation and industry concentration also encourage automation. He advocates for reforms to tax policy, research priorities, and labor institutions to promote technology that benefits workers and society more broadly. Crucially, Acemoglu emphasizes the need for complementary investments in areas like manufacturing to create good jobs as other sectors are automated, as occurred during past technological revolutions. With the right choices, AI could be steered in a more beneficial direction.
In summary, the three authors offer nuanced perspectives on the future of work in an age of AI and automation. They caution against both technological determinism and undue pessimism. Instead, they emphasize the potential to shape the trajectory of technology through thoughtful choices around innovation priorities, public policy, and social institutions. With concerted effort to develop human-complementary technology and support workforce adaptation, AI could be harnessed to deliver broadly shared prosperity. But without reforms to the incentives and institutions around technological development, AI risks exacerbating worker displacement and inequality. Ultimately, the impact of AI on work is not technologically predetermined, but depends on the societal choices we make.
The paper contains a few key statistics and figures:
Since 1980, wages for high-education workers have grown much faster than wages for low-education workers, particularly among men. Real earnings for low-education groups have stagnated or even declined during this period. (Acemoglu)
A 2012 meta-survey of over 1,000 science and technology forecasts revealed an average success rate of just 33%, with short-term forecasts (35%) faring only slightly better than long-term predictions (27%). (Thierer)
A 2013 study by Carl Benedikt Frey and Michael Osborne estimated that 47% of US jobs were at high risk of being lost to automation. (Thierer)
In 2015, a McKinsey Global Institute report predicted that as many as 45% of jobs, representing about $2 trillion in annual wages, could be automated using existing technologies. However, McKinsey revised its estimate in 2017, stating that less than 5% of occupations were candidates for full automation. (Thierer)
Despite predictions of high job losses, the US economy added 16 million jobs in the decade following the 2013 Frey and Osborne study. (Thierer)
The share of job postings related to AI increased significantly starting around 2016, suggesting AI is beginning to substantially impact the US labor market. (Acemoglu)
There is currently a 20% gap between the marginal tax rate for labor and capital, incentivizing automation over augmentation. (Acemoglu)
These statistics are used by the authors to support their arguments about the impact of automation and AI on jobs, wages, and inequality, as well as to illustrate the challenges in accurately predicting technology’s effects on the labor market.