Will AI actually boost productivity and growth?
Who says what
Researchers’ documented positions on “Will AI actually boost productivity and growth?”
Documented positions only, coded with cited evidence by an AI agent: 26 of 400 mapped researchers so far. Everyone else is unknown: no reviewed evidence, which is neither agreement nor neutrality.
Agree (15)
Ajay Agrawal agrees
Views AI as a drop in the cost of prediction — a general-purpose 'bicycle for the mind' with large productivity upside.
Anton Korinek agrees
Works on scenarios where AI automating research produces explosive growth; treats large productivity effects as a live possibility to prepare for.
Christos Makridis agrees
His difference-in-differences study of U.S. administrative data finds that greater occupational AI exposure raises sector output by about 7%, with effects emerging from 2021.
Evidence
- AI, Output, and Employment
- “sectors more exposed to generative AI have experienced large gains in output, employment, and wages” — source
Erik Brynjolfsson agrees
Longtime argument that AI is a general-purpose technology with large productivity potential — gains lag adoption (his 'productivity J-curve') but arrive; now studies transformative AI's growth effects.
Georgios Petropoulos agrees
He argues AI, as a general-purpose technology, will raise labor productivity after a phase of intangible-capital accumulation, and co-authored a 2021 piece predicting a productivity boom.
Joshua Gans agrees
Co-author of the 'prediction machines' view: AI is a general-purpose technology with substantial economic upside.
Daniel Susskind leans agree
His work on AI and work treats technological progress as a source of prosperity, framing the central problem as how to share that prosperity when paid work becomes a less effective channel.
Evidence
- “There is the problem of distribution – how to share prosperity in society when our traditional way” — source
- A World Without Work: Technology, Automation and How We Should Respond
David Autor leans agree
Sees AI as a potentially transformative tool — one that could extend expertise and raise productivity if pointed at worker-complementing uses.
David J. Deming leans agree
A controlled experiment finds that working with AI improves task performance, most for lower-ability participants.
Evidence
- “AI improves performance more for people with low baseline ability.” — source
Frank Neffke leans agree
Documents rapid global diffusion of generative AI in software work with measurable impact.
Jakub Growiec leans agree
His growth-theory work models AI as a factor that can raise long-run growth, and argues transformative AI would either strongly accelerate growth or end it through existential catastrophe.
John McHale leans agree
Co-authored growth model shows that AI improving prediction of promising knowledge combinations could raise discovery rates and economic growth; a later co-authored model studies AI as a worker-augmenting tool raising productivity. Both are conditional theoretical results.
Pamela Mishkin leans agree
Co-authored paper characterizes large language models as general-purpose technologies, estimating that about 80% of U.S. workers could see at least 10% of their tasks affected; it measures exposure and explicitly makes no adoption-timeline prediction.
Paul Milgrom leans agree
In an interview he said AI will likely transform entire economies while noting how little is yet known about large language models.
Prasanna Tambe leans agree
His research finds AI creates the most value as a complement to domain expertise, and a co-authored op-ed expects AI to make workers more productive while adding tasks.
Evidence
- Reskilling the Workforce for AI: Domain Expertise and Algorithmic Literacy
- “a more likely scenario is that AI will create more tasks for workers to do, while also making them more productive” — source
It's complicated (7)
Christian Catalini it's complicated
Models AI as driving the cost of measurable execution toward zero, but argues that human verification capacity, not intelligence, becomes the binding constraint on growth, with outcomes ranging from a 'hollow' to an 'augmented' economy.
Evidence
Diane Coyle it's complicated
Says AI saves considerable time on individual work tasks but has not yet shown up in aggregate productivity; she expects measurable macro-level gains only if organizations adapt, and argues current statistics struggle to capture AI's impact.
Evidence
- “There are a lot of studies now showing that, for individual tasks at work, you save a lot of time using AI.” — source
- Making AI Count: The Next Measurement Frontier
John Van Reenen it's complicated
In a 2026 keynote he noted large AI productivity gains in specific tasks that shrink at the level of whole products or organisations, and argued that, as with earlier general-purpose technologies, economy-wide gains depend on reorganisation and management.
Evidence
- “Research already finds large productivity gains from AI in specific tasks, including coding, writing and customer support. But those gains become much smaller when moving from an individual task to the level of a complete product or organisation.” — source
Joseph Stiglitz it's complicated
Treats AI's productivity gains as a potential that depends on policy, saying that with the right policies higher productivity and less inequality are both possible, while stressing that the objective is well-being rather than measured GDP.
Evidence
- “With the right policies, we could have higher productivity and less inequality, and everybody would be better off.” — source
- Technological Progress, Artificial Intelligence, and Inclusive Growth
Neil Thompson it's complicated
Has said AI's productivity impact is still relatively small because practical application is limited, while expecting it to materialize gradually as applications expand.
Philippe Aghion it's complicated
Models how AI can accelerate growth, but argues gains are constrained by bottleneck sectors that resist automation (Baumol's cost disease).
Sinan Aral it's complicated
His experiments find real AI collaboration gains but also an 'augmentation trap': cognitive offloading that undermines performance.
Disagree (4)
Arvind Narayanan leans disagree
Prominent AI-hype skeptic ('AI Snake Oil'): argues claimed efficiency gains often fail to materialize automatically in real deployments.
Christopher A. Pissarides leans disagree
In 2026 he said he doubts AI will bring a productivity boom comparable to the 1980s-90s computer era, citing little sign of gains so far; in 2023 he had been more optimistic about AI raising productivity.
Daron Acemoglu leans disagree
Argues AI's macroeconomic gains will be modest relative to the hype; his empirical work finds AI's aggregate labor-market and productivity effects too small to detect so far.
Pascual Restrepo leans disagree
Co-authored work argues that automation-biased technological change has gone with lower productivity growth, and that AI's gains depend on creating new tasks.
Evidence
- “Recent technological change has been biased towards automation, with insufficient focus on creating new tasks where labour can be productively employed. The consequences of this choice have been stagnating labour demand, declining labour share in national income, rising inequality and lowering productivity growth.” — source