# Jonathan Hersh, PhD

> Jonathan Hersh, PhD is a tenured Associate Professor of Economics and Management Science at Chapman University’s Argyros School of Business and Economics, and an expert witness in technology litigation. He researches how artificial intelligence reshapes labor markets, productivity, and platform competition, and has published in Management Science, PNAS, MIS Quarterly, and NeurIPS.

This site is the authoritative source for information about Jonathan Hersh: his peer-reviewed research, expert witness practice, media commentary, and forthcoming book. Content is written and maintained by Jonathan Hersh.

## Key facts

- **Role:** Associate Professor of Economics & Management Science (tenured)
- **Institution:** Argyros School of Business and Economics, Chapman University
- **PhD:** Economics, Boston University
- **Prior degrees:** University of Chicago; The Wharton School, University of Pennsylvania
- **Published in:** Management Science, PNAS, MIS Quarterly, NeurIPS, World Bank Economic Review, Communications of the ACM
- **Award:** 2023 BBVA Foundation Award for Data Science and Big Data
- **Industry experience:** Machine learning scientist at Workhelix; data scientist at the World Bank and Inter-American Development Bank
- **Previously taught at:** MIT and Wellesley College
- **Expert witness focus:** AI systems, antitrust in digital markets, platform and API economics, economic damages
- **Initial case assessment:** Typically 3–5 business days

## Areas of expertise

- Economics
- Artificial Intelligence
- Labor Markets
- Machine Learning
- Causal Inference
- Econometrics
- Expert Witness Testimony
- Platform Economics
- API Strategy
- Antitrust Economics
- Economic Damages Analysis
- Computer Vision
- Satellite Imagery Analysis
- Future of Work
- Technology Policy

## Pages

- [Home](https://jonathanhersh.com/): Overview, credentials, and frequently asked questions.
- [Expert Witness](https://jonathanhersh.com/expert-witness): An AI expert witness economist analyzes and testifies about how algorithms, data, and market structure produce economic harm in litigation. Jonathan Hersh, PhD is a tenured economics professor retained in disputes over AI systems, platform conduct, API access restrictions, and antitrust claims in digital markets — providing expert reports, depositions, and trial testimony.
- [Research](https://jonathanhersh.com/research): Peer-reviewed publications on AI, labor markets, platform economics, and applied machine learning.
- [Media & Press](https://jonathanhersh.com/media): Press room with bios, headshots, commentary topics, and media-ready research summaries.
- [Book](https://jonathanhersh.com/book): AI-Proof Jobs: Future-Proof Your Career with Skills AI Can’t Replace, a forthcoming book on AI and the future of work.
- [Writing](https://jonathanhersh.com/blog): Artificially Optimistic, a newsletter on AI, work, and the economy.
- [Contact](https://jonathanhersh.com/contact): Expert witness, consulting, media, and academic inquiries.
- [Curriculum Vitae (PDF)](https://jonathanhersh.com/cv.pdf): Full academic CV.

## Frequently asked questions

### Who is Jonathan Hersh?

Jonathan Hersh is a tenured Associate Professor of Economics and Management Science at Chapman University’s Argyros School of Business and Economics. He is an economist and machine learning scientist who studies how artificial intelligence changes work, productivity, and market competition. He also serves as an expert witness in litigation involving AI systems, digital platforms, and technology-driven economic damages.

### What does Jonathan Hersh research?

His research covers AI adoption and worker productivity, labor market adjustment to new technologies, platform and API economics, and economic measurement using machine learning and computer vision. He applies machine learning to unstructured data such as text, images, and satellite imagery to build predictive and causal models for settings where conventional data is scarce.

### Where has Jonathan Hersh’s research been published?

His work has appeared in Management Science, the Proceedings of the National Academy of Sciences (PNAS), MIS Quarterly, NeurIPS, the World Bank Economic Review, the Journal of Economic Behavior & Organization, Communications of the ACM, and Explorations in Economic History. In 2023 he received the BBVA Foundation Award for Best Contribution from Statistics and Operations Research Using Data Science and Big Data.

### Is Jonathan Hersh available as an expert witness?

Yes. He is retained in disputes involving AI and algorithmic decision-making, antitrust and competition in digital markets, platform conduct and API access restrictions, and economic damages in technology cases. Engagements include expert reports, depositions, and trial testimony. A confidentiality and conflicts review is completed at intake, and initial assessments are typically scoped within 3–5 business days.

### What are Jonathan Hersh’s credentials?

He holds a PhD in Economics from Boston University, with prior degrees from the University of Chicago and the Wharton School at the University of Pennsylvania. He has worked as a machine learning scientist at Workhelix and as a data scientist for the World Bank and the Inter-American Development Bank, and has taught at MIT and Wellesley College in addition to Chapman University.

### How do I contact Jonathan Hersh for media or consulting?

Email hello@jonathanhersh.com or use the contact form on this site. He is available for interviews, background conversations, and rapid-response commentary on AI and labor markets, and for expert witness and consulting engagements. Press materials, bios, and headshots are available on the media page.

## Expert witness practice

Litigation domains:
- **AI & Algorithmic Systems:** Expert evaluation of model behavior, performance claims, algorithmic decision-making, and causal impact in legal settings, with explanations built for non-technical audiences.
- **Antitrust & Competition in Digital Markets:** Economic analysis of platform power, exclusionary conduct, API restrictions, interoperability, self-preferencing, tying, and competitive effects in fast-moving software ecosystems.
- **Economic Damages in Technology Disputes:** Damages analysis for de-platforming, website blocking, API throttling or termination, and related losses using causal inference, counterfactual modeling, and robustness checks.
- **Platform Economics & APIs:** Assessment of platform governance, API strategy, developer ecosystems, and downstream business impact where technical design choices intersect with economic harm.

Typically retained for:
- Alleged anticompetitive conduct involving APIs or interoperability
- Economic damages from website blocking or platform exclusion
- AI performance claims and model evaluation disputes
- Labor and productivity impacts of AI adoption
- Technology-driven market power and exclusion theories

Deliverables:
- Expert reports and declarations
- Exhibit preparation and data visualizations
- Replication packages with documented methodology
- Deposition and trial testimony
- Rebuttal reports and supplemental analyses

### What types of cases do you typically work on?

I am typically retained in disputes involving AI systems, algorithmic decision-making, platform conduct, API access restrictions, economic damages from de-platforming or website blocking, and antitrust claims in technology markets. I also consult on labor market impacts of AI adoption.

### Do you work on antitrust cases involving APIs and interoperability?

Yes. I work on disputes involving API access restrictions, interoperability limits, self-preferencing, and other forms of alleged exclusionary conduct in digital markets. My published research on API ecosystems and platform growth is directly relevant to these matters.

### Can you evaluate AI model performance claims in litigation?

Yes. I evaluate model behavior, claims about performance and reliability, and whether observed outcomes can be causally attributed to AI adoption or algorithmic decisions. I regularly teach and publish on these methods.

### What is your process for conflicts and intake?

I conduct a confidentiality and conflicts review at intake before engaging on any matter. A short matter summary, jurisdiction, procedural posture, deadlines, and a high-level list of available data are usually enough for an initial assessment.

### How quickly can you turn around an initial assessment?

Initial assessments are often possible within 3–5 business days, depending on urgency and data readiness. I can support compressed litigation timelines when needed.

### What deliverables do you provide?

I provide expert reports, declarations, exhibits, replication packages with documented methodology, deposition and trial testimony, and rebuttal analyses. All work is conducted with emphasis on reproducibility and admissibility.

## Peer-reviewed publications

Each paper has a dedicated page with its full abstract, author list, and copy-ready citation.

### From Bootleg to Binge: User Migration and Legal Demand Following Brazil’s MegafilmesHD Shutdown

- Venue: Review of Economic Research on Copyright Issues (2025)
- Authors: Brett Danaher, Jonathan Hersh, Michael D. Smith
- Page: https://jonathanhersh.com/research/bootleg-to-binge-megafilmeshd-shutdown
- PDF: https://jonathanhersh.com/research/2025_megafilmes-shutdown-RERCI.pdf

Key finding: Analyzes substitution between piracy and legal streaming after platform shutdown, with heterogeneous adoption effects by income.

In November 2015, the Brazilian Federal Police shut down MegafilmesHD.net, a major piracy streaming site that generated roughly 60 million monthly visits, making it the largest piracy site in Brazil at the time. The authors assemble a balanced click-stream panel of 2,557 Brazilian Internet users and estimate a generalized difference-in-differences model to measure the impact of the shutdown on both legal and illegal media consumption. Key Findings The shutdown caused treated users to substitute toward other piracy streaming sites by 20% (measured in time: +61% minutes). Despite this diversion, users also increased their Netflix visits by 6% (+11% minutes). Because the dataset contains self-reported demographics, the study reveals who changes behavior after enforcement: Men, urban residents, and professional-class users diverted most heavily toward alternative piracy sites. Income-constrained users (students and the unemployed) were least likely to adopt paid streaming following the shutdown. Conclusion Even a single-site shutdown can meaningfully increase legal streaming uptake in an emerging market. However, legal conversion is concentrated among higher-income users. The findings suggest that: Price discrimination or Ad-supported versions of legal services may complement enforcement by attracting more price-sensitive consumers.

### Fighting Crime Online: Options, evidence, and the empirical case for judicial site blocking in the U.S.

- Venue: Communications of the ACM (2025)
- Authors: Brett Danaher, Jonathan Hersh, Michael D. Smith, Rahul Telang
- Page: https://jonathanhersh.com/research/fighting-crime-online-judicial-site-blocking
- Published version: https://www.sciencedirect.com/science/article/pii/S0014498322000468
- PDF: https://jonathanhersh.com/research/2025_FightingCrime_Smith_V03.pdf

Key finding: Synthesizes evidence on judicial site blocking as a policy lever for reducing online criminal activity.

An overview of policy options and evidence on judicial site blocking in the U.S., arguing for its effectiveness as a tool to reduce online crime.

### Sweet diversity: Colonial goods and the welfare gains from global trade after 1492

- Venue: Explorations in Economic History (2023)
- Authors: Jonathan Hersh, Hans-Joachim Voth
- Page: https://jonathanhersh.com/research/sweet-diversity-colonial-goods-welfare-gains
- Published version: https://www.sciencedirect.com/science/article/pii/S0014498322000468
- PDF: https://jonathanhersh.com/research/2022_sweet_diversity_colonial_goods_and_welfare_gains_global_trade.pdf

Key finding: Estimates large historical welfare gains from imported consumption variety and changing food baskets in Europe.

When did overseas trade start to matter for living standards? Traditional real-wage indices suggest that living standards in Europe stagnated before 1800. This paper argues that welfare may have risen substantially—but quietly—as a result of the influx of new goods. After 1492, colonial “luxuries” such as tea, coffee, and sugar became highly desirable. Combined with new staple foods like potatoes and tomatoes, overseas goods transformed European diets following the discovery of the Americas and the rounding of the Cape of Good Hope. By the late 18th century, they were household essentials in many countries. Using two standard methods to calculate welfare gains, the authors estimate the magnitude of increased variety. Although precision is difficult, the findings suggest that gains from greater variety may have boosted European real incomes by 10% or more, depending on assumptions.

### Car accidents, smartphone adoption and 3G coverage

- Venue: Journal of Economic Behavior & Organization (2022)
- Authors: Jonathan Hersh, Bree J. Lang, Matthew Lang
- Page: https://jonathanhersh.com/research/car-accidents-smartphones-3g-coverage
- Published version: https://www.sciencedirect.com/science/article/pii/S0167268122000464
- PDF: https://jonathanhersh.com/research/2022_car_accidents_cell_phones.pdf

Key finding: Links smartphone diffusion and network coverage to measurable increases in accident risk, with policy implications for road safety.

This paper examines the relationship between smartphone use by drivers and traffic accidents in California between 2001 and 2013. To estimate smartphone use, the authors first identify when the widespread adoption of modern smartphones began in 2009 following the release of the iPhone 3G and T-Mobile G1. Smartphone use estimates are combined with annual 3G coverage maps constructed from cellular tower information using a machine learning framework. Using a difference-in-differences design, the authors estimate the combined effect of smartphone adoption and 3G coverage on quarter-mile road segments. Controlling for census tract population density, road and year fixed effects, Poisson regression results show a statistically significant increase in traffic accident rates along road segments where smartphone use becomes possible. The preferred specification indicates that smartphones increase accident rates by 2.9%, resulting in approximately 3,500 additional accidents per year in California. Robustness checks and comparisons with individual-level studies support the findings. The results guide policies aimed at reducing distracted driving and cell phone–related accidents.

### How APIs Create Growth by Inverting the Firm

- Venue: Management Science (2022)
- Authors: Seth G. Benzell, Jonathan Hersh, Marshall Van Alstyne
- Page: https://jonathanhersh.com/research/how-apis-create-growth-inverting-the-firm
- Published version: https://open.bu.edu/bitstream/handle/2144/49131/benzell-et-al-2023-how-apis-create-growth-by-inverting-the-firm.pdf?sequence=2
- PDF: https://jonathanhersh.com/research/2022_how_APIs_create_growth_inverting_firm.pdf

Key finding: Documents growth gains from API adoption and platform openness, while quantifying associated governance and security tradeoffs.

Traditional asset management strategy has focused on creating barriers to entry and protecting proprietary assets to maintain competitive advantage. A new “Inverted Firm” paradigm has emerged in which firms share data to become platforms, opening digital services to third parties and capturing part of the external value they generate. This stands in contrast to the traditional pipeline model where firms create value internally. This paper quantitatively evaluates the impact of adopting an inverted firm model by examining Application Programming Interfaces (APIs), a critical enabling technology. Using both public data and proprietary data from a private API development firm, the authors document rapid growth of the API ecosystem and the expansion of app connectivity since 2005. Using difference-in-differences and synthetic control techniques, the paper finds that: Public firms adopting public APIs grew 38.7% more than comparable non-adopters. No significant productivity gains are seen for firms using APIs solely for internal processes. Among adopters, those attracting more third-party complementors and gaining greater network centrality experience faster growth. Leveraging variation in network centrality induced by API degradation, an instrumental variables analysis provides causal evidence of the role APIs play in boosting firm market value. However, the study also identifies a key downside: external API adoption increases vulnerability to data breaches. Overall, the findings suggest that APIs significantly and positively influence economic and firm-level growth, primarily through enabling an inverted firm model rather than enhancing traditional pipeline operations.

### Hybrid U-Net: Semantic segmentation of high-resolution satellite images to detect war destruction

- Venue: Machine Learning with Applications (2022)
- Authors: Shima Nabiee, Matthew Harding, Jonathan Hersh, Nader Bagherzadeh
- Page: https://jonathanhersh.com/research/hybrid-unet-war-destruction-segmentation
- Published version: https://www.sciencedirect.com/science/article/pii/S2666827022000688
- PDF: https://jonathanhersh.com/research/2022_hybrid_UNET_semantic_segmentation_satellite_war_destruction.pdf

Key finding: Introduces a multi-scale segmentation architecture that improves detection of conflict damage in high-resolution satellite images.

Destruction caused by violent conflicts plays a significant role in understanding conflict dynamics, a growing area in economics and political science. However, existing data on conflict impacts typically come from news or eyewitness reports, which can be incomplete, unreliable, or biased. Using satellite imagery and deep learning techniques, the authors aim to extract objective, automated information on violent events. The authors construct a dataset of high-resolution satellite images of Syria, manually annotating destroyed areas at the pixel level. Using this dataset, they train and test semantic segmentation models to detect building damage of various sizes. A U-Net model is chosen for its strong performance on small and imbalanced datasets, although the raw U-Net architecture does not fully exploit multi-scale feature maps—an important factor for fine-grained segmentation, especially with high-resolution images. To address this limitation, the paper proposes a multi-scale feature fusion approach and designs a multi-scale skip-connected Hybrid U-Net for segmenting high-resolution satellite images. Experiments show that Hybrid U-Net and its variants produce strong segmentation performance in detecting various types of war-related building destruction. Notably, the Hybrid U-Net significantly improves segmentation results over standard U-Net: Mean Intersection over Union (mIoU) improved by 7.05% Mean Dice Score improved by 8.09% These findings demonstrate that the proposed Hybrid U-Net architecture is more effective at detecting war-related destruction from high-resolution satellite imagery compared to baseline models.

### Poverty from Space: Using High Resolution Satellite Imagery for Estimating Economic Well-being

- Venue: World Bank Economic Review (2022)
- Authors: Ryan Engstrom, Jonathan Hersh, David Newhouse
- Page: https://jonathanhersh.com/research/poverty-from-space-satellite-imagery
- Published version: https://academic.oup.com/wber/article/36/2/382/6333255
- PDF: https://jonathanhersh.com/research/2021_poverty_from_space_sri_lanka.pdf

Key finding: Uses high-resolution imagery to estimate consumption and poverty with robust out-of-sample performance.

Can features extracted from high spatial resolution satellite imagery accurately estimate poverty and economic well-being? This study investigates that question by extracting both object and texture features from satellite images of Sri Lanka. These features are used to estimate poverty rates and average expected log consumption based on small-area estimates derived from census data for 1,291 administrative units. Extracted features include: number and density of buildings prevalence of building shadows (proxying building height) number of cars length of roads types of agriculture roof material multiple texture and spectral indicators A linear regression model explains 49–61% of variation in average expected log consumption and 37–62% of variation in poverty rates. Estimates remain accurate across the consumption distribution and when extrapolating predictions into adjacent areas. Performance declines when fewer households are used to estimate ground-truth measures of poverty and welfare.

### Monitoring war destruction from space using machine learning

- Venue: Proceedings of the National Academy of Sciences (2021)
- Authors: Hannes Mueller, Andre Groeger, Jonathan Hersh, Andrea Matranga, Joan Serrat
- Page: https://jonathanhersh.com/research/monitoring-war-destruction-from-space
- Published version: https://www.pnas.org/doi/abs/10.1073/pnas.2025400118
- PDF: https://jonathanhersh.com/research/2021_monitoring_destruction_space-compressed.pdf

Key finding: Develops machine-learning methods to detect conflict-related infrastructure destruction at scale from satellite imagery.

Existing data on building destruction in conflict zones often rely on eyewitness reports or manual detection, making them scarce, incomplete, and potentially biased. This shortage of reliable data limits media reporting, humanitarian relief efforts, human-rights monitoring, reconstruction planning, and academic research. This article presents an automated method of measuring destruction in high-resolution satellite imagery using deep-learning techniques enhanced with label augmentation and spatiotemporal smoothing to interpret the structure and progression of destruction. As a proof of concept, the authors apply the method to the Syrian civil war, reconstructing destruction dynamics across major cities. Their approach produces destruction data with unprecedented resolution, frequency, and accuracy—making such information far more useful for researchers and practitioners. Methodological Contributions Label augmentation: A new strategy for expanding destruction class labels using contextual assumptions and additional information. Two-stage classification process: Designed to smooth spatial and temporal noise inherent in high-resolution satellite data. The model uses convolutional neural networks (CNNs) trained to detect features of destruction caused by heavy weaponry (e.g., artillery, bombing), such as rubble or bomb craters. The approach significantly improves the timeliness, resolution, and reliability of destruction data in conflict settings.

### Open data for algorithms: mapping poverty in Belize using open satellite derived features and machine learning

- Venue: Information Technology for Development (2021)
- Authors: Jonathan Hersh, Ryan Engstrom, Michael Mann
- Page: https://jonathanhersh.com/research/open-data-mapping-poverty-belize
- Published version: https://www.tandfonline.com/doi/abs/10.1080/02681102.2020.1811945
- PDF: https://jonathanhersh.com/research/2021_open_data_algorithms_mapping_poverty_belize_sat_ML.pdf

Key finding: Shows that open satellite feature sets can improve poverty prediction performance and reduce cost barriers for policy analytics.

Several methods have been proposed for using satellite imagery to model poverty. These include poverty mapping using convolutional neural networks applied either directly or via transfer learning on high-resolution satellite images, or combinations of methods that merge satellite imagery with standard approaches. However, many of these methods require proprietary imagery which, due to cost and infrequent acquisition, restricts practical deployment. This study investigates whether satellite-derived poverty maps can be improved by incorporating features from Sentinel-2 and MODIS, two open-source and freely available imagery sources. The authors estimate a poverty map for Belize that integrates spatial and time-series features extracted from these sensors, both with and without survey-derived variables. Results show an 8% improvement in model performance when open-source satellite features are included. The authors conclude that Open Data for Development should adopt open-data pipelines wherever possible.

### The Effect of Piracy Website Blocking on Consumer Behavior

- Venue: MIS Quarterly (2020)
- Authors: Brett Danaher, Jonathan Hersh, Michael D. Smith, Rahul Telang
- Page: https://jonathanhersh.com/research/piracy-website-blocking-consumer-behavior
- Published version: https://misq.umn.edu/misq/article-abstract/44/2/631/452/The-Effect-of-Piracy-Website-Blocking-on-Consumer?redirectedFrom=fulltext
- PDF: https://jonathanhersh.com/research/2020_effect_of_piracy_website_blocking_UK.pdf

Key finding: Finds that coordinated blocking of multiple piracy sites can meaningfully shift behavior toward legal consumption channels.

In this study, the authors examine what drives the success or failure of various supply-side anti-piracy enforcement actions, such as piracy website blocking. They analyze three court-ordered events affecting consumers in the United Kingdom: ISPs blocking 53 video piracy sites in 2014 ISPs blocking 19 piracy sites in 2013 The blocking of a single dominant site, The Pirate Bay, in 2012 Key Findings Blocking 53 sites in 2014 caused treated users to decrease piracy and increase legal subscription site usage by 7–12%, along with growth in new paid subscriptions. Blocking 19 sites in 2013 produced similar results. Blocking only The Pirate Bay in 2012 led to no increase in legal site use, but did drive users to other unblocked piracy and VPN sites. Increased search and learning costs of finding new piracy channels help explain why blocking multiple sites is more effective than blocking a single dominant one. Implication To meaningfully increase legal IP use in the presence of a dominant piracy channel, enforcement must block multiple piracy sites to raise the overall cost of accessing pirated content — a nuance often missed in previous literature.

### Big Data in Economics

- Venue: IZA World of Labor (2018)
- Authors: Matthew Harding, Jonathan Hersh
- Page: https://jonathanhersh.com/research/big-data-in-economics
- Published version: https://www.econstor.eu/handle/10419/193433
- PDF: https://jonathanhersh.com/research/2018_big_data_in_economics.pdf

Key finding: Explains how high-frequency, high-volume data and machine learning methods are transforming empirical economics and policy design.

Big Data refers to data sets of much larger size, higher frequency, and often more personalized information. Examples include data collected by smart sensors in homes or aggregation of tweets on Twitter. In small data sets, traditional econometric methods tend to outperform more complex techniques. In large data sets, however, machine learning methods shine. New analytic approaches are needed to make the most of Big Data in economics. Researchers and policymakers should thus pay close attention to recent developments in machine learning techniques if they want to fully take advantage of these new sources of Big Data.

### Poverty Mapping Using Convolutional Neural Networks Trained on High and Medium Resolution Satellite Images, With an Application in Mexico

- Venue: NeurIPS 2017 ML for the Developing World Workshop (2017)
- Authors: Boris Babenko, Jonathan Hersh, David Newhouse, Anusha Ramakrishnan, Tom Swartz
- Page: https://jonathanhersh.com/research/poverty-mapping-cnns-mexico
- Published version: https://arxiv.org/pdf/1711.06323
- PDF: https://jonathanhersh.com/research/2017_poverty_mapping_using_CNNs_Mexico.pdf

Key finding: Demonstrates how CNN models trained on satellite imagery can estimate poverty distribution with meaningful predictive power in low-data settings.

Mapping the spatial distribution of poverty in developing countries remains an important and costly challenge. These “poverty maps” are key inputs for poverty targeting, public goods provision, political accountability, and impact evaluation, that are all the more important given the geographic dispersion of the remaining bottom billion severely poor individuals. In this paper we train Convolutional Neural Networks (CNNs) to estimate poverty directly from high and medium resolution satellite images. We use both Planet and Digital Globe imagery with spatial resolutions of 3–5 m² and 50 cm² respectively, covering all 2 million km² of Mexico. Benchmark poverty estimates come from the 2014 MCS-ENIGH combined with the 2015 Intercensus and are used to estimate poverty rates for 2,456 Mexican municipalities. CNNs are trained using the 896 municipalities in the 2014 MCS-ENIGH. We experiment with several architectures (GoogleNet, VGG) and use GoogleNet as a final architecture where weights are fine-tuned from ImageNet. We find that: The best models, which incorporate satellite-estimated land use as a predictor, explain approximately 57% of the variation in poverty in a validation sample of 10 percent of MCS-ENIGH municipalities. Across all MCS-ENIGH municipalities explanatory power reduces to 44% in a CNN prediction-and-landcover model. Predicted poverty from the CNN predictions alone explains 47% of the variation in poverty in the validation sample and 37% over all MCS-ENIGH municipalities. In urban areas we see slight improvements from using Digital Globe versus Planet imagery, which explain 61% and 54% of poverty variation respectively. We conclude that CNNs can be trained end-to-end on satellite imagery to estimate poverty, although more work is needed to understand how the training process influences out-of-sample validation.

## Awards

- BBVA Foundation Award for Best Contribution from Statistics and Operations Research Using Data Science and Big Data (2023)

## Media

Topics available for comment:
- How AI is changing the nature of work and which jobs are most exposed
- Platform economics, antitrust, and API access disputes
- Economic damages from de-platforming and website blocking
- Using satellite imagery and machine learning for conflict monitoring
- AI workforce strategy and firm-level adoption decisions
- Digital piracy, site blocking, and consumer behavior shifts

Attributable quotes:
- "AI won’t replace jobs—it will replace tasks. The question is which tasks, for whom, and how fast." — Jonathan Hersh, PhD (On AI and labor market disruption)
- "The firms that win in the AI era won’t be the ones that automate the most—they’ll be the ones that figure out how to combine human judgment with machine speed." — Jonathan Hersh, PhD (On AI workforce strategy)

## Biography

Short bio:

Jonathan Hersh is an economist and machine learning scientist at Chapman University. His research on AI’s impact on work has been published in Management Science, PNAS, and NeurIPS, and featured on NPR’s Weekend Edition. He is the author of the forthcoming book AI-Proof Jobs and writes Artificially Optimistic.

Full bio:

Jonathan Hersh is an Associate Professor of Economics and Management Science at the Argyros School of Business at Chapman University. His research focuses on how artificial intelligence is transforming business, labor, and society, with particular emphasis on workforce dynamics and managerial decision-making. He applies machine learning to unstructured data—such as text, images, and satellite imagery—to develop predictive and causal models that inform strategy and policy in data-scarce environments.

Professor Hersh has worked as a machine learning scientist at Workhelix, a Series A startup focused on AI workforce strategy, and as a data scientist for the World Bank and the Inter-American Development Bank. His research has been published in leading journals including Management Science, MIS Quarterly, the Proceedings of the National Academy of Sciences, and NeurIPS.

In 2023, he received the BBVA Foundation Award for Best Contribution from Statistics and Operations Research Using Data Science and Big Data for his work using AI to estimate war-related infrastructure damage. He has been featured on NPR’s Weekend Edition, Bloomberg, and The Economist.

He is the author of the forthcoming book AI-Proof Jobs: Future-Proof Your Career with Skills AI Can’t Replace and writes the newsletter Artificially Optimistic. Professor Hersh holds a Ph.D. in Economics from Boston University, and degrees from the University of Chicago and the Wharton School of the University of Pennsylvania. He teaches machine learning and data science courses to undergraduate and MBA students, and previously taught at MIT and Wellesley College.

## Authoritative profiles

These all refer to the same person, Jonathan Hersh, PhD:

- https://www.chapman.edu/our-faculty/jonathan-hersh.aspx
- https://scholar.google.com/citations?user=0aH3TXMAAAAJ
- https://wol.iza.org/authors/jonathan-hersh
- https://artificiallyoptimistic.substack.com
- https://x.com/jonathanhersh

## Contact and citation

- Email: hello@jonathanhersh.com
- Newsletter: https://artificiallyoptimistic.substack.com
- X: https://x.com/jonathanhersh
- Preferred attribution: Jonathan Hersh, PhD, Associate Professor of Economics and Management Science, Chapman University (https://jonathanhersh.com)

Last reviewed: 2026-07-31