We Don't Know What We Don't Know

One fascination I have with public policies is how often we have no idea whether our policies actually work. That perhaps is not surprising, given the complexity of most “human, civic and social problems.” And, for many reasons, not the least of which is ethical, we never can do controlled studies. 

Some of that uncertainty can be seen in public policies to support home broadband, where we still do not have conclusive and consistent evidence that municipal networks actually produce outcomes greater than the opportunity costs and actual investment.  

Study / report	Year	Geography	Method	Headline finding (summary)
Christopher S. Yoo & Timothy Pfenninger, “Municipal Fiber in the United States: A Financial Assessment” (UPenn)	2017 (report); published versions 2022	United States (sample of municipal FTTH projects)	Financial statement analysis of 20 municipal fiber projects (multi-year cash flow and debt repayment projections)	Found 11 of 20 municipal fiber projects generated negative cash flow over the sample period; only 2 of 20 were on track to recover total project costs within expected useful life — authors conclude many municipal projects would not cover costs without subsidies or external support. (Penn Carey Law)
Casey J. Mulligan / Jonathan Kolko (Public Policy Institute of California), “Does Broadband Boost Local Economic Development?” (Kolko, PPIC)	2010	U.S. counties / metro areas (United States)	Econometric analysis of broadband penetration vs local economic indicators	Concluded broadband expansion had limited measurable effects on local employment and wages in their models — economic benefits to residents appear limited and do not clearly outweigh large public deployment costs in some settings. (Public Policy Institute of California)
Grant S. Ford, “The rewards of municipal broadband: An econometric assessment” (Journal article / working paper)	2021	U.S. cities with municipal investments	Econometric evaluation of labor-market / economic outcomes after municipal broadband investment	Found no economically or statistically significant effect of municipal broadband on labor-market outcomes — casts doubt on large local economic returns sufficient to justify big public subsidies. (ScienceDirect)
C. S. Yoo (earlier working material / analyses summarized in press), “Municipal Fiber in the United States: An Empirical Assessment of Financial Performance” (UPenn summary & press)	2017 (widely reported)	Sample U.S. municipal FTTH projects	Empirical accounting of cash flows, break-even projections	Reported multiple high-profile municipal projects that would not repay costs within realistic timeframes (e.g., extremely long payback estimates for some cities), concluding that fiscal risks to municipalities can be material without subsidies. (Penn Carey Law)
ITIF / policy analyses (myth-debunking & affordability critiques), “Are High Broadband Prices Holding Back Adoption? / Broadband Myths” (ITIF)	2021	United States (policy analysis)	Policy literature review & data analysis	Argues that affordability/subsidy programs are likely to be a blunt tool in many contexts; recommends targeted subsidies instead of broad infrastructure subsidies because wide public subsidies may not be cost-effective in driving adoption or economic gains. (Policy critique relevant to subsidy cost-effectiveness.) (ITIF)

The issue, in all cases, is that careful investigators do point out that correlation is not causation. 

They argue that there might be a correlation between higher home broadband investment and economic outcomes, though not suggesting the home broadband investment “caused” the increases. 

The broad problem is that it never is clear whether home broadband investment follows economic growth and reflects it, or somehow enables it. Economic growth, when it happens, is likely the result of a lot of interconnected causes, and home broadband might not even be among the drivers. 

Study / report	Year	Geography	Method	Headline finding (short)
Qiang, Rossotto & Kimura (World Bank — Information and Communications for Development)	2009	120 countries (developed + developing)	Cross-country growth regressions (endogenous growth framework)	Found broadband diffusion associated with higher GDP growth: estimated sizable positive effects of broadband penetration on GDP per capita for both developing and developed countries. (World Bank)
Koutroumpis — The economic impact of broadband on growth (Oxford / OECD analyses)	2009 (and follow-ups)	OECD countries (multi-country panels)	Simultaneous macro + micro modelling / panel IV	Estimates that faster broadband adoption and higher speeds measurably raise GDP — e.g., a 10% increase in penetration or speed changes produce nontrivial % gains in GDP. (ITU)
Czernich, Falck, Kretschmer & Woessmann — Broadband Infrastructure and Economic Growth (Economic Journal)	2011	OECD panel (1996–2007)	Instrumental-variable panel regressions	A 10 percentage-point increase in broadband penetration raised annual per-capita growth by ~0.9–1.5 percentage points (IV estimates). (OUP Academic)
Briglauer et al. — Socioeconomic benefits of high-speed broadband (peer-reviewed / 2024)	2024	Cross-country / country-level analyses	Econometric analysis of adoption & speed vs GDP outcomes	Reports positive short-run and pandemic-era effects of increased adoption/speeds on GDP; quantifies significant returns to adoption increases. (ScienceDirect)
Brattle Group — Economic Benefits of Fiber Deployment	2024	United States (nationwide modeling)	Benefit-cost modeling (NPV of housing value, income, employment, social benefits)	Finds large net present value benefits from fiber deployment (authors estimate substantial NPV and argue public support may be justified because private returns under-capture social benefits). (Brattle)
Brattle Group — Paying for Itself: ACP delivers more than it costs (Affordable Connectivity Program analysis)	2025	United States (program level)	Program cost-benefit modeling (health, education, labor market savings)	Concludes reinstating ACP yields net economic benefits greater than program cost via health, education, and labor productivity gains. (Brattle)
ITU / CITI (Columbia) — The Impact of Broadband on the Economy (Raul Katz)	2012	Global literature review + case analyses	Literature review + case studies; synthesis of empirical evidence	Summarizes broad evidence that broadband has positive effects on growth, productivity, and jobs and outlines policy issues for maximizing social returns. (ITU)
Broadband Commission / OECD syntheses	2013–2020	International	Literature syntheses / cross-country summaries	Survey of literature: typical estimates show a 10% rise in penetration can raise GDP growth by 0.24%–1.5% depending on context; policy reports argue public intervention can be warranted to capture social returns. (Broadband Commission)

Will AI Substitution Really Cut Jobs? Maybe Not

Artificial intelligence cost savings and job cuts are a tricky matter. Consider claims that AI allowed IBM to cut between 3500 and 8000 jobs by using AI. In fact, job losses might have been in the “hundreds,” not thousands.  

It is argued that IBM then redeployed its resources from AI automation to hire new talent for high-growth areas like software engineering, sales, and marketing. Headcount might have actually grown, as a result. 

The other issue is that some job “losses” were related to the prior company spinoff of Kendryl, the managed services business with 90,000 employees, about 25 percent of IBM’s total workforce. 

That led to later job cuts at both Kendryl and IBM. 

The point is that it is hard to quantify how AI affects jobs. There could be a combination of cuts as well as new jobs created. There will be some displacement of existing human jobs as automation creates substitutes.

Also, there are other effects. New technology gets used because it boosts productivity. If that lowers the cost of producing goods and services, lower prices happen, increasing consumer demand. 

The increased demand often requires more production, which can offset initial job losses by increasing demand for labor in other parts of the value chain or other growing sectors of the economy.

For example, the introduction of the Model T made cars cheaper, increasing demand for cars, which ultimately required more, not fewer, auto workers.

In addition, technology doesn't just automate old tasks. It creates entirely new ones that only humans can perform, particularly those involving non-routine cognitive, creative, and social skills (managing the new technology, developing software, data analysis, ethical oversight, or personalized services). 

That can lead to the formation of whole new industries and occupations that did not exist before.

Historically, this pattern has played out repeatedly, from the Industrial Revolution (replacement of artisans with factory workers and creating engineering/maintenance roles) to the information technology revolution (eliminating typists and switchboard operators while creating software developers, network administrators, and digital content creators). 

Artificial Intelligence automation should have the same net positive impact on jobs. That said, the immediate focus often will be on "lost jobs."

Over the past six years, IBM has continuously refined their internal virtual agent, AskHR, to automate more than 80 HR tasks and handle over 2.1 million employee conversations annually. Most recently, in 2025, the team integrated IBM watsonx Orchestrate.  

The AI agent helped contribute to a 40-percent reduction in the HR team’s operational costs over the past four years, IBM says. AskHR also achieved a 94 percent containment rate of common questions, has led to a 75 percent reduction in support tickets raised since 2016, and has created more than 10 million employee interactions each year since 2023.

That might suggest that AI will only work in one direction: substituting itself for existing jobs. That will happen, but is only part of the story.

The point is that AI might often not be a zero-sum game, where AI only destroys jobs. It should also create the need for new jobs.

Fourth Industrial Revolution Now is More About AI than IoT

Until artificial intelligence seemingly swept nearly everything before it, the “Fourth Industrial Revolution” or “Industry 4.0” was mostly about applying sensors and the internet of things to machinery used in factories. Recently, we are more likely to hear it said about applied AI. 

When the term “Fourth Industrial Revolution”  was popularized (notably by Klaus Schwab of the World Economic Forum starting around 2016), its initial core was the fusion of physical and digital realms. 

The Internet of Things was viewed as the key enabler, referring to the network of physical devices, sensors, machines and software that allows machines and computers to collect and exchange data. In a factory setting, this meant equipping machines with sensors to gather real-time data on their performance, environment, and output.

Of course, the issue, as always, is not “data” as such but the ability to wring useful insights from data. For that reason, we are starting to hear AI mentioned as driving the Fourth Industrial Revolution. 

Revolution	Key Characteristics	Approximate Time Period
First Industrial Revolution	Mechanization of production using water and steam power. The transition from agrarian and handicraft economies to industry and machine manufacturing.	Late 18th to mid-19th Century (c. 1760s - 1840s)
Second Industrial Revolution	Mass production driven by the widespread use of electricity and the advent of the assembly line (e.g., in steel, oil, and automobile industries).	Late 19th to early 20th Century (c. 1870s - World War I)
Third Industrial Revolution	The Digital Revolution, involving the use of electronics, IT, and automated production. The rise of computers, the internet, and early automation.	Mid-to-late 20th Century (c. 1950s - 1970s onwards)
Fourth Industrial Revolution?	Fusion of the physical, digital, and biological spheres, use of artificial intelligence. `	21st Century (2024 and forward)