Can Pandemic AI Work With Incomplete Data?

Introduction

Pandemic forecasting AI is often described as a way to see outbreaks coming before hospitals fill, borders close or supply chains break. In the most optimistic versions of the AI abundance story, increasingly powerful systems could help humanity detect emerging diseases early, coordinate responses faster and reduce one of the major threats to long-term human flourishing. But there is a basic constraint that advanced models cannot escape: they can only forecast what they can observe.

The hardest period of any outbreak is usually the beginning, when information is sparse, delayed, politically contested or simply missing. The same conditions that make early warning valuable also make reliable prediction difficult. During COVID-19 and many previous epidemics, reporting delays, unequal testing access, undercounted populations and fragmented surveillance systems distorted the data flowing into forecasting models. As a result, AI systems often faced a version of the classic problem of “garbage in, garbage out”: sophisticated algorithms operating on incomplete pictures of reality. PMC[OUP Academic]academic.oup.comOUP AcademicA widely neglected impact factor in COVID-19 forecastsby L Ma · 2024 · Cited by 10 — In our work, through the analysis of the…

For pandemic forecasting to become a serious pillar of civilisational resilience, improvements in AI capability may matter less than many people assume. Equally important is building the global disease-data infrastructure that allows those systems to see what is actually happening.

Can Pandemic AI Work With Incomplete Data?

In principle, machine learning systems are well suited to finding patterns across large, messy datasets. Modern outbreak forecasting can combine clinical reports, mobility data, environmental information, news reports, internet searches and other signals to estimate where disease may be spreading.[PMC]pmc.ncbi.nlm.nih.govReporting delays: A widely neglected impact factor in COVID…by L Ma · 2024 · Cited by 10 — We develop a statistical framework to de…[ScienceDirect]sciencedirect.comThey also enhance risk assessment by…Read more…

The problem is that the first weeks of a new outbreak rarely generate large, reliable datasets.

Several distortions appear simultaneously:

From a machine-learning perspective, these are not minor inconveniences. They alter the underlying distribution of the data itself. A model trained on historical outbreaks assumes that incoming information reflects reality with reasonable consistency. During an emerging epidemic, that assumption can fail dramatically.[PNAS]pnas.orgHere, we detail three regional-scale models for forecasting and assessing the course of…

This creates a paradox. Forecasting systems are usually most accurate when diseases are already well characterised and data collection is stable. Yet the moments when society most needs prediction are the moments when information quality is worst.

Why Early Outbreak Data Is Unreliable

Reporting delays can make outbreaks look smaller than they are

One of the clearest examples comes from COVID-19 reporting delays.

Researchers analysing the first pandemic wave found substantial delays between infections occurring and those infections appearing in official datasets. These delays propagated through forecasting systems, producing distorted estimates of disease growth and epidemic timing. After correcting for reporting delays, forecast errors in some cases fell by as much as 50 percent.[OUP Academic]academic.oup.comOUP AcademicA widely neglected impact factor in COVID-19 forecastsby L Ma · 2024 · Cited by 10 — In our work, through the analysis of the…

For AI systems, delayed data creates a dangerous illusion. The model may appear to be observing real-time conditions when it is actually looking into the past.

Imagine an outbreak doubling every few days. If official case reports lag by a week, a forecasting system may underestimate current spread precisely when intervention decisions are being made. Public-health leaders may conclude that an outbreak remains manageable when transmission has already accelerated beyond that point.

Testing changes can create false trends

COVID-19 also demonstrated how changes in testing behaviour distort forecasts.

When testing expands rapidly, reported cases can rise even if transmission remains stable. When testing contracts, case counts may fall despite worsening spread. Forecasting systems that rely heavily on confirmed case numbers can mistake changes in surveillance intensity for changes in disease dynamics.[arXiv]arxiv.orgDisease Outbreak Detection and Forecasting: A Review of…Statistical and machine learning techniques have been applied to the predictio…

This issue is particularly severe during the opening phase of a pandemic because testing infrastructure is often being built at the same time the disease is spreading.

The resulting data stream does not simply contain noise. It contains systematic bias.

Novel pathogens provide very little training data

Machine learning generally benefits from large historical datasets. Emerging pathogens provide the opposite.

During the early stages of COVID-19, researchers repeatedly noted that forecasting systems had only small numbers of observations available for training. Models therefore had to extrapolate from extremely limited evidence while key epidemiological characteristics remained uncertain.[arXiv]arxiv.orgDisease Outbreak Detection and Forecasting: A Review of…Statistical and machine learning techniques have been applied to the predictio…

This is one reason outbreak prediction often appears more successful in retrospective studies than in real-time deployment. Once a pandemic has generated months or years of observations, patterns become easier to identify. Early warning systems do not have that luxury.

Bias And Missing Populations In Surveillance

The people missing from the data are often the most vulnerable

Disease surveillance is not evenly distributed across society.

Communities with weaker healthcare access, lower testing rates, poorer digital connectivity or limited public-health infrastructure often contribute less data to surveillance systems. Forecasting models may therefore become most accurate for populations that are already relatively visible and least accurate for populations facing the greatest risks.[dpe.gospub.com]dpe.gospub.comArtificial Intelligence for Public Health Surveillanceby GC Ikechukwu · 2026 — Though, actual implementations reveal significant traps: b…

This creates a feedback problem.

If surveillance systems systematically undercount certain populations, forecasts generated from those systems can underestimate disease burden in those same groups. Resources may then be allocated away from areas that need them most, reinforcing the original bias.

For a future in which AI contributes to broad human flourishing rather than concentrating advantages among already well-observed populations, this problem is not peripheral. It is central.

Global blind spots weaken global forecasts

Pandemics do not respect national borders, but disease-data systems remain heavily fragmented.

High-income countries generally produce richer surveillance datasets than low-income countries. Yet outbreaks often emerge or expand in places where laboratory capacity, reporting infrastructure and healthcare coverage are more limited. A forecasting system attempting global prediction may therefore receive its weakest information from regions where early detection is most important.[PMC]pmc.ncbi.nlm.nih.govArtificial intelligence in early warning systems for infectious…by I Villanueva-Miranda · 2025 · Cited by 84 — Using historical dat…

An AI model cannot reliably infer an outbreak that is largely invisible in its input data.

This is one reason many researchers argue that pandemic preparedness is as much an infrastructure challenge as an algorithm challenge. Better models cannot fully compensate for absent observations.

Digital surveillance introduces new biases

Many modern forecasting systems supplement official reports with alternative signals such as news articles, internet searches and social-media activity. These sources can sometimes identify unusual disease activity before formal reporting systems respond. ScienceDirect[2publichealthaihandbook.com]publichealthaihandbook.comAI in Disease Surveillance and Outbreak DetectionAI transforms outbreak detection by analyzing diverse data streams in real-time, using a…

However, these data sources bring their own distortions.

Search activity may reflect public anxiety rather than actual infections. News coverage can spike because of media attention rather than epidemiological change. Social-media participation varies substantially by age, geography and income.

The result is not necessarily bad forecasting. In some cases these signals improve prediction. But they do not eliminate surveillance bias. They often shift it into different forms.

The Wastewater Surveillance Lesson

Wastewater monitoring became one of the most celebrated innovations of the COVID-19 era because it could detect viral activity without requiring individuals to seek testing or medical care. Researchers showed that sewage monitoring could reveal disease circulation before large numbers of clinical cases appeared.[OAE Publishing]oaepublish.comOAE PublishingIntegrated environmental surveillance: the role of…by M Oliveira · 2025 · Cited by 3 — Sewage water analysis allows the…

For AI forecasting systems, wastewater data offered a valuable independent signal.

Yet even this approach has limits.

Recent research suggests that wastewater surveillance can contain its own coverage gaps. Communities lacking sewage infrastructure or comprehensive monitoring may remain underrepresented. Vulnerable populations can therefore continue to disappear from supposedly population-wide datasets.[News-Medical]news-medical.netStudy reveals inequities in wastewater-based diseaseStudy reveals inequities in wastewater-based disease…May 28, 2026 — 4 days ago — Wastewater surveillance was hailed during…Published: May 28, 2026

The lesson is broader than wastewater monitoring itself.

There is no perfectly objective surveillance stream. Every dataset captures some aspects of reality while missing others. Effective forecasting increasingly depends on combining multiple imperfect signals rather than relying on a single source of truth.

Why Better AI Alone Cannot Solve The Problem

A common assumption in discussions of advanced AI is that more capable models will eventually overcome data limitations.

To some extent this is true. Modern systems can integrate heterogeneous data sources, identify anomalies, estimate missing values and detect patterns that traditional statistical methods might miss.[PMC]pmc.ncbi.nlm.nih.govInfectious Disease Surveillance in the Era of Big Data and AIby CO Idahor · 2025 · Cited by 13 — This review explores the potential of…[PMC]pmc.ncbi.nlm.nih.govin public health surveillance: An overview of novel…by H Rilkoff · 2024 · Cited by 26 — Wastewater surveillance (WWS) has evolved as a…

But there are hard limits.

If infections are not being tested, reported or observed through alternative surveillance systems, the information may simply not exist in usable form. AI can infer hidden patterns, but it cannot perfectly reconstruct realities that leave almost no measurable trace.

This distinction matters for long-term visions of AI-enabled civilisational resilience.

Forecasting systems may eventually become far more powerful than today’s tools. Yet even highly advanced intelligence requires sensory input. In pandemic forecasting, disease surveillance functions as the sensory layer. Weak surveillance constrains what prediction systems can know.

The bottleneck is therefore partly computational but also institutional, logistical and political.

Improving Global Disease Data Infrastructure

The strongest response to distorted outbreak forecasts is not merely building larger models. It is improving the quality, speed and coverage of disease observation itself.

Several approaches are increasingly important:

Faster reporting systems. Reducing delays between diagnosis and reporting can significantly improve forecast quality and situational awareness.[OUP Academic]academic.oup.comOUP AcademicA widely neglected impact factor in COVID-19 forecastsby L Ma · 2024 · Cited by 10 — In our work, through the analysis of the…

Multi-source surveillance. Combining clinical reports, laboratory testing, wastewater monitoring, environmental sensing and digital signals reduces dependence on any single flawed dataset. OAE Publishing[2publichealthaihandbook.com]publichealthaihandbook.comAI in Disease Surveillance and Outbreak DetectionAI transforms outbreak detection by analyzing diverse data streams in real-time, using a…

Coverage of underserved populations. Forecasts become more reliable when surveillance systems deliberately include regions and groups that are traditionally undercounted.[News-Medical]news-medical.netStudy reveals inequities in wastewater-based diseaseStudy reveals inequities in wastewater-based disease…May 28, 2026 — 4 days ago — Wastewater surveillance was hailed during…Published: May 28, 2026

International interoperability. Data standards that allow information sharing across borders can reduce blind spots during rapidly spreading outbreaks.[PMC]pmc.ncbi.nlm.nih.govArtificial intelligence in infection surveillance: Data integration…by JH Li · 2025 · Cited by 10 — AI-driven infection surveillanc…

Uncertainty-aware forecasting. Models should communicate confidence intervals and data limitations rather than presenting precise predictions that imply more certainty than the evidence supports. Researchers increasingly emphasise that outbreak forecasting must account explicitly for missing and delayed information.[PNAS]pnas.orgHere, we detail three regional-scale models for forecasting and assessing the course of…

These investments are less glamorous than visions of superintelligent disease prediction. Yet they may generate larger real-world gains.

What This Means For The AI Bloom Vision

One of the strongest arguments for advanced AI is that it could help humanity become better at anticipation rather than merely reaction. Pandemic forecasting is an important test case.

The experience of COVID-19 suggests that prediction systems can provide useful warning signals and improve public-health decision-making. But it also shows that intelligence alone is not enough. The quality of civilisation’s forecasts depends heavily on the quality of civilisation’s observations.[PMC]pmc.ncbi.nlm.nih.govReporting delays: A widely neglected impact factor in COVID…by L Ma · 2024 · Cited by 10 — We develop a statistical framework to de…[ScienceDirect]sciencedirect.comreporting, timely outbreak detection, and effective response..Read more…

For the broader vision of human flourishing and long-term resilience, this carries a wider lesson. Many future AI systems may be limited not primarily by reasoning power but by the quality of the information flowing into them. Better sensors, better reporting networks, better institutions and broader access to data may prove just as important as better algorithms.

If advanced AI is eventually to help protect humanity from biological threats at global scale, the path is unlikely to be a simple story of smarter models. It is more likely to involve a combination of stronger intelligence and stronger observation: a world where disease surveillance becomes faster, more representative and more trustworthy, giving forecasting systems a chance to see emerging dangers before they become disasters.

Amazon book picks

Further Reading

Books and field guides related to Can Pandemic AI Work With Incomplete Data?. Use these as the next step if you want deeper reading beyond the article.

Book

The Rules of Contagion

By Adam Kucharski

Explains how forecasting works when data are noisy or incomplete.

See on Amazon

Book

The Premonition

By Michael Lewis

Explores how weak data and institutional blind spots hinder early epidemic response.

See on Amazon

Book

Deadliest Enemy

By Michael T. Osterholm, Mark Olshaker

Highlights surveillance gaps and preparedness challenges.

See on Amazon

Book

Epidemics and Society

By Frank M. Snowden

Shows how surveillance and information quality shape epidemic understanding.

See on Amazon

Browse more on Amazon:The Rules of ContagionThe PremonitionDeadliest Enemy

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Example eBay listing

"John is Not Really Dull" Vintage Style 1936 WPA Public Health Poster - 16x24

SearcheBay.co.uk: public health poster

Browse similar oneBay.co.uk

Example eBay listing

Don't Feed Rats Public Health Service poster during WWII 13 x 19" Photo Print

SearcheBay.co.uk: public health poster

Browse similar oneBay.co.uk

Example eBay listing

Vintage 1950s public health awareness advert poster

SearcheBay.co.uk: public health poster

Browse similar oneBay.co.uk

Example eBay listing

VINTAGE Public Health Poster Poison Ivy Poisonous Plants VINTAGE Advertising

SearcheBay.co.uk: public health poster

Browse similar oneBay.co.uk

Browse more oneBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC11156234/

2. Source: academic.oup.com
Link:https://academic.oup.com/pnasnexus/article-pdf/3/6/pgae204/58334731/pgae204.pdf

3. Source: pnas.org
Link:https://www.pnas.org/doi/10.1073/pnas.2006520117

4. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC12230060/

5. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2319417025001039

6. Source: arxiv.org
Link:https://arxiv.org/html/2410.17290v1

7. Source: arxiv.org
Link:https://arxiv.org/abs/2010.05079

8. Source: arxiv.org
Title: arXiv COVID-19: The unreasonable effectiveness of simple models
Link:https://arxiv.org/abs/2005.11085

Published: May 22, 2020

9. Source: arxiv.org
Link:https://arxiv.org/abs/2003.10776

Published: March 24, 2020

10. Source: arxiv.org
Link:https://arxiv.org/abs/2004.04019

11. Source: dpe.gospub.com
Link:https://dpe.gospub.com/dpe/article/view/13

12. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC12587752/

13. Source: publichealthaihandbook.com
Link:https://publichealthaihandbook.com/applications/surveillance.html

14. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC11075801/

15. Source: news-medical.net
Title: Study reveals inequities in wastewater-based disease
Link:https://www.news-medical.net/news/20260527/Study-reveals-inequities-in-wastewater-based-disease-outbreak-detection-systems.aspx

Published: May 28, 2026

16. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC13091077/

17. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2667193X26000827

18. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S0010482522001342

19. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2949924026000066

20. Source: oaepublish.com
Link:https://www.oaepublish.com/articles/wecn.2024.80

Additional References

21. Source: elib.dlr.de
Link:https://elib.dlr.de/219883/1/21621819_1_Selection_of_Robust_Features_for_Gaussian_Process_Regression_to_Predict_COVID_19_Case_Rates.pdf

22. Source: researchgate.net
Link:https://www.researchgate.net/publication/396264159_Infectious_Disease_Surveillance_in_the_Era_of_Big_Data_and_AI_Opportunities_and_Pitfalls

23. Source: researchgate.net
Link:https://www.researchgate.net/publication/400355604_Artificial_Intelligence-Driven_Approaches_to_Disease_Surveillance_and_Outbreak_Prediction

Published: January 2026

24. Source: facebook.com
Link:https://www.facebook.com/ECDC.EU/posts/%F0%9D%97%AA%F0%9D%97%B5%F0%9D%97%AE%F0%9D%98%81-%F0%9D%97%B0%F0%9D%97%AE%F0%9D%97%BB-%F0%9D%98%84%F0%9D%97%AE%F0%9D%98%80%F0%9D%98%81%F0%9D%97%B2%F0%9D%98%84%F0%9D%97%AE%F0%9D%98%81%F0%9D%97%B2%F0%9D%97%BF-%F0%9D%97%BF%F0%9D%97%B2%F0%9D%98%83%F0%9D%97%B2%F0%9D%97%AE%F0%9D%97%B9-%F0%9D%97%AE%F0%9D%97%AF%F0%9D%97%BC%F0%9D%98%82%F0%9D%98%81-%F0%9D%97%B1%F0%9D%97%B6%F0%9D%98%80%F0%9D%97%B2%F0%9D%97%AE%F0%9D%98%80%F0%9D%97%B2%F0%9D%98%80-%F0%9D%97%AE%F0%9D%97%BB%F0%9D%97%B1-%F0%9D%97%B6%F0%9D%98%80-%F0%9D%97%B6%F0%9D%98%81-%F0%9D%97%BF%F0%9D%97%B2%F0%9D%97%AE%F0%9D%97%B9%F0%9D%97%B9%F0%9D%98%86-%F0%9D%98%81%F0%9D%97%B5%F0%9D%97%B2-%F0%9D%97%9B%F0%9D%97%BC%F0%9D%97%B9%F0%9D%98%86-%F0%9D%97%9A%F0%9D%97%BF%F0%9D%97%AE%F0%9D%97%B6%F0%9D%97%B9-%F0%9D%97%BC%F0%9D%97%B3-%F0%9D%97%B2%F0%9D%97%BD%F0%9D%97%B6/1376463817839442/

25. Source: youtube.com
Title: From the Unseen to Foresight: Epidemic Modeling in the Age of Computing and AI
Link:https://www.youtube.com/watch?v=gD0obKGIWVk

26. Source: medrxiv.org
Title: 2025.06.24.25330210v1.full text
Link:https://www.medrxiv.org/content/10.1101/2025.06.24.25330210v1.full-text

27. Source: youtube.com
Link:https://www.youtube.com/watch?v=Ohs7mnl0tzc

28. Source: durableproject.org
Title: A I can strengthen pandemic preparedness
Link:https://durableproject.org/news/ai-can-strengthen-pandemic-preparedness/

29. Source: nature.com
Link:https://www.nature.com/articles/s41746-025-01606-1

30. Source: youtube.com
Title: AI, qualitative data, and the case for statistical rigour
Link:https://www.youtube.com/watch?v=IhCTCfW-pdg