Within Conflict AI
How VIEWS Forecasts Conflict and Manages False Alarms
This page examines how the VIEWS conflict forecasting system handles rare events and produces false positives.
On this page
- Overview of VIEWS architecture
- Examples of accurate and false forecasts
- Implications for policy and trust
Page outline Jump by section
Introduction
The Violence & Impacts Early-Warning System (VIEWS) is one of the most influential attempts to use machine learning to forecast armed conflict before violence escalates. It is often presented as a promising example of how advanced data analysis might help governments, humanitarian organisations and researchers act earlier, potentially reducing human suffering and improving civilisational resilience. But VIEWS faces a problem common to many AI forecasting systems: conflict is rare, unpredictable and heavily shaped by human choices. A model can identify many genuine risks while also generating large numbers of warnings that never become wars.
This creates a difficult balance. If a forecasting system only warns when violence is almost certain, it may miss emerging crises. If it warns too often, decision-makers may stop trusting it. Understanding VIEWS therefore requires looking not only at where it succeeds, but also at how it handles false alarms, uncertainty and the statistical challenge of predicting rare events.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…
How VIEWS Tries to Forecast Rare Events
VIEWS was developed by researchers associated with the Peace Research Institute Oslo (PRIO), Uppsala University and collaborating institutions. Rather than predicting a specific attack or coup, it estimates the probability that organised violence will occur in a country or geographic area over future months. The system combines multiple models into an ensemble and draws on historical conflict patterns, political indicators, economic conditions, geographic variables and other predictors. Forecasts are updated regularly and made publicly available. Sage Journals[Peace Research Institute Oslo]prio.orgPeace Research Institute OsloVIEWS: Violence & Impacts Early-Warning SystemThe Violence Early-Warning System (ViEWS) is a publicly availa…
The rarity of major conflict creates a fundamental forecasting challenge. In a typical dataset, peaceful months vastly outnumber violent ones. A model that simply predicts peace everywhere could achieve high overall accuracy while being useless for early warning. Because of this imbalance, VIEWS researchers place less emphasis on simple accuracy measures and more emphasis on metrics that evaluate whether the system successfully identifies genuinely dangerous cases.[Diva Portal]diva-portal.orgUnlike accuracy, they give a better evaluation of how the algorithm is performing when dealing with…Read more…
This means that false alarms are not treated as a side issue. They are built into the evaluation problem itself. Any system trying to catch rare outbreaks of violence must decide how many false positives it is willing to tolerate in exchange for detecting more genuine conflicts.
Why False Alarms Are Hard to Avoid
Conflict forecasting resembles medical screening more than weather prediction. Imagine a disease that affects only a small fraction of the population. Even a highly capable screening system may generate many positive tests that ultimately turn out to be harmless simply because the condition itself is uncommon.
Political violence creates a similar statistical problem. Most regions that exhibit warning signs do not descend into major armed conflict. Economic stress, political repression, ethnic tensions or environmental shocks may increase risk without producing violence. As a result, a forecasting system can correctly identify a dangerous environment while still appearing to have issued a false alarm if conflict never materialises.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…
There are several reasons why a VIEWS warning may not be followed by observed violence:[youtube.com]youtube.comThe Violence & Impacts Early-Warning System (VIEWSExploring the Societal Potential of the VIEWS Early Warning System – the PoC Project in Review…
- The model overestimated the risk.
- Local conditions changed after the forecast.
- Governments, mediators or aid organisations intervened successfully.
- Violence occurred at a different time than predicted.
- Violence occurred below the threshold used in the dataset.
- Reporting systems failed to capture relevant events.
These possibilities make false alarms difficult to interpret. A warning that appears wrong may sometimes reflect a near miss rather than a meaningless prediction.
What Performance Evaluations Actually Show
Published evaluations generally find that VIEWS performs substantially better than random guessing and better than many traditional baseline approaches. Researchers report that the system successfully captures the persistence of ongoing conflicts and can identify some geographic diffusion patterns, including violence spreading into neighbouring regions. Evaluations have highlighted cases such as northern Mozambique and parts of Cameroon where elevated risk estimates appeared before major escalations became widely recognised.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…
At the same time, the research literature is unusually candid about limitations. The original VIEWS paper and later revisions acknowledge that earlier versions struggled to distinguish clearly between low-risk and high-risk observations. Many predictions clustered in the middle range rather than producing strong separation between likely and unlikely conflict cases. Researchers subsequently revised the system to improve classification and ensemble weighting methods.[OUP Academic]academic.oup.comOUP AcademicRevising and evaluating the ViEWS political Violence Early…by H Hegre · 2021 · Cited by 54 — This article presents an upda…
One revealing finding from later evaluations was that the original system generated relatively few very high-confidence forecasts and relatively few very low-confidence forecasts. Instead, conflict and non-conflict cases often overlapped within similar probability ranges. In practical terms, this meant policymakers could not always treat forecast scores as sharply distinct signals.[OUP Academic]academic.oup.comOUP AcademicRevising and evaluating the ViEWS political Violence Early…by H Hegre · 2021 · Cited by 54 — This article presents an upda…
The revisions introduced after the first deployment were specifically designed to improve the system’s ability to separate high-risk and low-risk cases and to better forecast conflict onsets in places without extensive recent violence.[OUP Academic]academic.oup.comOUP AcademicRevising and evaluating the ViEWS political Violence Early…by H Hegre · 2021 · Cited by 54 — This article presents an upda…
Where VIEWS Produces the Most False Positives
False alarms tend to appear in several recurring situations.
Regions With Persistent Structural Risk
Some locations possess many characteristics historically associated with conflict: weak state institutions, armed groups, political instability or economic fragility. VIEWS may repeatedly assign elevated risk to these regions even during periods when violence does not occur.
From a statistical perspective this is often rational. The underlying conditions genuinely resemble past conflict environments. Yet repeated warnings can create an impression that the system is crying wolf.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…
Forecasting New Wars Rather Than Continuing Ones
Conflict continuation is generally easier to predict than conflict onset.
If a civil war is already active, historical patterns often provide strong clues about future violence. Predicting entirely new conflicts is much harder because the triggering events may depend on leadership decisions, protests, military defections or other developments that are difficult to observe in advance. Evaluations of VIEWS and related forecasting systems consistently show stronger performance for continuing conflicts than for genuinely novel outbreaks.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…[OUP Academic]academic.oup.comOUP AcademicRevising and evaluating the ViEWS political Violence Early…by H Hegre · 2021 · Cited by 54 — This article presents an upda…
Geographic Spillover Forecasts
VIEWS is designed partly to detect diffusion effects, where violence spreads across borders or into neighbouring regions. This can improve sensitivity to emerging crises, but it also increases the risk of forecasting violence in places that remain stable despite being geographically exposed.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…
The Trade-Off Between Missing Wars and Warning Too Often
A central insight from conflict forecasting research is that reducing false alarms usually increases missed conflicts, and vice versa.
This trade-off is often discussed using precision and recall. Precision measures how many warnings prove correct. Recall measures how many actual conflicts are successfully detected. Increasing one commonly reduces the other. Researchers working on conflict forecasting frequently use precision-recall analysis precisely because conflict events are so rare.[Barcelona School of Economics]bw.bse.eu1355 fileWe therefore focus on presenting precision/recall curves for armed conflict onset.Read more…
For policymakers, the preferred balance depends on the consequences of error.
- A humanitarian agency may prefer more warnings if missing a crisis could cost thousands of lives.
- A government allocating limited resources may prefer fewer false alarms.
- Diplomatic actors may worry that repeated inaccurate warnings could damage credibility.
This means there is no single correct threshold for action. VIEWS provides probabilities rather than definitive predictions partly because different users face different costs from false positives and false negatives.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…
A Less Obvious Problem: Good Scores Can Reward Conservative Forecasts
One of the more surprising findings in the conflict forecasting literature is that evaluation systems themselves can distort behaviour.
Researchers analysing a VIEWS forecasting challenge found that some scoring approaches rewarded conservative predictions that stayed close to “no major change”. Under certain evaluation rules, a simple model predicting little movement could outperform more ambitious forecasts attempting to anticipate major escalations.[arXiv]arxiv.orgarXiv Direction Augmentation in the Evaluation of Armed Conflict PredictionsDirection Augmentation in the Evaluation of Armed Conflict PredictionsApril 24, 2023…
This matters because forecasting systems are partly shaped by the metrics used to judge them. If evaluation methods penalise bold predictions too heavily, models may become overly cautious. If they reward sensitivity too strongly, systems may generate excessive false alarms.
The challenge therefore extends beyond machine learning architecture. It also involves deciding what kinds of forecasting mistakes society considers most costly.
Data Problems Can Create Apparent Forecast Errors
Not every false alarm originates in the model itself.
Conflict forecasting depends heavily on event databases such as those maintained by the Uppsala Conflict Data Program and ACLED. These datasets are among the best available, but they face unavoidable reporting delays and information gaps. Recent research has shown that conflict events are often reported weeks after they occur and that reporting speed varies systematically across countries and event types.[arXiv]arxiv.orgarXiv Direction Augmentation in the Evaluation of Armed Conflict PredictionsDirection Augmentation in the Evaluation of Armed Conflict PredictionsApril 24, 2023…
This creates a subtle problem. A forecast may appear incorrect because the relevant violence has not yet entered the dataset. Alternatively, violence in remote or politically closed regions may be undercounted altogether.
For early-warning systems, data quality therefore becomes part of forecasting performance. Some apparent prediction failures reflect weaknesses in observation rather than weaknesses in inference.
Why Trust Depends on Transparency
One reason VIEWS receives attention within the conflict forecasting field is its emphasis on public evaluation. Forecasts are published, archived and regularly reassessed rather than being hidden inside government systems. Researchers have repeatedly revisited earlier forecasts and compared them with subsequent outcomes.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…
This transparency helps users understand false alarms instead of treating forecasts as mysterious outputs from a black box.
Several features support trust:
- Publicly available forecasts.[prio.org]prio.orgPeace Research Institute OsloVIEWS: Violence & Impacts Early-Warning SystemThe Violence Early-Warning System (ViEWS) is a publicly availa…
- Published methodological descriptions.
- Regular performance evaluations.
- Open discussion of failures and revisions.
- Comparison with benchmark models and forecasting competitions.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…[OUP Academic]academic.oup.comOUP AcademicRevising and evaluating the ViEWS political Violence Early…by H Hegre · 2021 · Cited by 54 — This article presents an upda…
For conflict prevention, transparency may matter almost as much as raw predictive performance. Policymakers are more likely to use uncertain forecasts responsibly when they can inspect assumptions and understand limitations.
What VIEWS Suggests About AI and Human Flourishing
Within the broader debate about AI’s role in humanity’s long-term future, VIEWS illustrates both the promise and the limits of predictive systems.
The optimistic case is straightforward. If AI systems can identify escalating violence earlier than human analysts alone, they may help preserve lives, institutions and social stability. Better anticipation of conflict could support humanitarian planning, reduce displacement and strengthen the conditions needed for scientific progress, economic development and international cooperation.[Observatory of Public Sector Innovation]oecd-opsi.orgObservatory of Public Sector Innovation Vi EWSObservatory of Public Sector InnovationViEWS - The Political Violence Early-Warning System23 Jan 2023 — The Violence Early-Warning System…
The cautionary lesson is equally important. Forecasts are not neutral facts. False alarms can affect diplomatic decisions, resource allocation and public perceptions of entire regions. A system that predicts conflict everywhere becomes unusable; a system that predicts conflict nowhere becomes irrelevant.
VIEWS therefore offers a realistic picture of what high-stakes AI may look like in practice. The value does not come from perfect prediction. It comes from producing probabilistic warnings that are informative enough to improve human judgement while remaining transparent about uncertainty. The challenge is not eliminating false alarms entirely, but managing them in ways that preserve trust without ignoring genuine risks.[Sage Journals]journals.sagepub.comSage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political…[OUP Academic]academic.oup.comOUP AcademicRevising and evaluating the ViEWS political Violence Early…by H Hegre · 2021 · Cited by 54 — This article presents an upda…
eBay marketplace picks
Marketplace Samples
Example marketplace items related to this page. Use the search link to explore similar finds on eBay.
Endnotes
1.
Source: academic.oup.com
Link:https://academic.oup.com/jpr/article/58/3/599/8365273
Source snippet
OUP AcademicRevising and evaluating the ViEWS political Violence Early...by H Hegre · 2021 · Cited by 54 — This article presents an upda...
2.
Source: prio.org
Link:https://www.prio.org/projects/1977
Source snippet
Peace Research Institute OsloVIEWS: Violence & Impacts Early-Warning SystemThe Violence Early-Warning System (ViEWS) is a publicly availa...
3.
Source: diva-portal.org
Link:https://www.diva-portal.org/smash/get/diva2%3A1222824/FULLTEXT01.pdf
Source snippet
Unlike accuracy, they give a better evaluation of how the algorithm is performing when dealing with...Read more...
4.
Source: viewsforecasting.org
Title: ViEWS Overview 25June2018
Link:https://viewsforecasting.org/wp-content/uploads/ViEWS-Overview-25June2018.pdf
Source snippet
ViEWS: A political Violence Early Warning System25 Jun 2018 — The forecasts indicate a very strong persistence of conflict in regions in...
5.
Source: academic.oup.com
Title: Academic Can We Predict Armed Conflict?
Link:https://academic.oup.com/isq/article/65/3/660/6124679
Source snippet
How the First 9 Years of...by H Hegre · 2021 · Cited by 61 — This article sheds light on this question by evaluating 9 years of, at the...
6.
Source: arxiv.org
Title: arXiv Direction Augmentation in the Evaluation of Armed Conflict Predictions
Link:https://arxiv.org/abs/2304.12108
Source snippet
Direction Augmentation in the Evaluation of Armed Conflict PredictionsApril 24, 2023...
Published: April 24, 2023
7.
Source: arxiv.org
Title: arXiv Assessing Reporting Delays in ACLED Conflict Event Data
Link:https://arxiv.org/abs/2603.25964
Source snippet
Assessing Reporting Delays in ACLED Conflict Event DataMarch 26, 2026...
Published: March 26, 2026
8.
Source: prio.org
Link:https://www.prio.org/news/3602
Source snippet
Peace Research Institute OsloNew Results from Prediction Challenge That Seeks to...26 Feb 2025 — PRIO launched a conflict prediction cha...
9.
Source: arxiv.org
Link:https://arxiv.org/html/2410.14507v1
Source snippet
Bin-Conditional Conformal Prediction of Fatalities...Oct 18, 2024 — We apply the bin-conditional conformal prediction algorithm to forec...
10.
Source: arxiv.org
Link:https://arxiv.org/pdf/2506.14817
Source snippet
Next-Generation Conflict Forecastingby SP von der Maase · 2025 — The goal of conflict forecasting is to generate reliable predictions con...
11.
Source: youtube.com
Title: The Violence & Impacts Early-Warning System (VIEWS)
Link:https://www.youtube.com/watch?v=dx2P7jvYTsk
Source snippet
Exploring the Societal Potential of the VIEWS Early Warning System – the PoC Project in Review...
12.
Source: journals.sagepub.com
Link:https://journals.sagepub.com/doi/10.1177/0022343319823860
Source snippet
Sage JournalsViEWS: A political violence early-warning systemby H Hegre · 2019 · Cited by 247 — This article presents ViEWS – a political...
13.
Source: oecd-opsi.org
Title: Observatory of Public Sector Innovation Vi EWS
Link:https://oecd-opsi.org/innovations/views-the-political-violence-early-warning-system/
Source snippet
Observatory of Public Sector InnovationViEWS - The Political Violence Early-Warning System23 Jan 2023 — The Violence Early-Warning System...
14.
Source: bw.bse.eu
Title: 1355 file
Link:https://bw.bse.eu/wp-content/uploads/2022/06/1355-file.pdf
Source snippet
We therefore focus on presenting precision/recall curves for armed conflict onset.Read more...
15.
Source: viewsforecasting.org
Link:https://viewsforecasting.org/
Source snippet
VIEWS | Open-source project leveraging AI to predict armed...The article highlights advances in conflict forecasting and examines how re...
Additional References
16.
Source: eprints.whiterose.ac.uk
Link:https://eprints.whiterose.ac.uk/id/eprint/226622/1/hegre-et-al-2025-the-2023-24-views-prediction-challenge-predicting-the-number-of-fatalities-in-armed-conflict-with.pdf
Source snippet
the number of fatalities in armed conflict, with...by H Hegre · 2025 · Cited by 18 — Levels of analysis: forecasts of (logged) armed con...
17.
Source: researchgate.net
Title: 370227473 Direction Augmentation in the Evaluation of Armed Conflict Predictions
Link:https://www.researchgate.net/publication/370227473_Direction_Augmentation_in_the_Evaluation_of_Armed_Conflict_Predictions
Source snippet
Direction Augmentation in the Evaluation of Armed Conflict...Apr 25, 2023 — Governmental and nongovernmental organizations have increasi...
18.
Source: euridice.eu
Title: Beyond Closed Doors An Open Source AI Framework for Forecasting Armed Conflict 1
Link:https://euridice.eu/wp-content/uploads/2025/09/Beyond-Closed-Doors-An-Open-Source-AI-Framework-for-Forecasting-Armed-Conflict-1.pdf
Source snippet
An Open-Source AI Framework for Forecasting Armed ConflictSep 8, 2025 — Surges in conflict-related news coverage were shown to act as ear...
19.
Source: gppac.net
Link:https://gppac.net/files/2018-12/Early%20Warning%20and%20Early%20Response.pdf
Source snippet
auses of conflict, predict the outbreak of violence and mitigate the conflict, an early warning system...Read more...
20.
Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S0010482525005177
Source snippet
Explainable deep stacking ensemble model for accurate...by R Haque · 2025 · Cited by 71 — For performance evaluation, we focused on metr...
21.
Source: viewsforecasting.org
Title: new data release conflict forecasts for may 2026 april 2029
Link:https://viewsforecasting.org/news/new-data-release-conflict-forecasts-for-may-2026-april-2029/
Source snippet
New Data Release: Conflict Forecasts for May 2026–April 20292 days ago — The latest VIEWS forecasts for state-based armed conflict are no...
Published: may 2026
22.
Source: demscore.se
Link:https://demscore.se/data/static-datasets/ucdp-views-static-datasets/views-prio-grid-month-conflict-predictions-last-input-data-august-2025/
Source snippet
VIEWS PRIO-GRID-Month Conflict Predictions (Last Input...VIEWS PRIO-GRID-Month Conflict Predictions (Last Input Data: August 2025) versi...
Published: August 2025
23.
Source: researchgate.net
Title: 331132518 ViEWS A political violence early warning system
Link:https://www.researchgate.net/publication/331132518_ViEWS_A_political_violence_early-warning_system
Source snippet
(PDF) ViEWS: A political violence early-warning system2 May 2026 — This article presents ViEWS-a political violence early-warning system...
Published: May 2026
24.
Source: homepages.ucl.ac.uk
Title: ucl.ac.ukwe predict conflict better then we thought!
Link:https://www.homepages.ucl.ac.uk/~uctqnm0/web2_files/isaPaper.pdf
Source snippet
taking timeby G CIFLIKLI · Cited by 2 — The prediction of conflict and peace dynamics has become a centerpiece of academic output in the...
25.
Source: royalsocietypublishing.org
Title: Data driven conflict classification exposes weak
Link:https://royalsocietypublishing.org/rsos/article/12/12/250897/366130/Data-driven-conflict-classification-exposes-weak
Source snippet
Data-driven conflict classification exposes weak predictive...17 Dec 2025 — Even when considering a wider pool of algorithms—such as fro...
Topic Tree