GCRI · Labs
媒体未来实验室
Information ecosystems decide what people know about risk: which warnings reach them, which claims they believe and which institutions they trust. This lab studies news, platforms, recommender 系统, synthetic media and verification practice, and tests media innovations before they scale. This page sets out how misinformation and information gaps develop and spread, how interventions are evaluated, and who decides on the evidence.
工作如何开展。
Testbed operations
Future of media lab · Field lab coast · Campaign 3
Sensor deployment
示意海岸线
Sensors
| S-01 tide | 95 | 0.3 | Nominal |
| S-04 salinity | 86 | 0.8 | Recalibrate |
| S-07 wave | 90 | 0.1 | Nominal |
Sensor vs reference
图Testbed operations: example names and illustrative values, not a record or a live system.
Future of media lab: how the risk develops.
Information ecosystems and responsible media innovation. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Amplification by design
Ranking 系统 optimise for engagement, and content that provokes strong reactions can travel further. Small changes in ranking rules alter what large audiences see, which makes platform design a research object in its own right.
Synthetic media lowers the cost of deception
Generated images, audio and video can impersonate people and events cheaply. Provenance standards and detection help, but neither is complete, and the existence of fakes also lets genuine evidence be dismissed.
News deserts and trust gaps
Where local journalism closes, fewer sources report on councils, courts and hazards. Gaps are filled by rumour or partisan content, and trust in institutions shapes how warnings are received when they arrive.
Interventions can backfire
Labels, corrections, prebunking and friction aim to reduce false beliefs. Effects measured in surveys may fade, may not change behaviour, or may raise scepticism of accurate content too, which is why field evaluation matters.
存量、流量、延迟和一个临界阈值。
两个存量，即压力和 适应能力，由一个增强的增长回路、一个延迟的平衡响应以及压力超过能力时产生的损害相连；越过阈值后，系统可能发生突变。这是使用明示示例数值的教学模型，不是预测。
控制措施
- 压力增长
- 每年 6.0%
- 能力在降低压力方面的效力
- 每年 4.0%
- 对感知压力的投资响应
- 20%
- 感知延迟
- 4.0 年
- 压力的临界阈值
- 60 项指数
- 越过阈值后的放大
- 2.0 倍
- 模拟年数
- 50 年
读数
- 峰值压力
- 113 指数
- 期末能力
- 186 指数
- 超过阈值的年数
- 15.8 年
数字、方法与来源
| 年份 | 压力 | 能力 | 感知 |
|---|---|---|---|
| 0 | 20.0 | 50.0 | 20.0 |
| 4 | 23.3 | 44.9 | 21.2 |
| 8 | 27.2 | 42.2 | 24.0 |
| 12 | 31.7 | 42.3 | 27.7 |
| 16 | 36.8 | 45.5 | 32.2 |
| 20 | 42.3 | 51.5 | 37.2 |
| 24 | 47.9 | 60.3 | 42.6 |
| 28 | 53.2 | 71.3 | 48.0 |
| 32 | 57.8 | 84.3 | 53.1 |
| 36 | 66.4 | 98.4 | 58.4 |
| 40 | 82.7 | 115.9 | 69.7 |
| 44 | 97.8 | 139.7 | 83.8 |
| 48 | 109.1 | 169.4 | 97.3 |
方法
- 系统动力学 Forrester (1961), Industrial Dynamics, MIT Press; Sterman (2000), Business Dynamics
- 临界转变与临界点 Scheffer and others (2009), Early-warning signals for critical transitions, Nature 461
- 因果回路图 Sterman (2000), Business Dynamics, McGraw-Hill
- 不确定性扇形图（参数蒙特卡罗） Spiegelhalter, Pearson and Short (2011), Visualising uncertainty about the future, Science 333
- 单因素敏感性（龙卷风图） Saltelli and others (2008), Global Sensitivity Analysis: The Primer, Wiley
数据
No observed data enters this model: it runs on the teaching values set by the controls, to be replaced with a programme’s own evidence.
贯穿六大支柱，由 Nexus Consortium 牵头
How future of media lab is modelled, measured and tested.
模型、其假设、数据和不确定性，以科学家或工程师会核查的形式呈现。结构遵循已发表的方法。所取数值仅为示意，不是研究发现。
模型、方程与假设
The model structure used for future of media lab, the assumptions it makes and how each is tested.
Ri = f(Hi, Ei, Vi, Ci)危害、风险暴露、脆弱性和应对能力EAL = ∑k pk · Lk对年概率为 pkxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error| 假设 | 为何制作 | 如何检验 |
|---|---|---|
| 在一次事件中，危害与暴露相互独立 | 使损失积分保持可计算 | 与基于联合记录拟合的 copula 进行比较 |
| 耦合在工作点附近是线性的 | 让冲击通过 A 传播 | 在线性范围之外施加扰动并比较 |
| 脆弱性曲线可以在不同地点之间迁移使用 | 很少有本地曲线可用 | 在有观测损失的地方据此进行验证 |
- 输入
- 带溯源信息的版本化数据集
- 成果
- 分布，绝不是单一数字
- 代码
- 连同模型卡和测试一起发布
- 复现
- 相同的数据、代码、环境和随机种子
衡量什么，以及衡量得如何。
每个指标都注明其单位、频率、背后的来源类型以及数值将附带的证据标签。此处不显示数值：数值由计划依据其自身权限提供。
| 指标 | 单位 | 节奏 | 来源类型 | 标签 |
|---|---|---|---|---|
| Exposure to flagged content | proportion of impressions | 每项研究 | Platform researcher data access | 观测 |
| Belief accuracy | proportion of true and false claims correctly rated | 每项研究 | Survey experiments | 观测 |
| Truth discernment | difference between belief in true and false items | 每项研究 | Survey experiments | 估算 |
| Trust in news | 受访者百分比 | 年度 | Audience surveys | 已报告 |
| Provenance credential coverage | percent of published media carrying provenance data | 年度 | Publisher and platform audits | 观测 |
| Local news coverage | news outlets per local area | 年度 | News outlet census | 估算 |
| Researcher data access granted | count of approved access requests | 年度 | 平台透明度报告 | 已声明 |
证据来自何处。
具名公共数据集与机构记录，各自注明许可类别。依协议获得的来源仅按其条款使用，绝不再次发布。
| 来源 | 它提供什么 | 许可类别 |
|---|---|---|
| Reuters Institute Digital News Report (University of Oxford) | Cross-national survey of news use, trust and platform reliance | Report free access; data by request |
| Media Pluralism Monitor (European University Institute) | Risks to media pluralism by country and area | Open access, attribution |
| DSA Transparency Database (European Commission) | Statements of reasons for content moderation decisions by platforms | Free and open, EU reuse terms |
| Code of Conduct on Disinformation transparency centre | Signatory reports on measures against disinformation | Free access, terms of use |
| GDELT Project | Global news event and tone records from monitored media | 免费开放，须署名 |
| ClaimReview structured data (schema.org) | Structured fact checks published by verification organisations | 使用条款 |
由谁决定，以及证据服务于什么。
实验室 prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| 谁来决定 | 问题 | 本页面服务的证据 |
|---|---|---|
| Media and communications regulators | Whether platform risk assessments and mitigation measures meet legal duties | Independent audits, transparency database records, researcher access evidence |
| Newsroom editors and publishers | Whether to adopt a verification, labelling or provenance practice | Field evaluations of the practice with audience and trust outcomes |
| 平台信任与安全团队 | Which interventions to deploy at scale and how to monitor side effects | Randomised evaluations with effects on accurate as well as false content |
| Public health and emergency communicators | How to issue warnings that reach, and are believed by, different audiences | Reach, trust and comprehension evidence by audience group |
该主题产生的产出
- Preregistered field and survey experiments on corrections, labels and prebunking
- Audit designs for ranking and recommender 系统 under researcher access rules
- Provenance and verification test protocols for synthetic media
- Maps of local information gaps with sources stated
What experts ask about information ecosystems.
- How do accurate and inaccurate claims spread, and what slows each?
- Which sources do different communities trust during an emergency, and why?
- How do recommendation 系统 and platform design shape what people see?
- Which corrections, labels or warnings change belief or behaviour, and for how long?
- How can synthetic media be detected, and what are the limits of detection?
How information systems behave and fail.
Information spreads through networks of people and platforms. Network science shows that a few highly connected nodes can make diffusion fast and that clustering can create communities where some claims circulate and others do not. Platform design adds feedback: content that draws engagement is shown more, and engagement does not track accuracy.
Trust is a central variable. During disasters and health emergencies, people combine authority warnings with what they hear from family, local leaders and online sources. Where trust in institutions is low, accurate warnings may be ignored and rumours fill the gap. Misinformation, shared in error, and disinformation, shared deliberately, have different drivers and call for different responses.
Generative AI lowers the cost of producing convincing text, images, audio and video. This affects both the volume of misleading content and the credibility of genuine evidence, which can be dismissed as fabricated.
Research methods and their limits.
- 01
Content and network analysis
Mapping how material moves across accounts and platforms. Platform data access is restricted and changes, which limits replication.
- 02
Survey and panel research
Measuring trust, exposure and belief over time. Self-reports are imperfect and samples may miss key groups.
- 03
Randomised experiments
Testing labels, corrections and warning formats under ethics review. Laboratory effects may differ from real-world use.
- 04
Detection evaluation
Testing tools that identify synthetic media, using machine learning methods. Detection is an arms race and accuracy declines as generators improve.
Research on information is itself sensitive. The lab is designed to protect participant privacy, avoid amplifying harmful content and respect freedom of expression.
Where this page’s references lead.
阅读本页时所对照的机构和文件。状态和版本会变化；每个链接都指向发布机构。
问题与解答。
Why not simply measure how much misinformation exists?
Do corrections work?
How do researchers get platform data?
Does the lab rate news sources or decide what is true?
How does the lab handle harmful content in research?
Can a media organisation propose a study?
从本页贡献证据。
报告 is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.
- 电子邮件网关 GW-GCRI从本页发送证据、更正、案例、方法、审查、翻译或概况投稿。它会带上本页标签送达，由负责的编辑接收。
- Zenodo开放存缴存缴并提交到社区用您的 ORCID iD 在 Zenodo 上发布您自己的记录，然后提交到那里的 Reports 社区。由策展人审核；DOI 仍归您所有。
- 报告GCRI查看所有贡献方式贡献类型、各自的评审方式、您获得的署名、许可协议，以及绝不接受的内容。
切勿发送个人数据、敏感基础设施细节或您无权分享的任何内容。贡献在使用前会经过审查；发送贡献不赋予署名权或编辑权，赞助也绝不能换取研究结论。
法律文件。
本网站上的地图、数据和图表仅作为背景和一般信息展示。它们不是建议、预测或预警；信息的局限见免责声明和数据说明。