Coordination

Introduction

AI could help civilisation coordinate, but only if it is treated as a decision-support technology rather than a substitute for public judgement. The optimistic case is real: AI can scan floods of data, model trade-offs, spot emerging crises, summarise competing proposals, and help institutions act before problems become disasters. That matters for an AI bloom because abundance, longevity, clean energy, scientific progress, and civilisational resilience all depend on people, firms, governments, and countries being able to make compatible decisions at scale.

Overview image for Coordination The danger is just as central. The same systems that could improve planning and negotiation can also flood societies with plausible falsehoods, synthetic identities, automated lobbying, deepfakes, and overconfident advice. Coordination depends on shared facts and mutual trust. If AI weakens those foundations faster than it improves institutions, it could make civilisation less able to handle the future it is creating. The real question is therefore not whether AI can “solve coordination”, but what kinds of institutions can use AI without becoming confused, captured, or brittle.

Why coordination is a bottleneck in the AI bloom

Many of the largest benefits imagined in an AI-enabled human bloom are not limited by invention alone. A cure has to be tested, approved, manufactured, paid for, and distributed. Clean energy has to pass through grids, planning rules, supply chains, and political consent. Safer superintelligence requires labs, states, researchers, investors, and publics to make decisions under uncertainty rather than racing blindly. Even when technical knowledge improves, societies still need ways to agree what to do with it.

This is why coordination deserves its own place in the wider AI bloom story. Intelligence can be abundant in the narrow sense while collective action remains scarce. A world full of powerful models but weak institutions could become faster, richer, and more unstable at the same time. The optimistic version is different: AI becomes a tool for shared situational awareness, better negotiation, and more competent public action.

There are three main mechanisms:

Seeing the whole system sooner. AI can combine satellite images, sensor feeds, public records, logistics data, economic indicators, and local reports into usable summaries.
Exploring options before acting. Models can simulate scenarios, reveal trade-offs, and test plans against constraints.
Helping groups find workable agreement. AI can summarise many viewpoints, detect common ground, translate expert knowledge, and support negotiation teams.

None of these mechanisms removes politics. They make politics more informed only when people can inspect the evidence, challenge assumptions, and hold decision-makers accountable.

Coordination illustration 1

Better planning and crisis response

The strongest near-term case for AI-assisted coordination is crisis response. Disasters create exactly the kind of situation where human institutions struggle: information arrives too quickly, damage is unevenly distributed, roads and communications fail, and agencies need to decide where to send scarce help. AI is useful here because it can turn messy, high-volume signals into a rough operational picture.

A concrete example is disaster damage assessment from imagery. FEMA’s published AI use-case inventory says its Response Geospatial Office is exploring computer vision and machine learning to prioritise satellite, aerial, and radar imagery after disasters, so analysts can focus on areas likely to contain damaged buildings, debris, or concentrated need. The agency still keeps geospatial analysts in the loop, which is crucial: the AI narrows the search, but people interpret and act on the result. [Department of Homeland Security]dhs.govDepartment of Homeland SecurityFederal Emergency Management Agency – AI Use Cases | Homeland SecurityApril 29, 2025…Published: April 29, 2025

Recent research shows why this matters. A 2026 AAAI paper described an AI system deployed during US federally declared disaster responses to Hurricanes Debby and Helene. Disaster drone teams were producing between 47GB and 369GB of imagery per day, more than experts at the scene could realistically transmit and interpret quickly. The deployed model assessed 415 buildings in about 18 minutes after being trained on a large post-disaster drone imagery dataset and with 91 disaster practitioners trained in its operational use. [AAAI Publications]ojs.aaai.orgAAAI PublicationsDeploying Rapid Damage Assessments from sUAS Imagery for Disaster Response | Proceedings of the AAAI Conference on Artif…

This is not a story about replacing emergency managers. It is a story about reducing the delay between “something has happened” and “we know roughly where the worst damage is”. In a flood, wildfire, earthquake, or storm, that delay can determine whether aid routes, medical teams, generators, shelter, and repair crews are sent to the right places.

Weather and flood warning show the same coordination lesson. Reuters reported in 2024 that AI-enhanced forecasting helped predict intense European rainfall, but did not prevent damage because forecasts still had to be translated into communication, preparedness, infrastructure, and public action. Better prediction is necessary but not sufficient; warnings have to reach the right people in time and be trusted enough to change behaviour. [Reuters]reuters.comAI enhances flood warnings but cannot erase risk of disasterAI enhances flood warnings but cannot erase risk of disaster The World Meteorological Organization made the same point after deadly Spanish floods: early warnings need to reach those who need them and lead to informed early action. [Reuters]reuters.comEarly action can mitigate flood destruction, UN climate agency saysEarly action can mitigate flood destruction, UN climate agency says

For civilisation-scale coordination, that is the key pattern. AI can improve the map, the forecast, and the option set. It cannot by itself create competent agencies, social trust, maintained infrastructure, emergency budgets, or clear chains of responsibility.

From prediction to joined-up public action

Governments already use AI in ways that are closer to coordination than simple automation. The OECD’s 2025 report on AI in core government functions analysed 200 public-sector AI use cases. It found that 57% supported automated, streamlined, or tailored services; 45% enhanced decision-making, sense-making, or forecasting; and 30% aimed to improve accountability or anomaly detection. [OECD]oecd.orggoverning with artificial intelligence 398fa287governing with artificial intelligence 398fa287

Those categories matter because coordination often fails in the boring middle of government: incompatible databases, slow paperwork, fragmented budgets, poor procurement, duplicated services, and a lack of timely feedback. AI can help institutions notice patterns across these silos. It may flag fraud, predict service demand, triage cases, summarise consultation responses, or help civil servants compare policy options.

But the OECD also warns that government AI adoption is often stuck in pilots, held back by skill gaps, legacy IT, poor data access, limited impact measurement, outdated rules, tight budgets, and public resistance. [OECD]oecd.orgOpen source on oecd.org. These obstacles are not peripheral. They are the difference between AI as a demo and AI as a durable institutional capability.

The G7 public-sector AI toolkit makes a similar point: AI can improve internal operations, policy effectiveness, responsiveness, transparency, and accountability, but only if principles for safe, secure, and trustworthy AI are translated into practical policies and governance arrangements. [OECD]oecd.orgg7 toolkit for artificial intelligence in the public sector 421c1244 eng7 toolkit for artificial intelligence in the public sector 421c1244 en In plain terms, governments need boring foundations before they can use powerful tools well: data quality, procurement competence, audit trails, skilled staff, privacy protections, redress mechanisms, and the ability to stop a system that is causing harm.

The bloom-relevant version of public-sector AI is therefore not “AI government” in which models rule. It is better institutional memory, faster synthesis, earlier warning, clearer accountability, and less administrative friction. Used well, AI can help a state remember what it already knows and act before small failures accumulate into large ones.

AI as a negotiation assistant, not an automated diplomat

Another promising but delicate use is negotiation. Civilisation’s hardest problems often involve parties that do not trust one another: military conflicts, climate finance, water sharing, migration, trade rules, AI safety standards, and crisis de-escalation. AI may help mediators by organising information, tracking commitments, identifying possible package deals, modelling incentives, and detecting when a proposal gives each party enough to keep talking.

A 2026 Center for Strategic and International Studies report argues that AI could support conflict mediation because mediation teams are often small, time-constrained, and short of secure analytical support. The report highlights tasks such as managing fragmented information, mapping negotiation space, designing confidence-building measures, anticipating spoilers, and creating verification mechanisms for more durable agreements. [CSIS]csis.orgAI and the Future of MediationAI and the Future of Mediation

This is a plausible coordination gain. Negotiations fail not only because people disagree, but because they misunderstand each other’s constraints, miss mutually acceptable trades, or cannot verify promises. AI could help a mediator ask: which issues are linked, which concessions are symbolic rather than material, which guarantees matter most, and where might a spoiler derail implementation?

Research on AI negotiation is still immature, however. Work on AI agents in the board game Diplomacy shows that AI can model negotiation and strategic communication in constrained environments, but game settings are far simpler than real diplomacy, where legitimacy, history, trauma, law, domestic politics, and moral stakes matter. [Nature]nature.comOpen source on nature.com. More recent work on AI negotiation agents suggests that classic human negotiation principles, such as warmth, preparation, relationship-building, and assertiveness, still matter even when agents negotiate with each other; AI-specific tricks do not erase the underlying social nature of agreement. [arXiv]arxiv.orgSource details in endnotes.

The sensible conclusion is restrained optimism. AI can help negotiation teams prepare, search the space of possible deals, and monitor implementation. It should not be treated as a neutral world arbiter. Models inherit data biases, can optimise for the wrong objective, and may make a settlement appear cleaner than it is. A peace agreement, climate bargain, or AI treaty needs legitimacy, not merely optimisation.

Coordination illustration 2

Civic deliberation and the search for common ground

AI may also help societies coordinate below the level of formal diplomacy. Public debates often collapse into binary camps even when many people share intermediate concerns. Tools that map opinion, summarise arguments, and surface areas of agreement could make democratic decision-making less performative and more useful.

Taiwan’s vTaiwan process is an important early example, although it should not be overstated. It used digital participation tools, including Polis, to help citizens and officials discuss contentious technology policy questions such as Uber regulation. Reporting on Audrey Tang’s work in Taiwan describes the aim as making common ground more visible rather than pushing people deeper into polarised threads. [WIRED]wired.comTaiwan is making democracy work again. It's time we paid attentionTaiwan is making democracy work again. It's time we paid attention

The AI governance world has begun experimenting with similar ideas. The Collective Intelligence Project’s Alignment Assemblies argue that decisions about model releases, acceptable risks, and model behaviour are already being made by a small number of actors, while the consequences affect many more people. Its OpenAI risk-prioritisation process involved 1,000 demographically representative Americans in 2023, using a wiki-survey format to rank and submit statements about what should matter when making AI safe for the public. Participants prioritised governance, oversight, avoiding overreliance on poorly understood systems, and misuse risks such as misinformation. [The Collective Intelligence Project]cip.orgSource details in endnotes.

This points to a more democratic version of AI-assisted coordination: not asking a model what society should value, but using AI and digital tools to hear more people, structure disagreement, and show decision-makers where legitimacy problems lie.

There are limits. Online participation can be shallow, unrepresentative, gameable, or too easily reduced to survey aggregation. A New Yorker critique of AI “constitutions” and corporate democratic-input exercises warns that scalable digital opinion-gathering may still leave a democratic-legitimacy deficit if it replaces deeper forms of representation and deliberation. [The New Yorker]newyorker.comThe New Yorker Does A.I. Need a Constitution?The article sets these developments against the broader backdrop of constitutional instability in the U.S., especially under Donald Trump… That warning is healthy. AI can help publics speak more clearly, but it cannot make private companies or governments democratically legitimate by branding feedback as participation.

Misinformation and trust failure

The deepest risk is that AI improves coordination for bad actors faster than for public institutions. Cheap generation makes it easier to produce fake articles, forged images, realistic audio, synthetic videos, mass comments, personalised scams, fake local campaigns, and automated influence operations. Coordination can be weaponised.

The UK government’s assessment of generative AI risks warned that synthetic media could pollute the public information ecosystem, influence public debate, manipulate markets, undermine criminal justice, and erode trust in government. [GOV.UK]GOV.UKsafety and security risks of generative artificial intelligence to 2025 annex bsafety and security risks of generative artificial intelligence to 2025 annex b The point is not just that people may believe individual fakes. It is that people may stop trusting real evidence because fakes are everywhere. A society that cannot agree what happened struggles to coordinate on what to do next.

Recent election evidence shows the problem is no longer theoretical. The UK Electoral Commission launched a deepfake detection pilot ahead of May 2026 elections in England, Scotland, and Wales, saying AI tools had made convincing deepfakes faster and cheaper to produce. It noted that during the 2024 UK general election, more than half of surveyed voters reported seeing misleading information about parties or candidates, and around a quarter reported seeing or hearing a deepfake. [Electoral Commission]electoralcommission.org.ukSource details in endnotes.

A May 2026 Guardian report on Demos research found that major AI chatbots gave inaccurate election-related answers during testing before the Scottish election, with 34% of answers containing misinformation across tested systems. Reported failures included fabricated scandals, incorrect election dates, and false voter requirements. [The Guardian]theguardian.comThe Electoral Commission warned that AI-driven misinformation compromises public trust and urged ministers to impose legal duties on AI p… That example matters because misinformation did not have to come from a hostile campaign. It came from systems that many users treat as convenient information tools.

The evidence is not one-sided. A 2025 study on conversational AI and political knowledge found that, in experiments, AI conversations increased political knowledge about as much as self-directed internet search, while also finding that 13% of eligible UK voters had used conversational AI for political information around the 2024 election. [arXiv]arxiv.orgSource details in endnotes. Another study found that LLM-assisted “prebunking” reduced belief in specific election myths in a US experiment and increased confidence in election integrity for at least a week. [arXiv]arxiv.orgSource details in endnotes.

That tension is important. AI can mislead voters, but it can also help correct rumours. The outcome depends on model reliability, source grounding, platform incentives, public media literacy, independent auditing, and whether trusted institutions can move quickly enough.

AI swarms and the collapse of cheap coordination

A further risk is not a single deepfake but many coordinated artificial actors. A 2025 paper on malicious AI swarms warns that future influence operations could use persistent agent networks to infiltrate communities, fabricate grassroots consensus, test messages continuously, harass targets, suppress or mobilise voters, and contaminate information environments. [arXiv]arxiv.orgSource details in endnotes.

This is a coordination problem in a sharper form. Social media already made persuasion faster and cheaper. Generative AI could make coordinated persuasion cheaper still, while also creating the appearance that many independent people have reached the same conclusion. That attacks one of the basic signals humans use for social judgement: “other people like me seem to believe this.”

The danger is not only persuasion. It is exhaustion. If every public consultation, comment period, election discussion, product review, local planning dispute, or crisis rumour becomes vulnerable to synthetic participation, institutions may find it harder to know what citizens actually think. Genuine minority concerns could be dismissed as bots. Fake outrage could be mistaken for public opinion. Officials may retreat into closed decision-making because open channels feel unmanageable.

Possible defences include provenance systems, stronger platform accountability, identity and reputation tools that preserve privacy, rapid institutional correction, independent audits, and “prebunking” that helps people recognise manipulation techniques before encountering a false claim. But content detection alone is unlikely to be enough. As synthetic media improves, coordination systems will need to verify processes, not just inspect artefacts.

Institutions that can use AI wisely

The central institutional challenge is to capture AI’s coordination benefits without handing over judgement, legitimacy, or accountability. That requires design choices, not just better models.

Good AI-assisted institutions will need several habits:

Human responsibility must remain visible. A minister, judge, emergency manager, regulator, or mediator should not be able to hide behind “the algorithm said so”. AI can inform decisions, but named institutions must own them.

Models need bounded roles. Disaster triage, document summarisation, fraud detection, public consultation analysis, and negotiation support are different tasks. Each needs different accuracy thresholds, audit methods, and appeal routes.

Evidence should be traceable. In high-stakes settings, AI outputs should point back to sources, data, assumptions, and uncertainty. A confident summary without provenance can be worse than no summary at all.

Pluralism should be designed in. A coordination system should preserve minority views, local knowledge, and dissenting expert judgement rather than averaging them away. This is especially important in public health, climate adaptation, policing, welfare, and AI governance.

Public capacity matters. If only large firms and powerful states can afford advanced coordination tools, AI may increase centralised control rather than shared flourishing. Public-interest infrastructure, open standards, civic technology, and independent research capacity are part of the bloom case.

The OECD’s findings point in the same direction: trustworthy public-sector AI depends on governance, data, infrastructure, skills, investment, procurement, and partnerships, plus guardrails for transparency, oversight, and risk management. [OECD]oecd.orgfull reportfull report These may sound mundane compared with superintelligence, but they are exactly the foundations a civilisation needs if intelligence becomes much more powerful.

Coordination illustration 3

What would count as real progress?

AI helping civilisation coordinate would not look like a single world-planning machine. It would look like many institutions becoming better at sensing, deliberating, deciding, and correcting themselves.

In the near term, progress would include faster disaster assessment, more reliable early warnings, better public-service triage, transparent use of AI in government, stronger election information safeguards, and participatory tools that help citizens see where they agree and where they genuinely differ.

In the medium term, it would include AI-supported treaty monitoring, climate adaptation planning, safer financial-risk detection, better pandemic preparedness, and public institutions that can absorb scientific progress without losing legitimacy. In the long term, if AI capabilities become vastly more powerful, coordination may become one of the decisive questions for humanity’s future: whether advanced intelligence amplifies rivalry and manipulation, or helps humans build institutions wise enough to manage abundance, risk, and expansion.

The fairest answer is therefore conditional. AI can help civilisation coordinate when it improves shared understanding, widens meaningful participation, and strengthens accountable institutions. It can harm coordination when it manufactures false consensus, corrodes trust, concentrates decision power, or lets leaders outsource judgement. For an AI bloom, coordination is not a side benefit. It is one of the gates through which every larger benefit must pass.