eduKate Singapore

VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Next useful routeCanonical hubHELP

How Intelligence Works | Common Knowledge — Public Signals, Nested Beliefs and Group Coordination

—

by

eduKate Asia

Common knowledge is more than several people knowing the same fact. It means each knows the fact, each knows that the others know it, each knows that the others know that everyone knows it, and so on through every level. This hierarchy matters when successful action depends on what people believe other people will do. Public signals can help a group coordinate, while private messages, ambiguous audiences and unreliable delivery can leave a dangerous gap between information and coordinated action.

The useful practice is to identify the decision that requires coordination, establish a shared source of truth, distinguish who received a message from who acknowledged its meaning, and choose action rules that work under finite and imperfect confirmation. A school team, emergency group or distributed organisation rarely needs to achieve the philosopher’s ideal of infinite shared knowledge. It needs an explicit operational agreement about membership, state, timing, authority and failure recovery, with no pretence that one announcement magically solved all uncertainty.

Contents

Start with the fact, then ask who knows who knows

Suppose a school announces that tomorrow’s assembly has moved indoors. If the notice is delivered privately to Alice, she may know the venue, but she may not know whether Ben received the same notice. If Ben independently knows the venue, the two have mutual first-order knowledge. Yet Alice may still hesitate to go indoors if she expects Ben to wait at the old location and she needs Ben to bring equipment. Their problem is not lack of knowledge about the venue; it is lack of knowledge about the other’s knowledge.

At the next level, Alice learns that Ben knows the venue. But Ben may not know that Alice knows he knows it. A subsequent acknowledgement helps one belief while leaving uncertainty about whether that acknowledgement was itself received. Repeating the acknowledgement adds another level of nested understanding, without guaranteeing an infinite hierarchy over an unreliable communication channel.

A shared public announcement in a model where every group member observes the same event and knows that everyone observes it can change the epistemic situation. Those model assumptions are demanding. Real organisations often approximate them through broadcasts, meetings, openly visible status and agreed decision rules. The distinction between ideal common knowledge and practical shared coordination is the backbone of this subject.

Individual, mutual, distributed and common knowledge

Individual knowledge concerns what one person knows. Mutual knowledge means that everyone in a named group knows the relevant fact, but it does not automatically follow that everyone knows that everyone knows. Distributed knowledge is different again: a group could collectively infer something by combining pieces of information held separately even when no single member knows the answer. Common knowledge is the recursively shared structure of everyone’s knowledge.

Imagine four adults organizing a science fair. One knows the hall capacity, one knows the registration count, one knows the fire-safety rule and one knows which doors are unavailable. As a group they possess the ingredients to infer whether admission is possible. Until they combine these ingredients they may have distributed knowledge of the answer but no shared conclusion. A meeting that combines facts may create mutual knowledge of a decision without establishing perfect common knowledge of every supporting detail.

Calling all of these concepts team awareness hides important failure modes. A group can have all the facts but no common decision; everyone can know the final decision but assume somebody else will implement it; or everyone can interpret a public label differently. Diagnose which level is missing before trying to fix a coordination failure with another blanket reminder.

The hierarchy of beliefs has no fixed short stopping point

Let E mean that a schedule change is valid. At level one, Alice knows E and Ben knows E. At level two, Alice knows Ben knows E, and Ben knows Alice knows E. At level three, Alice knows that Ben knows that Alice knows E, and the symmetric statements also hold. Common knowledge requires every finite level of this nested pattern in the ideal formal model, not merely three or ten acknowledgements.

Why would the higher levels matter? A participant can believe the schedule is correct yet refuse a joint action if they fear others will not act. A second person can likewise refrain because they worry the first will refrain. With high stakes and no tolerance for acting alone, the uncertainty is recursively about other people’s reasoning, not simply about the original event.

This is a conceptual tool, not an instruction to send infinite messages. In real work the remedy is to define an agreed action policy with realistic guarantees. A coordinator can specify a quorum, a fallback time, a stop rule and what to do if confirmation fails. Such rules reduce the need to infer infinitely many beliefs from incomplete conversational clues.

Why a public signal changes coordination

Private information tells someone what happened. Public information, under suitable visibility assumptions, also tells participants something about what others could have seen. Consider an emergency announcement shown on a display visible to all staff during a meeting. People can see colleagues listening and can hear the same terms, giving them reasons to believe the announcement has been jointly registered. The social setting supplies information about audiences as well as content.

That does not turn every meeting into mathematically perfect common knowledge. Someone may have been distracted, may interpret a code word differently or may join remotely with a delay. Nor does public visibility ensure the message is true. A public mistake can become highly coordinated misinformation. Authenticity, interpretation and action readiness are separate checks.

A practical public signal identifies the specific group, the effective date, the version of the decision and who has authority to change it. It should distinguish a proposed action from an approved one. If the message says all volunteers should prepare to move, that does not necessarily mean the final move is authorised. Clarity about the action state reduces the distance between hearing a signal and coordinating safely on it.

Numerical case: why acting alone can be rationally avoided

Assume two organisations must launch a joint test on the same morning. If both launch, each gains 10 utility points. If exactly one launches, that organisation loses 8 points because it incurs costs without the partner; the organisation that stays out scores 0. If neither launches, both score 0. These are illustrative utility points, not money or empirically measured payoffs. The example reveals why knowing the technical plan is not enough.

Suppose organisation A believes B will launch with probability p. A’s expected utility from launching equals 10p minus 8 times (1 minus p), or 18p minus 8. Staying out has expected utility zero. Launching is strictly preferred if 18p minus 8 is greater than zero: p must exceed 8/18, which simplifies to 4/9, approximately 44.44 percent. At p = 0.40, launching has expected utility 18 times 0.40 minus 8 = minus 0.8 points; at p = 0.60 it is 2.8 points.

But p is not necessarily B’s true probability of launching, and the payoff model hides assumptions about risk and independence. A’s beliefs can depend on whether B thinks A received the plan, which in turn depends on what A thinks B believes. A public go-or-no-go process with explicit confirmations and a cancellation rule can outperform endless attempts to persuade each side that the other’s confidence must be high.

A threshold is useful but not a magic safety rule

The example’s threshold came from a stipulated payoff table. If launching alone instead loses 30 points, A’s expected payoff would be 10p minus 30(1 minus p) = 40p minus 30, making the threshold p greater than 30/40, or 75 percent. Changing the costs changes the rational threshold; common knowledge does not eliminate the need to quantify consequences and decide who can bear them.

Organisations often substitute vague declarations such as everyone is ready for a concrete decision protocol. A better protocol defines what counts as ready, which evidence proves readiness, what percentage or named quorum must be confirmed, when the commitment becomes binding and what happens after an absence. The protocol may deliberately permit asynchronous execution or cancellation instead of demanding simultaneous action.

Notice the separation between epistemic certainty and expected-value arithmetic. A can calculate a threshold perfectly while still have an unreliable belief about B. B can know A’s threshold without knowing whether A observed the latest network status. The correct response is not to repeat the equation louder; it is to gather the missing coordination evidence or choose an action that remains acceptable under uncertainty.

The two-generals problem and lost acknowledgements

Picture two commanders separated by an unreliable messenger route. Their joint attack succeeds only if both move. One sends a message with the agreed time. The other acknowledges it, but the acknowledgement might be lost. A follow-up acknowledgement of the acknowledgement could also be lost. After any finite number of messages, the sender of the final message cannot be certain that it was delivered. This creates a persistent asymmetry in what each party can know about the other’s confidence.

The story is a simplified model of coordinated action over unreliable channels, not a literal recommendation for military behaviour. Halpern and Moses analysed the role of common knowledge in distributed environments and the limits of attaining it under communication uncertainty. The useful engineering parallel is that confirmation of a fact and confirmation that a confirmation arrived are different states.

For organisations, the lesson is to avoid pretending that a chain of acknowledgements proves absolute certainty. Implement robust alternatives: choose a safe default if acknowledgement is absent; make actions reversible where possible; allow redundant channels; define a commit authority; or use a quorum that is explicitly adequate for the risk. The appropriate policy depends on costs and consequences, not just on the elegance of a communication protocol.

Almost common knowledge may still behave differently

The 1989 electronic-mail game studied by Ariel Rubinstein shows why a finite but long chain of nested knowledge can yield strategically different behaviour from ideal common knowledge. One player learns which coordination game is active, and messages can pass between players with uncertainty. Even a large number of successful confirmations does not necessarily produce the same incentives that would follow if the state were common knowledge. This result depends on a particular game and information structure, not a claim that people never coordinate by email.

The counterintuitive point is that probability of message delivery and depth of nested beliefs are not interchangeable measures. If each additional acknowledgement is very likely, participants can become quite confident about lower-order facts. Yet their decisions may still be affected by the possibility that the other party occupies a different level of confidence.

In practical teamwork, the lesson is not to reject email but to limit what the decision depends on. A calendar invitation is often enough for an ordinary meeting. A joint release that would cause irreversible financial or physical harm should use explicit transaction-like states, responsibility checks and a recoverable stop condition. Higher stakes require stronger coordination machinery.

What Aumann’s agreement theorem does and does not say

Robert Aumann’s 1976 paper Agreeing to Disagree proves a result under specific assumptions: two Bayesian agents have a common prior, and their posterior probabilities for the same event are common knowledge. Under those assumptions, their posteriors must agree. The theorem concerns consistency of beliefs and information, not a social rule that honest people can never reasonably disagree in ordinary life.

Real people have different priors, different models, inconsistent evidence and uncertain interpretations. They may also disagree about preferences rather than factual probabilities. An organisation should not use the theorem as a shortcut to silence dissent or assert that a committee’s disagreement proves irrationality. The theorem is most useful as a lens on which assumptions would have to hold for information exchange to force agreement.

Read the author-hosted citation and statement at https://ma.huji.ac.il/~raumann/pdf/Agreeing%20to%20Disagree.htm and the publisher DOI at https://doi.org/10.1214/aos/1176343654. For a later computational analysis of how Bayesian agents may reach approximate agreement under assumptions, see Scott Aaronson’s original paper at https://arxiv.org/abs/cs/0406061. Different results answer different questions: common knowledge, truthful Bayesian updating, finite communication and computational efficiency should not be collapsed into a single slogan.

Worked probability case: the same evidence is not the same prior

Suppose a binary event initially has probability 0.20 for a particular decision maker. The prior odds are 0.20 / 0.80 = 0.25. A piece of evidence has likelihood ratio 3 in favour of the event: it is three times as likely if the event is true as if false, under the stated model. Bayes’ odds rule multiplies prior odds by the likelihood ratio, giving posterior odds 0.25 times 3 = 0.75. The posterior probability is 0.75 / (1 + 0.75) = 3/7, approximately 42.86 percent.

A second decision maker with prior probability 0.50 has prior odds 1. With exactly the same likelihood ratio 3, posterior odds become 3 and posterior probability becomes 3/4, or 75 percent. Both used the same public evidence and valid arithmetic, yet their posteriors differ because their priors differ. Calling the signal common knowledge does not impose Aumann’s common-prior assumption.

The case is an invented numerical lesson, not an observation of human participants. If the public signal itself is unreliable or differently interpreted, the gap can widen further. When colleagues disagree after an announcement, ask whether they disagree on priors, likelihoods, the event definition, the evidence’s authenticity or the action’s value. That diagnosis is more productive than repeating the final percentages.

Make announcements that genuinely narrow ambiguity

A reliable group announcement specifies who is included, what proposition is being asserted, what evidence supports it, what decision has been made, when it takes effect and what observers must do. It also identifies the authoritative revision channel. Without those elements a message can spread rapidly while different participants act on different versions. The danger grows when groups span shifts, languages or organisations with separate procedures.

Consider a notice that says tomorrow’s event starts earlier. Earlier than what, and whose local time? One person may understand 8:00, another 8:30, and a remote participant might interpret a timezone differently. A better notice gives an absolute date and local timezone, a previous and current time, and the authority responsible for the change. That makes the factual content clearer before any deeper knowledge problem is considered.

Do not add meaningless read-receipt bureaucracy to every message. Information that only affects independent actions may need nothing beyond readable instructions. Common-knowledge considerations matter most when a person’s best action depends on other people’s awareness, commitments or synchronized steps. Match the communication design to the coordination requirement.

A school coordination example with three decision states

Imagine teachers preparing a science demonstration with a room change and restricted equipment. Stage one is proposed: a coordinator suggests moving. Stage two is approved: the designated leader confirms the new room has sufficient space and the required supervision. Stage three is ready-to-execute: the responsible teacher confirms materials and safety checks at the new room. The word moved can refer to any of these stages unless the school defines it.

Suppose there are eight responsible adults, and the safety policy requires six named acknowledgements plus the supervisor’s explicit approval before proceeding. Five acknowledgements out of eight equal 62.5 percent of the group, but they do not meet a six-person count requirement. Six acknowledgements out of eight equal 75 percent; if the supervisor has not approved, the separate approval condition is still false. The denominator and the Boolean conditions matter more than a vague everyone seems informed.

Once the trigger conditions are satisfied, the organisation can record a decision version and execution time. Participants should know how to discover the current state and who can revoke approval. The process does not prove metaphysical common knowledge. It supplies a practical coordination policy that does not rely entirely on guesses about what absent or distracted colleagues might think.

Quorums, named roles and why percentages can mislead

A quorum is a decision rule about how many or which participants must be represented, not a direct measure of everyone’s nested knowledge. A team of 10 may require seven valid acknowledgements to proceed. Seven of 10 is 70 percent, but the rule may also require a particular risk owner. Seven acknowledgements from people lacking that authority could still leave the decision invalid under the policy.

Response counts can be distorted by duplicate accounts or individuals responding twice. Define the denominator as unique eligible participants, not messages. If a dashboard reports 12 acknowledgements but there are only nine eligible staff, something is wrong with identity or counting. The right repair is not a higher completion percentage; it is deduplication, authorisation checks and a traceable record of whose confirmation counts.

Quorums can also leave meaningful minorities uninformed. The rule might still be justified for a reversible low-risk operation, but a high-risk task may demand confirmation from every named operator. Choose the quorum against the actual failure consequences and avoid describing a sufficient operational quorum as proof that all participants now hold common knowledge.

Shared dashboards, audience visibility and stale state

A dashboard can create a strong common reference point if users know its URL, can access it, trust its authorship and agree that it displays the authoritative version. Yet a shared screen in a meeting and a cached screenshot in a private chat are not the same epistemic situation. Someone may act on a stale image believing that others are watching the live dashboard. The interface has made the proposition seem public without guaranteeing everyone has the same version.

A useful dashboard visibly labels its update time, state identifier, responsible owner and the distinction between requested, approved, executed and reconciled actions. It should allow a participant to check whether a colleague’s confirmation applies to the same version. If a change invalidates earlier confirmations, the product must state that explicitly instead of leaving a reassuring but misleading green badge.

Design for uncertainty. An expired or unreachable status page should say unknown or unavailable, not silently imply unchanged. A participant should not interpret failure to reach a dashboard as evidence that the previous decision remains safe. The dependable system includes a failure rule and fallback authority beyond the display itself.

Common ground in conversation is related, not identical

People maintain conversational common ground: background assumptions and propositions they treat as mutually available for the current exchange. If two friends have already agreed to meet at the library, one may say see you there without repeating the address. In a familiar setting this saves effort. With new participants, however, there may be no shared referent for there or tomorrow.

This everyday mechanism is related to common knowledge but need not meet the infinite formal definition. People tolerate ambiguity and repair it conversationally. They infer a listener’s knowledge from context, tone, gestures and prior discussion. These skills enable flexible cooperation yet can also produce false certainty. A manager’s belief that everyone understood a rushed instruction is an inference, not verified evidence.

Useful repair language is concrete: which version did you see, what will you do first, when does the rule take effect, and what would make you stop? These questions assess shared understanding and intended actions, not only whether a box was ticked. They can reveal differences that a public announcement alone failed to remove.

A public declaration is not a public preference

People may know that a decision was announced while privately opposing it. A public show of hands can also be affected by social pressure, role hierarchy or fears about dissent. Common knowledge of the announced decision does not imply common knowledge of genuine preferences. This matters in committees where the public record says unanimous but several members believe the choice is harmful.

Separate evidence of receipt, evidence of comprehension, evidence of approval and evidence of willingness or capability to act. A team may acknowledge an instruction while lacking equipment. A person may agree with a goal while objecting to its timing. A leader who collapses those distinct questions can build a seemingly coherent plan on an imaginary consensus.

For a closely related but different social mechanism, see the Abilene paradox: false agreement and hidden preferences. That article concerns why group members can collectively choose something nobody truly wants. Common knowledge asks a separate question about what each knows the others know. The problems can interact without becoming the same problem.

Being collectively certain is not being collectively correct

A rumour can be visible to everyone, repeatedly endorsed and treated as public fact. It may then support remarkably coordinated behaviour while remaining false. Common knowledge concerns the nested distribution of knowledge under an ideal factive model, but real people often coordinate on beliefs or accepted messages that are mistaken. The distinction is especially important when a group treats social repetition as verification.

Before a high-stakes public signal is issued, verify the proposition independently. For a change in examination instructions, trace the message to the institution empowered to issue it. For operational safety, ensure the approval is from an authorised role, not merely a person with a loud voice. Shared access to a wrong document increases consistency without increasing accuracy.

This is why a sound process combines two functions: evidence validation asks whether the state is correct, and coordination validation asks whether those who must act have access to the correct state and understand the action rule. Neither function can replace the other. Reliability is an interaction between truth, communication, decision authority and follow-through.

Why organisations need explicit commitment states

Cross-organisation work often fails at boundaries. One team says ready because its documents are complete; the partner interprets ready as every approval received. Both can truthfully report readiness by their local definitions. A common-status label does not solve the problem unless its definition is shared and measurable. A joint release should therefore use named states that explain what has and has not happened.

For example, prepared may mean files exist, validated may mean independent checks passed, authorised may mean a named approver signed off, committed may mean the action is now planned for an exact time, and completed may mean independent readback confirms the intended external effect. The sequence is a design illustration, not a universal standards taxonomy. Each organisation must specify which transitions require joint approval and which can happen asynchronously.

Versioning matters because announcements can cross. If one partner commits to version 8 and the other to version 9, both may believe they agreed while referencing different content. Shared identifiers, expiry times and supersession rules help. More communication is not always better; communication anchored to an explicit state machine is more useful than a high volume of messages without a common meaning.

Distributed systems expose the limits of perfect agreement

Computers also face problems about who knows which state and whether messages were received. Halpern and Moses’s research formalised knowledge in distributed environments and showed why ideal common knowledge is difficult or unattainable under certain realistic uncertainty assumptions. It also explored weaker knowledge notions appropriate to practical systems. The analogy to human organisations is useful, but the formal results should not be extended to every social conversation without matching their assumptions.

Distributed logs, consensus protocols and acknowledgements aim to make state transitions reliable despite faults. A committed record may be authoritative for a service even though no single operator has personally read the record. The engineering property required for safety may be quorum agreement, linearizable state, eventual convergence or a designated leader’s confirmed decision rather than an epistemic claim about infinite shared belief.

Do not conflate a technical commit with human assent. A protocol can establish that a value was durably recorded according to a quorum rule; it cannot establish that participants understood its ethical implications. Human decision processes need accountable authority and meaningful dissent channels in addition to reliable storage and transport.

Membership, timing and the boundary of everyone

Common knowledge is always relative to a defined group and proposition. If everyone means five current employees, what happens when a sixth joins later? If the policy changed between a morning meeting and an afternoon shift, what does it mean for the evening team to know the decision? A static statement can conceal dynamic membership and time-dependent validity.

Write down the group boundary and effective interval. For example, all nine on-duty laboratory supervisors at 09:00 Singapore time might be required to confirm version 12 of a safety notice. If a tenth supervisor takes over at 10:00, the organisation needs a handover rule. Counting nine original responses indefinitely is not proof that the new operator has seen the relevant instructions.

Authority can change, too. A former manager may know the old policy and still lack permission to issue an update. The organisation’s commitment process must authenticate who is speaking and which revisions supersede earlier messages. Clear membership and authority are prerequisites for a meaningful public signal, not bureaucratic accessories added after a failure.

How to build a workable communication protocol

Start by defining the action dependency. Ask whether everyone must move simultaneously, whether a quorum is enough, or whether people can safely act independently. Then state what counts as the shared proposition: an event time, equipment status, approval or versioned document. Only after those steps choose a medium. A broadcast chat can support awareness; a signed release record can support authority; a transactional commit process can support irreversible coordinated work.

Next, design for nonresponses and conflicting reports. A missing acknowledgement should have a predefined interpretation such as do not proceed, proceed with named quorum, or ask a specific decision owner. Silence should not be casually interpreted as assent in situations where approval has operational or ethical significance. An observer needs a way to see whether a requirement is still pending.

Finally, close the feedback loop through external verification. An announcement that a task is complete is not necessarily evidence that the expected world state changed. Verify the result independently when possible, record exceptions and make it easy to correct an earlier decision. Strong coordination includes an honest path for stopping and revising work, not only a path for launching it.

Transfer exercise: remote exam-room coordination

Four staff members must move equipment before an examination. The rule requires at least three of the four to confirm the room is empty, plus a separate authorisation from the exam officer. At 08:30, three staff confirm, and the fourth has not responded. The officer has not approved. Can the move begin? No: although three of four equals 75 percent and meets the numeric count, the independent officer condition is false.

At 08:35 the officer approves version 3 of the move plan. But one of the earlier confirmations refers to version 2, which used a different equipment route. The team now has only two valid confirmations for version 3 if confirmations are version-specific, so the move remains blocked. At 08:40 the staff member reconfirms version 3. Under the stated policy, the numeric and officer conditions are now met. The example illustrates versioned operational coordination, not mathematically perfect common knowledge.

The changed-context question is what happens if a fifth staff member joins at 08:42. The answer depends on policy: if the rule is at least three valid confirmations from the original four, membership does not change the threshold; if it is 75 percent of currently responsible staff, three of five is 60 percent and no longer enough. The denominator and membership rule must be specified before taking action.

Where the concept is useful and where it misleads

The concept is useful when people hesitate because they are uncertain what others know or believe, especially in synchronized tasks, public announcements and large-group coordination. It is also useful when diagnosing whether a team’s shared information is actually common, merely distributed or only privately available. It can turn vague frustration about communication into a precise question about audiences, versions, commitment states and dependencies.

It can mislead when used as a universal demand for consensus or absolute certainty. Human groups cannot literally verify infinite nested belief levels by checking a few chat messages. Formal impossibility results depend on specified communication models, and practical systems can still achieve useful coordination without ideal common knowledge. Leadership should not force costly acknowledgements where independent action or safe defaults would work.

Keep preferences and evidence separate from shared awareness. A disagreement can persist because values differ even when everyone knows the same facts. An announcement can be common yet false. A quorum can authorize an action while leaving some people uninformed. Each failure requires its own remedy; an additional public notification is not always the appropriate one.

Frequently asked questions

If everyone receives the same email, is the content common knowledge? Not generally. Delivery, attention and interpretation are uncertain, and each recipient may be unsure who else received it. In a formal model with a guaranteed public observation the answer can differ, but ordinary email rarely supplies all those assumptions.

Does a read receipt solve it? A read receipt adds information about one layer of receipt or access. It does not automatically guarantee comprehension, acceptance or that the sender knows the recipient knows the sender saw the receipt. For practical work it may be enough under a defined decision rule.

Is common knowledge the same as consensus? No. People can know a proposal is unanimously rejected; then their rejection may be commonly known, but they still disagree with the proposal. Conversely, people may privately favour the same plan without knowing others share that preference.

What is the difference between common knowledge and common ground? Conversational common ground is a useful practical set of assumptions treated as shared in discussion. The formal recursive definition of common knowledge is stronger. Language users often proceed with imperfect shared context and repair misunderstandings as they occur.

Can teams achieve useful coordination without perfect common knowledge? Yes. Many real systems use reliable commitments, policies, designated authorities, a quorum, safe defaults and rollback conditions. The required operational guarantee depends on the task’s risk.

Does Aumann’s theorem say rational people must agree on everything? No. It applies to common-prior Bayesian agents under common knowledge of posteriors for a specified event, among other formal assumptions. Ordinary disagreements about evidence, definitions, models and preferences lie outside the simple slogan.

Why mention computer networks in an intelligence article? Distributed-system research makes the nested-knowledge problem unusually explicit. It illustrates how messages alter knowledge states and why protocol safety often relies on carefully defined weaker conditions rather than imaginary perfect information.

Primary research and boundaries

Robert J. Aumann, Agreeing to Disagree, The Annals of Statistics (1976), original author-hosted citation and abstract: https://ma.huji.ac.il/~raumann/pdf/Agreeing%20to%20Disagree.htm. Publisher DOI: https://doi.org/10.1214/aos/1176343654. The theorem is conditional on a common prior and common knowledge of relevant posteriors; it is not a behavioural claim about every disagreement.

Joseph Y. Halpern and Yoram Moses, Knowledge and Common Knowledge in a Distributed Environment, Journal of the ACM (1990). Original research record: https://research.ibm.com/publications/knowledge-and-common-knowledge-in-a-distributed-environment ; author manuscript: https://arxiv.org/abs/cs/0006009. This work provides a formal distributed-systems framework and analyses knowledge limits under particular system assumptions.

Ariel Rubinstein, The Electronic Mail Game: Strategic Behavior under Almost Common Knowledge, American Economic Review (1989), publication record: https://ideas.repec.org/a/aea/aecrev/v79y1989i3p385-91.html. Scott Aaronson, The Complexity of Agreement, original paper: https://arxiv.org/abs/cs/0406061. The game-theoretic and computational claims should be read with their distinct assumptions. All numerical examples in this article are constructed for explanation, not reported empirical findings.

For a different failure of group understanding, see How Intelligence Works | Abilene Paradox — False Agreement, Hidden Preferences and Group Decisions. It concerns a group mistakenly acting as though a disliked choice is mutually wanted, which can occur despite extensive public talk.

For deadline planning where personal expectations differ from reference-class evidence, see How Intelligence Works | Planning Fallacy — Inside Views, Reference Classes and Reliable Deadlines. To explore other human and organisational mechanisms, use the How Intelligence Works hub. These pages address different reasoning jobs; a reader can move between them without confusing consensus, coordination and prediction.

Discover more from eduKate Singapore

Subscribe now to keep reading and get access to the full archive.

Continue reading