Foundations and Models of Communication

Introduction

Communication is the coordinated creation and interpretation of meaning. It is central to learning, teamwork, leadership, research, software development, and the responsible use of artificial intelligence. In an AI and machine-learning workplace, technical competence alone is insufficient. Professionals must explain model behaviour, document assumptions, interpret user requirements, present evidence, negotiate priorities, and communicate uncertainty to people with different levels of technical knowledge.

Communication is often described as the transfer of information from one person to another. This description is useful but incomplete. Communication is not merely movement of data. It includes purpose, interpretation, context, relationships, culture, feedback, and the possibility of misunderstanding. A message can be transmitted accurately at the physical level yet interpreted incorrectly at the human level.

This chapter introduces the terminology, purposes, contexts, and processes of communication. It then compares linear, interactional, and transactional models, with particular attention to the Shannon–Weaver model, the SMCR model, encoding, decoding, channel selection, feedback, and noise. The chapter concludes by applying these ideas to a realistic AI project.

Learning Objectives

By the end of this chapter, you should be able to:

1. Define communication and distinguish related concepts such as message, meaning, channel, feedback, and noise.

2. Explain the purposes and contexts of communication.

3. Describe encoding, decoding, channel selection, feedback, and barriers.

4. Compare linear, interactional, and transactional models.

5. Explain the Shannon–Weaver and SMCR models and their limitations.

6. Select communication channels appropriately for different audiences and purposes.

7. Analyse communication problems in an AI/ML project and recommend improvements.

8. Apply ethical and professional principles to technical communication.

1. Communication as Meaning-Making

Communication is a process through which people or systems create, exchange, interpret, and respond to messages within a context. The term process emphasizes that communication unfolds over time rather than occurring as a single, isolated act. Meaning is not contained entirely in words; it is constructed through the relationship between a message, its sender, its receiver, and the situation in which it occurs.

A message is the content being communicated. It may consist of spoken words, writing, images, gestures, code, a graph, a facial expression, or a combination of these. Information refers to data or knowledge that reduces uncertainty for a receiver. Meaning is the significance a receiver assigns to a message. These concepts overlap but are not identical. A dashboard may contain accurate information, but a manager may misunderstand its significance if the labels, scales, or context are unclear.

Communication may be intentional or unintentional. A data scientist intentionally sends a project update, while their silence in a meeting may unintentionally suggest disagreement or lack of preparation. Communication may also be verbal or non-verbal. Verbal communication uses linguistic symbols; non-verbal communication includes posture, timing, visual design, tone, and physical distance. In written digital communication, formatting, response time, punctuation, and emojis may also influence interpretation.

Communication is effective when the intended meaning is understood sufficiently for the purpose at hand. Effectiveness does not require identical interpretations in every detail. A technical briefing is effective if its audience understands the decision required, the evidence supporting it, and the relevant limitations.

2. Purposes of Communication

Communication serves several purposes, often simultaneously.

Informing

Informative communication provides facts, explanations, procedures, or status updates. Examples include a model card, an experiment report, an API specification, and a lecture. Effective information is accurate, relevant, organized, and appropriately detailed.

Expressing and sharing ideas

Communication allows individuals to express observations, opinions, emotions, values, and creative ideas. In professional environments, people need to communicate uncertainty and concern, not only conclusions.

Persuading and influencing

A proposal, presentation, or recommendation seeks to influence decisions or behaviour. Ethical persuasion uses accurate evidence, acknowledges alternatives, and avoids manipulation.

Coordinating activity

Projects depend on communication about roles, deadlines, dependencies, and changes. In an ML project, coordination may involve data engineers, domain experts, product managers, legal advisers, and users.

Building relationships and identity

Communication establishes trust, belonging, credibility, and professional identity. Respectful listening and reliable follow-through are communication behaviours as important as eloquent speaking.

Controlling and regulating

Policies, instructions, standards, and feedback guide behaviour. Communication in this function should be clear about authority, responsibility, consequences, and avenues for appeal.

3. Contexts of Communication

The same words can have different effects in different contexts. Context includes the physical, social, cultural, psychological, temporal, and technological conditions surrounding communication.

Intrapersonal communication

Intrapersonal communication occurs within an individual, including reflection, planning, and self-talk. A programmer mentally tests an explanation before presenting it. Critical thinking and self-monitoring improve intrapersonal communication, although internal assumptions can also create bias.

Interpersonal communication

Interpersonal communication occurs between individuals. It may be face-to-face, written, synchronous, or asynchronous. It is influenced by roles, status, trust, prior experience, and emotional state.

Group and team communication

Group communication involves several participants who share tasks or decisions. It requires turn-taking, coordination, conflict management, and shared records. Team members may have different technical vocabularies and professional priorities.

Organizational communication

Organizational communication flows vertically, horizontally, and across departments. Formal channels include policies, meetings, reports, and ticketing systems. Informal channels include conversations and professional networks. Both can influence decisions, but important decisions should be documented through appropriate formal channels.

Public and mass communication

Public communication addresses a larger audience, while mass communication distributes messages through media to many people. A public announcement about an AI system must account for accessibility, public trust, potential harm, and the possibility that excerpts will circulate beyond the original audience.

Computer-mediated and human–computer communication

Email, collaboration platforms, video meetings, dashboards, chatbots, and interfaces mediate communication through technology. AI systems also produce outputs interpreted by humans. Designers therefore need to consider not only whether an output is computationally generated, but how users will understand its confidence, scope, and limitations.

4. Elements of the Communication Process

Although models simplify reality, most communication processes include the following elements:

• Source or sender: the person, group, organization, or system initiating a message.
• Receiver: the person, group, or system for which the message is intended.
• Message: the symbolic content being expressed.
• Encoding: converting an idea or intention into symbols, language, images, or signals.
• Decoding: interpreting symbols and constructing meaning from them.
• Channel: the medium used to carry the message, such as speech, email, video, a report, or an interface.
• Feedback: a response that indicates reception, interpretation, agreement, confusion, or action.
• Noise: any factor that interferes with message transmission or interpretation.
• Context: the circumstances and relationships in which communication occurs.
• Effect: the resulting change in knowledge, attitude, emotion, or behaviour.

Encoding and decoding are not mirror-image mechanical operations. A sender chooses symbols using their knowledge and assumptions; a receiver interprets them using a different knowledge base and set of expectations. Consequently, even a grammatically correct message may fail.

For example, the statement “the model is accurate” is incomplete unless accuracy is defined, the evaluation data are identified, and the relevant benchmark is stated. A domain expert may interpret accuracy as clinical usefulness, while a machine-learning engineer may interpret it as a metric calculated on a test set.

5. Linear Models of Communication

A linear model represents communication as a one-way process from source to receiver. It is useful when the primary concern is transmission, reach, or sequence. A public announcement, recorded lecture, or automated notification may be analysed linearly.

The basic structure is:

Source → Encoding → Message/Channel → Decoding → Receiver

Linear models highlight the importance of message design and channel quality, but they tend to understate feedback, context, relationships, and the receiver’s active role.

The Shannon–Weaver model

Claude Shannon developed a mathematical theory of communication for electronic signal transmission, and Warren Weaver later discussed its broader implications. The original model contains an information source, transmitter, signal, communication channel, receiver, and destination. A noise source may distort the signal during transmission.

In a human communication interpretation:

Information source → transmitter/encoder → signal through channel → receiver/decoder → destination

Noise can be physical, technical, semantic, psychological, or organizational. Physical noise includes poor audio or a crowded room. Technical noise includes a failed connection or corrupted file. Semantic noise occurs when words or symbols have different meanings. Psychological noise includes anxiety, distraction, or prejudice. Organizational noise can arise from hierarchy, unclear responsibilities, or excessive approval layers.

The model is particularly valuable for diagnosing where a message may be disrupted. It also encourages attention to redundancy, clarity, and channel reliability. However, it was not originally designed as a complete theory of human interaction. It treats communication primarily as signal transmission and does not fully represent mutual influence, culture, power, or continuously changing context.

Example of Shannon–Weaver analysis

Suppose a project lead sends an email stating that a model must be “production-ready by Friday.” The lead encodes an intended deadline and standard. The email is the channel. A developer decodes “Friday” as the end of the working day, while the lead means the beginning of Friday. If the organization has no shared definition of “production-ready,” semantic noise produces different interpretations even though the email is delivered perfectly. A feedback question such as “Which acceptance criteria must be met, and by what time?” can expose and resolve the ambiguity.

6. The SMCR Model

David Berlo’s SMCR model organizes communication into four components: Source, Message, Channel, and Receiver. It emphasizes the characteristics of communicators and the design of messages.

Source and receiver factors

Berlo identifies factors that influence both source and receiver:

• communication skills;
• attitudes toward the topic, self, and other participants;
• knowledge level;
• social system, including roles and norms; and
• culture.

A machine-learning researcher may have strong technical knowledge but limited knowledge of a particular industry. A domain specialist may have the opposite profile. Successful communication bridges these differences rather than assuming that one audience is universally knowledgeable.

Message factors

Message considerations include content, elements, treatment, structure, and code. Content is what is said. Elements include verbal and non-verbal components. Treatment concerns how the content is presented. Structure concerns organization and relationships among ideas. Code refers to the symbolic system, such as natural language, mathematics, diagrams, or programming notation.

Channel factors

Berlo discusses channels in relation to sensory modes, such as hearing, sight, touch, smell, and taste. In professional practice, channel selection also includes media properties: speed, permanence, interactivity, reach, richness, accessibility, and confidentiality.

The SMCR model is helpful for planning an audience-centred message. Its limitation is that it may appear too static and may not explicitly show feedback or the continuously changing nature of interaction.

7. Interactional and Transactional Models

Interactional model

The interactional model improves on the linear model by adding feedback. Communication is viewed as a cycle: one participant sends a message, the other interprets it and responds, and the first participant interprets that response. The participants alternate between sender and receiver roles.

Feedback may be verbal (“I understand”), non-verbal (a nod or pause), written (a revised document), or behavioural (implementing a recommendation). Feedback can be immediate or delayed. A live conversation permits rapid clarification, whereas feedback on a published report may arrive days later.

The interactional model also emphasizes field of experience: the knowledge, values, language, and experiences that participants bring to communication. The greater the overlap between fields of experience, the easier shared interpretation may be, although shared experience does not guarantee agreement.

Transactional model

The transactional model presents communication as simultaneous and relational. Participants send and receive messages at the same time, and meaning is co-created within a context. Tone, silence, timing, posture, and environmental conditions influence the interaction while it occurs.

This model is especially suitable for face-to-face meetings, teamwork, negotiation, interviews, and intercultural communication. It recognizes that communication changes relationships and that power can affect whose interpretation is accepted. It also explains why a later clarification cannot always erase the effect of an earlier message.

Comparing the models

Linear models are useful for examining one-way announcements, transmission quality, and message reach. Interactional models are useful for feedback, clarification, and turn-taking. Transactional models are useful for complex, relational, culturally situated communication in which participants jointly construct meaning.

No model is universally best. Models are analytical tools. A recorded safety instruction may be designed linearly but followed by an interactional question session. A team meeting may be transactional, while its minutes create a later linear record.

8. Feedback, Noise, and Barriers

Feedback is not merely a reply; it is evidence about how communication has been received and interpreted. Effective feedback should be timely, specific, relevant, and directed toward improvement. In technical work, confirmation should often be behavioural or task-based rather than limited to “yes.” Asking a colleague to restate the decision or demonstrate a procedure can reveal misunderstanding.

Noise may occur at any stage. Common barriers include:

• Language barriers: unfamiliar terminology, translation problems, or ambiguous wording.
• Cognitive barriers: information overload, poor attention, confirmation bias, and premature conclusions.
• Emotional barriers: fear, defensiveness, stress, or lack of trust.
• Cultural barriers: different norms for disagreement, hierarchy, time, silence, and politeness.
• Status barriers: junior staff may hesitate to challenge senior decisions.
• Technological barriers: inaccessible documents, unstable connections, poor interface design, or incompatible formats.
• Organizational barriers: unclear ownership, fragmented tools, and undocumented decisions.

Noise reduction requires more than repeating a message. A sender may simplify language, define terms, use examples, provide a visual representation, select a richer channel, invite questions, and verify understanding. Receivers also have responsibility: they should ask clarifying questions, identify assumptions, and avoid pretending to understand.

9. Channel Selection

A channel should be selected according to communication purpose, audience, urgency, complexity, sensitivity, need for interaction, accessibility, and record-keeping requirements.

A brief routine update may be appropriate for a messaging platform. A complex design decision may require a meeting supported by diagrams and followed by written documentation. Sensitive performance feedback generally requires a private, interactive channel rather than a public message. A formal compliance decision requires an authorized and retrievable record.

Channel richness refers broadly to the capacity of a medium to convey cues, enable immediate feedback, and support personalization. Richer channels are often useful for ambiguity, conflict, and emotionally sensitive issues. Leaner channels can be efficient for routine, standardized, and unambiguous information. Richness does not mean that a channel is always superior: video meetings can create fatigue, exclude participants with poor connectivity, and make later retrieval difficult.

Best practice is often a channel combination: discuss a complex issue synchronously, share accessible supporting material, and record the decision, responsible person, assumptions, and deadline in a durable system.

10. Communication in AI and Machine Learning

AI and ML work create distinctive communication challenges. Teams must translate among technical, business, legal, and public perspectives. Terms such as bias, accuracy, fairness, explainability, and confidence may have different meanings for different audiences.

A responsible technical explanation should identify:

• the system’s purpose and intended users;
• the data and assumptions involved;
• the evaluation method and relevant metrics;
• known limitations and uncertainty;
• the consequences of errors;
• human oversight and escalation procedures; and
• how users can challenge or correct an output.

A metric should never be presented without its context. For example, a high test accuracy does not establish fairness across demographic groups, robustness under distribution shift, or suitability for a high-stakes decision. Communication that hides these distinctions can create false confidence and contribute to harm.

AI-generated communication also requires governance. Users should know when content has been generated or substantially transformed by an AI system where that information affects trust or decision-making. Sensitive data should not be entered into unauthorized tools. Teams should maintain records of prompts, sources, review processes, and approvals when these are relevant to accountability. Human review is particularly important for high-impact decisions, but “human in the loop” is not meaningful if the reviewer lacks authority, time, training, or access to the evidence needed for review.

Accessibility is another professional obligation. Use readable contrast, meaningful headings, captions, alternative text for important images, plain language where possible, and formats compatible with assistive technologies.

11. Illustrative Case: Communicating an ML-Based Loan Screening System

Background

A fictional financial-technology company develops a machine-learning system to prioritize loan applications for human review. The data-science team reports that the system achieves strong predictive performance on historical test data. A product manager wants to deploy it quickly.

Problem

At a launch meeting, the data scientist says the model is “highly accurate and unbiased.” The compliance officer asks what those terms mean. The data scientist is referring to overall accuracy and a fairness metric calculated for one protected group comparison. The product manager interprets the statement as evidence that the system is safe for automatic approval decisions. The model card has not yet been completed, and no clear appeal process exists.

Analysis

The case illustrates semantic noise, status barriers, and channel mismatch. A short meeting statement compressed several technical claims into two broad adjectives. The audience members had different fields of experience and different decision responsibilities. Because the meeting was not followed by a structured written record, assumptions remained unexamined.

The linear transmission of the performance claim did not create shared meaning. An interactional process would have encouraged questions, while a transactional perspective would recognize that the participants’ roles and power relationships shaped what was said and accepted.

Approach and solution

The team redesigns the communication process:

1. The data scientist defines accuracy, the evaluation population, the comparison baseline, and the relevant error types.

2. Fairness is reported using explicitly named measures and groups, with an explanation that no single metric establishes overall fairness.

3. A concise executive summary is prepared for managers, while a technical appendix provides methods and results.

4. A private workshop includes compliance, affected-domain experts, engineering, and product staff.

5. Participants use a question-and-restatement process: each decision owner explains their understanding of the proposed use.

6. The team documents intended use, prohibited use, limitations, monitoring indicators, human review responsibilities, and an appeal route.

7. Deployment begins as a decision-support pilot rather than automatic approval, with periodic review for performance and disparate impact.

Outcome

The company delays full deployment but reduces the risk of treating a predictive score as a definitive decision. Managers understand what the evidence does and does not establish. Reviewers receive guidance on when to override or escalate a recommendation. The written record supports accountability and later evaluation.

Learning points

The quality of communication is part of system quality. Technical accuracy does not guarantee communicative accuracy. Effective AI communication requires defined terms, audience adaptation, feedback, documentation, transparency about uncertainty, and channels appropriate to risk.

12. Professional Best Practices

Professionals can improve communication by:

• identifying the audience, purpose, and desired action before composing a message;
• separating facts, interpretations, assumptions, and recommendations;
• defining specialized terms and avoiding unexplained acronyms;
• using concrete examples and appropriately designed visuals;
• presenting uncertainty and limitations rather than overstating confidence;
• choosing a channel that supports the required interaction and record;
• checking accessibility and confidentiality;
• inviting questions and verifying understanding;
• documenting decisions, owners, deadlines, and changes;
• listening actively and acknowledging legitimate disagreement; and
• reviewing the ethical consequences of how information is framed.

Communication competence is not a fixed personal trait. It develops through practice, feedback, reflection, and familiarity with different audiences and contexts.

Challenges and Limitations of Communication Models

Models necessarily simplify communication. A diagram may separate sender and receiver even though real participants influence each other simultaneously. Models can also give the impression that meaning is transferred intact, whereas meaning is interpreted. They may underrepresent culture, emotion, power, identity, accessibility, and institutional history.

The Shannon–Weaver model is strong for analysing signal transmission and noise but limited as a social theory. The SMCR model supports audience and message analysis but gives less explicit attention to feedback. Interactional models represent feedback but may imply alternating rather than simultaneous activity. Transactional models better represent complexity but can be less precise for diagnosing a particular transmission failure.

The appropriate response is not to reject models but to use several lenses. Ask both whether the message was transmitted and whether it was understood; both whether a channel was available and whether people felt able to respond; both whether the content was accurate and whether its framing encouraged a responsible interpretation.

Chapter Summary

Communication is a process of creating, exchanging, interpreting, and responding to messages in context. Its purposes include informing, expressing, persuading, coordinating, building relationships, and regulating behaviour. Communication operates intrapersonally, interpersonally, in groups, organizations, public settings, and technology-mediated environments.

The main elements are source, message, encoding, channel, decoding, receiver, feedback, noise, context, and effect. Linear models emphasize one-way transmission. The Shannon–Weaver model is useful for analysing channels and noise, although it does not fully capture human meaning and relationships. Berlo’s SMCR model emphasizes source and receiver characteristics, message design, and channel factors. Interactional models add feedback and alternating roles. Transactional models treat communication as simultaneous, contextual, and relational.

Effective communication requires appropriate terminology, audience analysis, channel selection, active listening, feedback, documentation, and attention to accessibility and ethics. In AI and ML, professionals must communicate metrics, assumptions, uncertainty, limitations, and human responsibilities without creating false confidence. Communication is therefore not an optional soft skill; it is part of responsible technical practice.

Key Terms

Communication: The process of creating, exchanging, interpreting, and responding to messages.

Message: The symbolic content communicated.

Meaning: The significance assigned to a message by an interpreter.

Source: The person, group, or system initiating communication.

Receiver: The intended interpreter or recipient of a message.

Encoding: Converting an idea or intention into symbols or signals.

Decoding: Interpreting symbols and constructing meaning.

Channel: The medium through which a message is carried.

Feedback: A response that provides information about reception or interpretation.

Noise: Any factor that interferes with transmission or understanding.

Context: The circumstances, relationships, culture, and environment surrounding communication.

Linear model: A one-way model in which a source transmits a message to a receiver.

Interactional model: A model that represents communication as an exchange involving feedback.

Transactional model: A model in which participants simultaneously influence one another and co-create meaning.

Shannon–Weaver model: A transmission model emphasizing source, signal, channel, receiver, destination, and noise.

SMCR model: Berlo’s model consisting of Source, Message, Channel, and Receiver.

Semantic noise: Interference caused by different meanings assigned to words or symbols.

Channel richness: The capacity of a medium to convey cues, support feedback, and enable personalization.

Field of experience: The knowledge, culture, values, and experiences participants bring to communication.

Audience adaptation: Adjusting content, language, structure, and channel for a particular audience.

Review Questions

Short Answer Questions

1. Define communication and explain why it is more than the transmission of information.

2. What is the difference between encoding and decoding?

3. Identify four types of noise and provide one example of each.

4. What role does feedback play in the interactional model?

5. State two limitations of the Shannon–Weaver model.

Long Answer Questions

1. Explain the purposes and contexts of communication, using examples from an AI/ML workplace.

2. Describe the Shannon–Weaver model and evaluate its usefulness for analysing human communication.

3. Explain Berlo’s SMCR model, including the factors that influence source, message, channel, and receiver.

4. Compare linear, interactional, and transactional models of communication.

5. Discuss principles for selecting an appropriate communication channel in a professional organization.

Analytical / Application Questions

1. An engineer sends a message saying, “The model is ready for deployment,” but the product manager assumes that security review and user testing are complete. Analyse the communication failure using encoding, decoding, semantic noise, feedback, and channel selection.

2. Design a communication plan for explaining an AI system to technical developers, senior managers, and members of the public. Explain how the message, language, channel, and feedback process should differ for each audience.

3. Using the illustrative loan-screening case, identify the ethical risks of vague communication and propose governance practices that would improve accountability and informed decision-making.

Multiple Choice Questions with Answers

Question 1

Which element of communication converts an idea into symbols or signals?

A. Decoding

B. Encoding

C. Feedback

D. Noise

Correct Answer: B. Encoding

Question 2

Which model is primarily associated with information source, transmitter, channel, receiver, destination, and noise?

A. SMCR model

B. Transactional model

C. Shannon–Weaver model

D. Interactional model

Correct Answer: C. Shannon–Weaver model

Question 3

A technical term is understood differently by a data scientist and a manager. This is an example of:

A. Physical noise

B. Semantic noise

C. Channel richness

D. Non-verbal feedback

Correct Answer: B. Semantic noise

Question 4

Which model represents communication as simultaneous and relational?

A. Linear model

B. Transactional model

C. Transmission-only model

D. One-way announcement model

Correct Answer: B. Transactional model

Question 5

In the SMCR model, which factor is most directly concerned with the symbolic system used, such as language or diagrams?

A. Code

B. Status

C. Destination

D. Effect

Correct Answer: A. Code

Question 6

Which channel is generally most suitable for a sensitive, complex issue requiring immediate clarification?

A. An anonymous mass announcement

B. A private interactive conversation

C. A short public status message

D. An unstructured data dump

Correct Answer: B. A private interactive conversation

Question 7

Why is feedback important in professional communication?

A. It guarantees that disagreement will disappear

B. It replaces the need for a clear message

C. It provides evidence about reception and interpretation

D. It eliminates all forms of noise

Correct Answer: C. It provides evidence about reception and interpretation

Question 8

Which statement best reflects responsible communication about an AI model?

A. A high accuracy score proves the model is fair

B. The model’s limitations and evaluation context should be explained

C. Technical audiences do not need definitions of metrics

D. Human oversight is unnecessary when a model is accurate

Correct Answer: B. The model’s limitations and evaluation context should be explained

Question 9

A team records a decision after a meeting, including the owner, deadline, and assumptions. Which communication benefit is most directly supported?

A. Improved documentation and accountability

B. Removal of all psychological noise

C. Conversion of transactional communication into intrapersonal communication

D. Elimination of the need for feedback

Correct Answer: A. Improved documentation and accountability

Question 10

Which statement best distinguishes an interactional model from a linear model?

A. The interactional model includes feedback between participants

B. The interactional model excludes context

C. The linear model always uses digital technology

D. The linear model requires simultaneous message exchange

Correct Answer: A. The interactional model includes feedback between participants