A note on appropriate and inappropriate
employment of generative AI to support Housing Studies students’ doing
dissertation projects
Using
generative AI to augment part-time Housing Studies students' learning when
doing their dissertation projects is a common practice. Suggest 4 ways of this
practice as conductive to effective learning and suggest 4 ways that are
ineffective, harmful to learning and/ or academic dishonest behaviour.
Generative AI can be a powerful learning aid for
part-time Housing Studies students working on dissertations, but only when used
to augment—not replace—human judgment, critical engagement, and scholarly
rigour. Below are four ways this practice can be conducive to effective
learning, followed by four ways it can be ineffective, harmful, or academically
dishonest.
Conducive
to effective learning
1.
Structured literature scoping and gap identification
AI tools can rapidly synthesise large volumes of
housing policy reports, planning documents, and academic papers to help
students map the field, identify recurring themes, and spot under-researched
areas. When students then verify and critique these AI-generated summaries
against primary sources, they develop stronger information literacy and sharper
research questions.pmc.ncbi.nlm.nih+2
2.
Iterative drafting and language refinement
For non-native English speakers or time-constrained
part-time students, generative AI can suggest alternative phrasings, improve
sentence clarity, and help structure arguments without generating whole
sections. This supports learning by letting students focus on argument logic
and evidence while refining academic writing skills through comparison and
revision.ru+1
3.
Methodological scaffolding and protocol design
AI can help students draft interview guides, survey
instruments, or coding frameworks by generating examples aligned with stated
research aims. When students adapt these templates to their specific housing
context (e.g., public rental housing in Hong Kong), they learn methodological
design principles while avoiding the pitfalls of starting from scratch.pmc.ncbi.nlm.nih+1
4.
Reflective feedback and self-assessment
Students can use AI to simulate examiner questions,
generate counter-arguments to their thesis, or highlight logical gaps in their
draft chapters. Engaging critically with this feedback—accepting, rejecting, or
refining it—builds metacognitive skills and prepares students for viva or
defence scenarios.frontiersin
Ineffective,
harmful, or academically dishonest practices
1.
Outsourcing critical reading and synthesis
Using AI to summarise papers or generate literature
review content without engaging with the original texts undermines deep
learning and critical analysis. This “cognitive offloading” leaves students
unable to defend sources, evaluate methodologies, or identify biases—core
competencies in Housing Studies research.pmc.ncbi.nlm.nih+3
2.
Fabricated or unverified citations and data
Generative AI is prone to “hallucinating”
references, statistics, or case studies that appear plausible but are entirely
fictitious. Submitting such content without verification constitutes academic
misconduct and risks propagating misinformation in the scholarly record.pmc.ncbi.nlm.nih+4
3.
Ghostwriting entire chapters or sections
Prompting AI to write full dissertation
chapters—especially without disclosure—violates academic integrity policies and
deprives students of the learning that comes from constructing arguments,
analysing data, and iterating drafts. This practice also erodes the credibility
of the degree and the student’s own professional development.aisel.aisnet+2
4.
Undisclosed use and lack of transparency
Failing to declare AI assistance in methodology,
writing, or data processing breaches emerging academic norms and journal
policies. Transparency is essential for reproducibility, peer review, and
maintaining trust in research—particularly in policy-relevant fields like
housing studies where findings may inform public decisions.files.eric.ed+3
Key principle: Treat
generative AI as a “research assistant” that accelerates routine tasks and
expands possibilities, not as a substitute for your intellectual labour. Always
verify outputs, disclose usage per institutional guidelines, and ensure that
every AI-augmented step ultimately strengthens your own understanding and
scholarly contribution.
A note on the topic of “devil’s advocate” and its usage in doing an MBA dissertation project
Highlight 4 main ideas of "devil's advocate" (the topic) in Irving Janis's book "Groupthink: Psychological Studies of Policy Decisions and Fiascoes" and 2 claims of the topic in terms of Toulmin's model of arguments.
Four main ideas
In Groupthink: Psychological Studies of Policy Decisions and Fiascoes, Irving Janis presents the devil’s advocate as a structured method for preventing consensus-seeking from suppressing critical thought. The role is not simply to be argumentative; it is designed to improve the group’s examination of policy alternatives.
1.Institutionalised dissent A group should formally assign at least one member to act as a devil’s advocate when evaluating important alternatives. This makes disagreement part of the decision procedure rather than an act of personal disloyalty. Janis recommends that the advocate challenge the majority position in the manner of a lawyer testing the strength of testimony.muni
2.Critical examination of assumptions The devil’s advocate questions the assumptions underlying the preferred proposal, including its feasibility, likely effectiveness, risks, and ethical consequences. This helps reveal weaknesses that group members may overlook when they are committed to consensus.
3.Generation of alternatives and warning signals The role encourages members to consider competing courses of action, negative evidence, external threats, and possible failure scenarios. This directly addresses groupthink symptoms such as collective rationalisation, self-censorship, the illusion of unanimity, and the dismissal of warnings.muni
4.A safeguard against premature consensus Devil’s advocacy introduces controlled conflict before a final decision is made. The objective is not necessarily to defeat the preferred policy but to ensure that it can withstand serious criticism. Janis also recommends related safeguards, such as inviting outside experts, dividing the group into subgroups, and holding a “second-chance” meeting for residual doubts.muni
A useful qualification is that devil’s advocacy is most effective when the criticism is taken seriously. If the role is merely symbolic, or if the advocate is punished for dissent, it will not overcome groupthink. Later organisational research similarly describes the technique as deliberately introducing conflict to expose potential pitfalls, while recommending that the role be rotated so that one person does not become permanently associated with negativity.nationalforum
Two Toulmin-model claims
In Toulmin’s model, a claim is the conclusion an argument asks the audience to accept. The following are two possible claims about devil’s advocacy, with their supporting components.
Toulmin element
Claim 1: Devil’s advocacy improves group decision-making
Claim
Organisations should use devil’s advocacy when making high-stakes or uncertain decisions because it improves the quality of group reasoning.
Grounds/ data
Groupthink can cause members to suppress doubts, ignore warnings, rationalise previous decisions, and consider too few alternatives.
Warrant
If a person is required to challenge the preferred proposal, the group is more likely to test assumptions, identify risks, and examine alternatives.
Backing
Janis identifies critical evaluation and devil’s advocacy as safeguards against groupthink and defective policy decisions. muni
Qualifier
Devil’s advocacy will generally improve decision quality when the group treats the criticism as a genuine part of analysis.
Rebuttal
It may be ineffective if the role is superficial, overly personal, or used only to delay decisions.
Toulmin element
Claim 2: Devil’s advocacy should be a formal and rotating role
Toulmin element
Claim 2: Devil’s advocacy should be a formal and rotating role
Claim
The devil’s-advocate role should be formally assigned and rotated among members rather than left to voluntary dissent.
Grounds/ data
Cohesive groups may create pressure on dissenters, encourage self-censorship, and mistake silence for unanimity.
Warrant
A formal and rotating role distributes responsibility for criticism and reduces the likelihood that dissent will be interpreted as personal disloyalty.
Backing
Janis recommends assigning a member to challenge the majority position; organisational treatments of the method also recommend rotating the role. muni+1
Qualifier
This is especially important in cohesive groups facing crisis, uncertainty, or strong leadership pressure.
Rebuttal
Formal assignment alone is insufficient if the leader has already announced a preferred answer or does not genuinely accept criticism.
The two claims differ in emphasis: Claim 1 concerns the effect of devil’s advocacy on decision quality, whereas Claim 2 concerns how the technique should be implemented.
Discuss the usefulness and limitations of using generative AI to play the role of devil's advocate for the task of evaluating a new organizational strategy option for a company, notably in the context of a part-time MBA student doing a dissertation project.
Generative AI can be a useful first-round devil’s advocate for evaluating a new organisational strategy because it can rapidly challenge assumptions, generate alternatives, and simulate different stakeholder perspectives. However, it should support—not replace—managerial judgment, empirical research, and human accountability, particularly in an MBA dissertation.
Usefulness for strategy evaluation
1.It reduces confirmation bias A manager or student who has developed a preferred strategy may unconsciously search for supporting evidence. Asking an AI system to argue against the proposal can expose hidden assumptions about customers, competitors, costs, capabilities, regulation, and implementation. Research on AI-supported strategic decision-making suggests that LLMs can generate and evaluate strategic ideas and can be used to critique proposed strategies through a devil’s-advocate approach.pubsonline.informs
2.It produces criticism quickly and economically A human devil’s advocate may require considerable time to understand the proposal and develop counterarguments. Generative AI can provide an initial critique within minutes, allowing a part-time MBA student with limited time to conduct several rounds of challenge before interviewing managers or collecting survey data. Its low social inhibition may also allow it to raise criticisms that junior employees or insiders hesitate to express to senior decision-makers.pubsonline.informs
3.It broadens the range of perspectives The student can ask the AI to evaluate the strategy from the viewpoints of different stakeholders—for example, customers, employees, suppliers, competitors, regulators, investors, and local communities. It can also apply frameworks such as SWOT, PESTLE, Porter’s Five Forces, scenario analysis, the resource-based view, or stakeholder theory.
4.It supports structured argument analysis The AI can convert a strategy proposal into a Toulmin-style argument: claim, evidence, warrant, backing, qualifier, and rebuttal. It can then identify unsupported claims, weak causal links, missing evidence, and overly confident conclusions. This is particularly useful when refining a dissertation research model or preparing interview questions.
5.It encourages strategic exploration rather than immediate acceptance AI can propose alternative business models, implementation paths, risk-mitigation measures, and “pre-mortem” scenarios. Generative AI has been described as useful in the “doubt” phase of innovation, where teams revisit the assumptions underlying a proposed strategy rather than merely generating ideas.bcg
Main limitations
1.It may produce plausible but false information AI can invent market statistics, competitors, citations, regulations, or case examples. Its fluent language can make weak evidence appear authoritative. Therefore, every factual claim relevant to the strategy must be checked against reliable sources, company data, industry reports, or primary research.
2.Its evaluation can be inconsistent AI judgments may change depending on wording, the order in which options are presented, or the assumptions included in the prompt. Research on AI evaluation of strategic decisions reports inconsistency and bias in individual evaluations, although aggregating multiple AI evaluations may produce results closer to human expert judgments.papers.ssrn+1
3.It lacks the organisation’s tacit knowledge A model may not understand informal power relationships, employee morale, founder preferences, organisational culture, customer trust, political sensitivities, or the practical difficulty of implementing change. It can identify a theoretical risk but cannot reliably judge whether that risk is decisive in the particular company.
4.It can create artificial disagreement A devil’s advocate should challenge the proposal constructively. AI may generate generic objections simply because it has been instructed to oppose the strategy. This can result in “false balance,” where a weak objection receives the same attention as strong evidence against the proposal.
5.It does not assume responsibility for the decision Generative AI can compare options, but it cannot accept legal, ethical, financial, or managerial accountability. The final decision must remain with the company’s decision-makers, who must consider the organisation’s objectives, risk tolerance, values, and available resources.
6.Confidentiality is a significant concern A student should not upload confidential strategy documents, commercially sensitive financial information, identifiable employee or customer data, or unpublished interview material into an unauthorised public AI tool. For example, Hong Kong Metropolitan University guidance states that confidential, sensitive, or personal data should not be entered into generative AI tools, especially tools that are not institutionally supported.hkmu.edu
Relevance to a part-time MBA dissertation
For a part-time MBA student, AI is especially useful as a research assistant and reflective critic, not as a source of primary evidence. It can help to:
·identify assumptions in the proposed strategy;
·generate alternative explanations and rival hypotheses;
·draft a preliminary stakeholder or risk matrix;
·develop interview and survey questions;
·challenge the logic of a conceptual framework;
·organise themes after the student has independently analysed the data;
·identify gaps requiring verification through academic literature or company evidence.
However, the student should not treat an AI-generated critique as evidence that the strategy is viable or unviable. The dissertation’s conclusions should be based on verified secondary sources and appropriately collected primary data, such as interviews, surveys, financial analysis, or case-study evidence. AI output can suggest what to investigate; it cannot substitute for that investigation.
Recommended procedure
A disciplined process could be:
1.Define the strategy clearly State the strategic objective, target market, resources, assumptions, expected benefits, risks, timeframe, and success measures.
2.Ask AI to act as a structured devil’s advocate Request separate challenges concerning market demand, competitive response, financial viability, organisational capability, implementation, ethics, regulation, and stakeholder effects.
3.Require evidence and confidence levels Ask the AI to distinguish between verified facts, assumptions, inferences, and questions requiring further research. Do not accept references without checking the original sources.
4.Use multiple perspectives and prompts Compare critiques from a cautious CFO, frontline employee, customer, competitor, regulator, and sustainability stakeholder. This reduces dependence on one prompt or one framing of the issue.
5.Triangulate the criticisms Test important AI-generated objections against academic literature, company records, market data, expert interviews, and survey responses. A criticism that repeatedly appears across independent evidence deserves greater attention.
6.Record and disclose the AI use Keep the tool name and version, date of use, prompts, outputs, revisions, and decisions about which suggestions were accepted or rejected. Research guidance commonly emphasises critical human oversight, transparent acknowledgement, and retaining records of AI use.gla.ac+1
Overall assessment
Generative AI is valuable because it makes structured dissent faster, cheaper, and more accessible—an important advantage for a time-constrained part-time MBA student. Its appropriate role is to expand the range of questions and counterarguments; its inappropriate role is to act as the final evaluator, generate unverified evidence, or replace the student’s original analysis and primary research.
Harvard reference
Janis, I.L. (1982) Groupthink: Psychological studies of policy decisions and fiascoes. 2nd edn. Boston, MA: Houghton Mifflin.med.stanford+1
In-text citation
·Parenthetical: (Janis, 1982)
·Narrative: Janis (1982) argues that groupthink occurs when the desire for unanimity overrides the realistic evaluation of alternatives.
If you are using the first edition, published in 1972 under the title Victims of Groupthink, cite it as:
Janis, I.L. (1972) Victims of groupthink: A psychological study of foreign-policy decisions and fiascoes. Boston, MA: Houghton Mifflin.
A note on appropriate
and inappropriate employment of generative AI to support MBA students’ doing
dissertation projects
Using generative AI to augment part-time MBA students'
learning when doing their dissertation projects is a common practice. Suggest 4
ways of this practice as conductive to effective learning and suggest 4 ways
that are ineffective, harmful to learning and/ or academic dishonest behaviour.
Generative AI can
support effective learning in a part-time MBA dissertation when it acts as a learning
partner rather than a substitute for the student’s research, judgement and
writing. Its educational value is greatest when the student evaluates, verifies
and improves the output; uncritical reliance can weaken independent learning
and critical thinking.link.springer+1
Conductive to effective learning
Practice
How it supports learning
Appropriate student responsibility
1. Brainstorming and refining the research
problem
After developing an initial idea, the
student can ask AI to suggest alternative perspectives, theoretical lenses,
variables or research questions. This can expose overlooked possibilities and
help narrow a broad managerial problem into a feasible dissertation question.
Compare
suggestions with business literature, discuss them with the supervisor and
formulate the final research question independently.
2. Using AI as a Socratic tutor
The student can ask AI to explain difficult
concepts—such as institutional theory, thematic analysis, sampling, validity
or regression—in different ways, or to question their understanding. This is
particularly useful for part-time MBA students balancing employment, study
and family commitments.
Treat
explanations as provisional; verify definitions and methodological guidance
against textbooks, peer-reviewed sources and supervisor feedback.
3. Critiquing arguments and testing
assumptions
The student can provide their own proposed
framework, interview questions or interpretation of findings and ask AI to
identify logical gaps, rival explanations, possible bias or counterarguments.
This encourages reflection and analytical questioning rather than passive
acceptance.
Decide which
criticisms are valid and support revisions with evidence. AI should not make
the final theoretical or managerial judgement.
4. Supporting organization, analysis and
revision
AI can help create a project plan, organise
themes from researcher-analysed material, suggest coding structures, generate
draft code or improve clarity, grammar and coherence. Guidance for
dissertation projects identifies ideation, project planning, data-analysis
support, outlining and proofreading as possible uses, subject to oversight. blogs.qub.ac
Maintain control
of the research process, check code and interpretations, protect confidential
data, and disclose the tool’s use where required by the institution.
Ineffective, harmful or dishonest
Practice
Why it is problematic
1. Asking AI to produce the dissertation or
substantial sections
This replaces
the learning activity—reading, synthesising, reasoning and writing—with
outsourced text production. It may also breach academic-integrity rules if
AI-generated work is presented as the student’s own. Guidance for graduate
research commonly distinguishes permitted support, such as organising ideas
and correcting minor errors, from generating or rewriting substantive
portions of a dissertation. gradschool.fiu
2. Accepting fabricated or unverified
sources and claims
Generative AI
can produce plausible but nonexistent references, inaccurate quotations,
distorted article summaries and unsupported statistics. Including these in a
literature review or analysis can undermine the dissertation’s validity and
constitute research misconduct if the student knowingly submits them. AI outputs
should be checked against the original scholarly source. link.springer+1
3. Uploading confidential research or
business data
Submitting
identifiable interview transcripts, customer information, proprietary sales
data or company documents to an external AI service may breach privacy,
confidentiality, research-ethics or data-protection obligations. It can also
damage the trust of participants and the student’s organisation. Sensitive
material should be anonymised or kept out of the tool unless explicitly
authorised. gradschool.fiu
4. Using AI to generate research findings
or conceal the student’s contribution
Asking AI to
invent survey responses, manipulate results, create interview quotations,
select a preferred interpretation or write conclusions unsupported by the
data is academically dishonest and may amount to data fabrication or
falsification. Even less extreme overuse can reduce knowledge acquisition,
critical thinking and problem-solving ability when the student follows
ready-made answers instead of engaging with the research process. sciencedirect+2
Practical boundary
A useful rule is: the
student may delegate routine assistance, but not intellectual ownership.
The student should personally select and justify the research question,
evaluate the literature, make methodological decisions, analyse evidence, draw
conclusions and write the final argument.
The exact boundary
should also follow the MBA programme’s policy and supervisor’s instructions. A
transparent AI-use record can state the tool, date, purpose, prompts or
relevant outputs, how the output was verified, and what the student changed;
this supports accountability without treating AI as an academic source.
