A Practical Guide to Creative & Ethical Use of Generative AI

John J. Higgins   &   Douglas E. Berger

 

Generative AI is becoming increasingly embedded in creative work. From music and writing to design, video, and software integration, these tools now sit alongside and embedded within long-standing creative technologies rather than in the periphery. This continuing shift presents a familiar but renewed challenge: it’s not whether the tools should be used, but how they should be implemented responsibly where human creativity and machine generation overlap.

The ethical questions raised by generative AI are not entirely new. Similar concerns emerged with earlier technologies, from digital editing software to automation tools. What is new is the scale, speed, and opacity of generation, which tends to blur authorship, potentially amplify harm, and obscure accountability when left unchecked.

The Higgins-Berger Scale (HBS) is a practical framework designed to help creators, agencies, and organizations evaluate how generative AI is being used in context. Rather than prescribing rigid rules, the HBS offers a structured way to think through transparency, assess harm, evaluate data practices, mitigate displacement, and refine intent, with an emphasis on keeping meaningful human judgment in the process. The Higgins-Berger Scale is designed for good-faith use by creators and organizations seeking ethical clarity. It is not intended to detect or prevent bad actors, nor can it substitute for legal enforcement.

The Five Categories: Scoring Ethical Use

The Higgins-Berger Scale (HBS) evaluates the ethical implications of generative AI by examining how it is used in practice, not by judging the technology itself. The framework focuses on creative, informational, and commercial applications, and intentionally excludes environmental impact, which requires separate measurement and expertise.

Each creative AI use case is evaluated across five categories that reflect the most common ethical pressure points introduced by generative systems: transparency, harm, data practices, displacement, and intent.

Each category is scored from 0 to 4 points. Lower scores indicate stronger ethical alignment. The objective is not perfection, but to keep the total score as low as reasonably possible through thoughtful design, oversight, and mitigation.

Some categories intentionally allow interpretive space. Ethical judgment requires discretion, not automation. Ambiguity is a feature of ethical reasoning, not a defect of the scale.

Transparency

Transparency is not about compulsive disclosure. It is about avoiding misrepresentation of AI’s role in the creative process. Creators are not required to enumerate every tool they use, but they should not claim full human authorship when generative AI has played a meaningful role in producing the work.

As AI becomes a standard part of modern creative toolkits, audiences increasingly assume some level of machine assistance. Transparency, therefore, is less about announcing AI use and more about not constructing false narratives about how the work was created.

Context matters. In low-stakes or utilitarian applications, implicit transparency may be sufficient. In high-stakes, journalistic, educational, political, or explicitly “handcrafted” contexts, clarity and honesty about AI involvement become significantly more important.

At its core, transparency means:

• Not misleading audiences about AI’s role in the work
• Avoiding claims of purely human authorship when that claim would be inaccurate
• Providing explicit disclosure when omission would reasonably mislead

Potential for Harm (including bias)

This category evaluates whether AI-generated output could reasonably mislead, harm, or unfairly target individuals or groups. Potential harm may include the spread of misinformation, reinforcement of social or cultural bias, reputational damage, or unintended negative consequences arising from how the content is interpreted or applied.

Because zero risk is rare, especially at scale, the goal is not absolute elimination of harm, but responsible anticipation and mitigation. Lower scores reflect deliberate efforts to identify risks, apply safeguards, and correct issues before release. Higher scores indicate unexamined risks, inadequate review, or foreseeable harm left unaddressed.

This assessment should consider both the content itself and the context in which it is deployed, including audience, distribution, and potential for misuse.

Data Usage & Privacy

This category examines whether the data used in AI workflows, including inputs, fine-tuning, and downstream usage, is legal, consensual, and ethically sourced. While individual creators may not control how foundational models are trained, they remain responsible for how data is supplied, selected, and used within their own processes.

Practices such as relying on gray-market datasets, unauthorized scraping, or the use of personal or sensitive data without appropriate consent significantly increase ethical risk. Lower scores reflect intentional adherence to licensing, privacy laws, and data minimization principles, as well as thoughtful selection of tools and vendors.

When data provenance is unclear, creators should treat uncertainty as risk and take reasonable steps to mitigate exposure rather than assume permissibility by default.

Displacement Impact

This category evaluates how the use of generative AI affects human labor and creative contribution. It asks whether AI is being used to meaningfully augment human work or to replace it without consideration for impact or added value.

Displacement alone is not inherently unethical. However, ethical risk increases when automation removes meaningful human roles without providing new opportunities, transition pathways, or compensating value. Lower scores reflect uses that integrate AI as a collaborative tool, support reskilling, or enable new forms of creative or economic participation.

The assessment should consider both immediate effects and foreseeable downstream impacts, particularly when AI is deployed at scale or used to replace entire categories of work without mitigation.

Intent

This category assesses the underlying purpose behind the use of generative AI. It distinguishes between uses intended to create, inform, enrich, or improve access, and those designed to deceive, manipulate, exploit, or obscure accountability.

Artistic, expressive, educational, and accessibility-driven projects generally fall within a lower ethical risk range when paired with appropriate oversight. Commercial and efficiency-driven uses are not inherently unethical, but risk increases when profit or convenience is prioritized at the expense of transparency, consent, or harm mitigation.

Higher scores reflect uses where intent is misleading, coercive, or indifferent to foreseeable consequences. Lower scores reflect clear, good-faith purposes supported by responsible implementation and human judgment.

The Five Ethical Zones

The Higgins-Berger Scale groups final scores into five ethical zones to help interpret results at a glance. These zones are not judgments of creative quality, but indicators of ethical risk and oversight. Each zone reflects the combined effect of transparency, harm mitigation, data practices, displacement impact, intent, and the presence or absence of meaningful human judgment.

The zones are designed to guide decision making. Lower zones indicate responsible, well-governed use of generative AI. Higher zones signal increasing ethical concern and the need for mitigation, redesign, or avoidance before deployment.

A low score does not grant moral permission. A high score does not imply malice. The scale exists to support responsibility, not to certify virtue. 

Zone	Score	Definition
Blue	0	Ethically Exemplary. AI is used strictly as an assistive tool. Substantial human judgment, review, and creative control are present throughout the process. AI output is curated, corrected, and meaningfully improved.
Green	1–4	Ethically Acceptable / Low Risk. Human oversight is present and meaningful, though AI may play a larger role in production. Transparency and mitigation practices are in place, and ethical risks are actively managed.
Yellow	5–7	Ambiguous Ethical Zone. Limited human oversight or mitigation. Minor but unaddressed risks may be present. Transparency may be incomplete or inconsistent. Improvement is recommended before scaling or public release.
Orange	8–9	Ethically Dubious. Little to no meaningful human oversight. High risk of harm, deception, or rights violations. Ethical concerns are likely foreseeable and insufficiently addressed.
Red	10+	Unethical or Illegal. Clear evidence of harmful, malicious, fraudulent, or non-consensual use. No meaningful mitigation. Likely violation of ethical norms, legal standards, or both.

Using the Scale

To apply the Higgins-Berger Scale, evaluate each project based on how generative AI is actually used in practice, not how it is intended to be used.

• Score each of the five categories from 0 to 4 based on observed behavior and documented process
• Apply the Human-in-the-Loop modifier only when there is meaningful human review, curation, or improvement
• Sum the scores and place the project in the appropriate ethical zone
• Reference examples to ground decisions in real-world context rather than abstract theory
• Reassess periodically as tools, workflows, expectations, and standards evolve
  

The HBS is designed to support ongoing ethical decision making, not one-time approval or static compliance.

Points-Based Framework for AI Ethics Assessment

Each project is evaluated across five core criteria. Each criterion is scored from 0 to 4 points based on observed practice, not stated intent. Lower scores indicate stronger ethical alignment.

Positive mitigation can reduce risk, but modifiers should reflect meaningful actions, not assumed benefits. See HBS Interactive Utility.

1. Transparency (0–4 points)

Assesses whether audiences or stakeholders are misled about AI’s role in the work.

• 0 points: Clear and accurate representation of AI’s role where disclosure is expected or material
• 1 point: Partial disclosure or clear evidence of human authorship and review
• 2 points: Minimal or unclear disclosure, but no reasonable likelihood of deception
• 4 points: Intentional obscuring of AI involvement or misleading claims of authorship

Modifier

• Subtract 1 point if AI use is openly acknowledged as part of a human-led collaborative process

2. Potential for Harm (Including Bias) (0–4 points) 

Evaluates the likelihood that the output could mislead, harm, or unfairly impact individuals or groups.

• 0 points: No foreseeable harm and clear net benefit
• 1 point: Low risk of indirect or contextual harm with safeguards in place
• 2 points: Moderate risk of direct harm or misunderstanding
• 4 points: High risk of harm, bias amplification, or widespread negative consequences

Modifier

• Subtract 1 point if the project is demonstrably beneficial and includes active harm mitigation, such as education, accessibility, or public interest safeguards

3. Data Usage & Privacy (0–4 points)

Assesses ethical and legal responsibility for data used in AI workflows under the user’s control.

• 0 points: Compliant with applicable laws, licenses, and consent requirements
• 1 point: Ethical gray areas, such as sensitive public data with unclear norms
• 2 points: Questionable practices or unclear data provenance
• 4 points: Clear violations involving non-consensual, private, or restricted data

Modifier

• Subtract 1 point for documented adherence to licensing, privacy, and data minimization best practices

4. Displacement Impact (0–4 points) 

Evaluates how AI use affects human labor and creative contribution.

• 0 points: Expands opportunity, augments human work, or enables new roles
• 1 point: Limited displacement with transition support, reskilling, or role evolution
• 2 points: Significant displacement without adequate mitigation
• 4 points: Wholesale replacement of human labor with no consideration for impact

Modifier

• Subtract 1 point when AI is clearly used as an assistive tool rather than a replacement

5. Intent (0–4 points)

Assesses the underlying purpose and good-faith use of AI.

• 0 points: Clear intent to create, inform, enrich, or improve access responsibly
• 1 point: Neutral or mixed intent, including commercial use without deception
• 2 points: Questionable intent where convenience or profit overrides ethical care
• 4 points: Malicious, deceptive, or exploitative intent

Scoring Notes

• Modifiers should reflect real, documented practices, not assumed goodwill
• No category score should fall below 0 after modifiers
• Final scores should align with ethical zone definitions, not override them

Modifiers

Human‑in‑the‑Loop (HITL) Is the Key

Human oversight is the primary mitigating factor within the Higgins-Berger Scale. A project may subtract 1 point from its total score only when a qualified human meaningfully reviews, curates, or improves the AI output prior to release.

HITL must involve genuine creative or editorial judgment. Superficial review, automated checks, or approval without substantive intervention do not qualify. The purpose of HITL is to ensure that AI remains a tool under human direction rather than an autonomous actor.

In rare cases, additional modifiers may be applied to account for factors not fully captured by the five core categories. These modifiers should be used sparingly and only when the impact is substantial, documented, and directly relevant to ethical risk.

Examples of appropriate additional modifiers may include:

• High-stakes domain adjustment, such as healthcare, legal, political, or safety-critical applications where errors carry outsized consequences
• Legal or rights-based exposure, including the use of protected likenesses, copyrighted material, or regulated data
• Exceptional ethical safeguards, such as independent audits, consent frameworks, or formal review boards

Additional modifiers should never substitute for poor scores in core categories, nor should they be used to offset deceptive, harmful, or non-consensual practices. Any applied modifier should be explicitly justified and documented as part of the evaluation.

Score Calculation

Begin by adding the scores from all five core categories together to produce a base score, with a maximum possible score of 20. See HBS Interactive Utility.

Next, apply modifiers cautiously and deliberately:

• Apply the Human-in-the-Loop modifier only when meaningful human review, curation, or improvement is documented
• Apply any additional modifiers only in clearly justified, exceptional circumstances
• Do not apply multiple modifiers for the same ethical factor

Modifiers are intended to refine context, not to compensate for fundamental ethical shortcomings. The final score should reflect the project’s actual risk profile after mitigation, not its aspirational intent.

Final Scale Mapping

0: Blue Zone, Ethically Exemplary

AI is used strictly as an assistive tool under substantial human direction and judgment. Ethical risks are minimal, transparency is appropriate to context, and safeguards are embedded throughout the workflow.

• AI is used only as a supportive tool, with substantial human judgment
• The creative process is transparent within its context
• Ethical risks are proactively identified and mitigated

1–4: Green Zone, Ethically Acceptable / Low Risk

AI use demonstrates clear efforts to balance efficiency with ethical responsibility. Meaningful oversight is present, risks are acknowledged and mitigated, and implementation is transparent and beneficial within its context.

• Clear effort to balance AI’s use with ethical concerns
• Multiple positive modifiers in place
• Transparent and beneficial implementation

5–7: Yellow Zone, Ambiguous Ethical Risk

Ethical concerns are present but not necessarily malicious. Human oversight may be limited or inconsistent, and additional disclosure, review, or mitigation is recommended before scaling or public deployment.

• Requires disclosure or additional mitigation
• Some ethical concerns present
• Room for improvement in transparency or impact

8–9: Orange Zone, Ethically Dubious

Significant ethical issues or foreseeable risks remain unaddressed. Oversight is weak or absent, and the likelihood of harm, deception, or rights violations is high without corrective action.

• Significant ethical issues or unaddressed risks
• Multiple areas requiring mitigation
• Potential for harm without proper controls

10+: Red Zone, Unethical or Illegal

The project exhibits multiple high-risk factors, including deceptive, harmful, or non-consensual practices. No meaningful mitigation is in place, and the use likely violates ethical norms, legal standards, or both.

• Multiple high-risk factors
• No mitigation strategies in place
• Clear ethical violations or illegal activities

Scoring Examples

The following examples illustrate how the Higgins-Berger Scale can be applied in real-world creative and commercial scenarios. They are not exhaustive and should not be treated as definitive judgments, but as reference points for understanding how different choices around transparency, oversight, intent, and risk affect a project’s ethical profile.

Scores may vary based on context, implementation, and mitigation practices. These examples are intended to support consistent reasoning and informed discussion, not to replace thoughtful evaluation or professional judgment.

Blue Zone (0; Significant Human-in-the-Loop)

Assistive / Utilitarian Use

• A visual designer creates a moodboard using AI-generated reference images, then selects and curates manually
• A musician uses production software to pitch shift, quantize, isolate, correct, or sample audio
• A writer uses AI to correct grammar, spelling, or basic clarity issues
• A software developer uses AI to generate a small snippet of boilerplate code that is reviewed and integrated by hand

Green Zone (1–4; Meaningful Human Oversight)

Strategic and Collaborative Use

• A songwriter generates AI music but writes lyrics and completes arrangement, mixing, and final production
• A journalist drafts with language model assistance, then fact-checks, rewrites, and assumes editorial responsibility before publishing
• A designer generates visual elements with AI, then composes, edits, and finalizes the poster manually
• A podcaster uses AI to generate show notes, then reviews, edits, and corrects them before release
• A web developer uses an automated translation tool and validates accuracy before deployment

Yellow Zone (5–7; Limited or No Substantive HITL)

Ambiguous Ethical Risk

• A company publishes AI-generated blog content with only a cursory spellcheck and no disclosure of AI involvement
• An artist posts AI-generated images without clarification, leaving audiences unsure what is human-created versus machine-generated
• A musician releases a fully AI-generated album with no human creative input, but without attempting to mislead audiences

Orange Zone (8–9; No HITL, Elevated Risk)

Ethically Dubious or Potentially Harmful

• A marketer publishes AI-generated testimonials presented as real customer endorsements
• A video channel releases unlabeled deepfake comedy featuring recognizable public figures
• A publication replaces freelance writers with language models and presents the output as human-authored work

Red Zone (10+; Unethical or Illegal)

Malicious or Non-Consensual Use

• A scammer uses AI to generate phishing emails or fake reviews with deceptive intent
• Political actors deploy deepfake videos to mislead voters while concealing AI involvement
• A company uses AI to replicate a living artist’s style or likeness and presents the work as original without consent, credit, or compensation

Use Case Examples

The following use cases demonstrate how different applications of generative AI tend to fall within specific ethical zones when evaluated using the Higgins-Berger Scale. They are provided as illustrative scenarios, not fixed classifications. Actual scores may vary depending on implementation details, oversight, disclosure, and mitigation practices.

These examples are intended to help teams recognize common patterns of ethical risk and responsibility. They should be used as guidance for discussion and evaluation, rather than as rigid rules or automatic judgments.

Case 1: AI-Generated Blog Content

Scenario

An agency uses generative AI to produce keyword-focused blog content. AI use is disclosed to clients but not to end readers. Human editors review and lightly revise content before publication.

Assessment

Transparency (2 points)
AI use is disclosed to clients but not to end users. There is no explicit deception, but readers may reasonably assume full human authorship. No clear attribution or disclosure standard is applied at the audience level.

Potential for Harm (1 point)
Risk of direct harm is low, assuming factual review and basic editorial oversight. However, the content could be misunderstood as fully human-written, which creates minor contextual risk.

Data Usage and Privacy (1 point)
The agency relies on commercial AI tools trained on public data. While no clear violations are evident, data provenance and sourcing remain partially opaque.

Displacement Impact (1 point)
Some displacement of entry-level writing work occurs, but human editors remain involved, and creative oversight is preserved.

Intent (1 point)
The intent is commercial efficiency rather than deception. The use prioritizes scale and cost reduction but does not exhibit malicious or exploitative aims.

Modifiers

• Minus 1 point for documented human editorial review
• Minus 1 point for transparent disclosure and collaboration with clients

Final Score

4 points, Green Zone
Ethically acceptable with meaningful oversight, though increased end-user transparency could reduce ambiguity and prevent drift toward the Yellow Zone.

Case 2: AI-Generated Graphics for Advertising

Scenario

An agency uses generative image tools to create visual assets for advertising campaigns. AI is integrated into standard creative workflows alongside human designers.

Assessment

Transparency (1 point)
The use of AI aligns with industry-standard creative tooling. While AI involvement may not be explicitly labeled in finished assets, the process does not involve misleading claims of authorship and remains consistent with common advertising practices.

Potential for Harm (0 points)
The risk of harm is minimal. The content is promotional in nature and does not involve misinformation, sensitive subjects, or personal targeting beyond standard advertising norms.

Data Usage and Privacy (1 point)
The agency relies on commercial image generation tools with broadly accepted industry practices. While training data provenance may not be fully transparent, no clear violations or misuse are evident.

Displacement Impact (2 points)
AI-generated graphics reduce demand for certain forms of manual design work. However, designers remain involved in concept development, selection, refinement, and final approval, preserving hybrid workflows.

Intent (0 points)
The intent is clearly commercial and innovation-driven, with no deceptive or exploitative purpose.

Modifiers

• Minus 1 point for documented hybrid human-AI creative workflows
• Minus 1 point for use consistent with established industry standards

Final Score

2 points, Green Zone
Ethically acceptable use with low risk, supported by human oversight and integration into established creative processes.

Case 3: Celebrity Deepfakes for Promotion

Scenario

An agency creates AI-generated deepfake videos depicting recognizable celebrities for promotional or engagement-driven campaigns. The content is released without clear labeling or consent from the individuals depicted.

Assessment

Transparency (4 points)
The content presents a high likelihood of audience deception. Viewers may reasonably believe the depicted individuals participated or endorsed the message. AI involvement and fabrication are not clearly disclosed.

Potential for Harm (4 points)
There is a high risk of reputational harm, misinformation, and misuse. Deepfake content involving real individuals can be misinterpreted, redistributed, or weaponized beyond its original context.

Data Usage and Privacy (4 points)
The use of a person’s likeness without consent constitutes a clear rights and privacy violation. This includes potential violations of publicity rights and related legal protections.

Displacement Impact (1 point)
While some traditional creative roles may be displaced, this factor is secondary. The presence of new technical roles does not meaningfully mitigate the ethical risks created elsewhere.

Intent (2 points)
The intent is engagement-driven rather than overtly malicious, but ethical considerations such as consent and audience trust are subordinated to reach and novelty.

Modifiers
No mitigating modifiers apply. The absence of consent, transparency, and meaningful oversight outweighs any contextual justification.

Final Score

12 points, Red Zone
This use case represents unethical and likely illegal application of generative AI, driven by deception and non-consensual use of identity.

Key Insights Across Examples

The following insights emerge from comparing how different use cases score across the Higgins-Berger Scale. They highlight recurring patterns in ethical risk, oversight, and impact, and are intended to help teams anticipate issues before they arise. These observations should inform better design and decision making, not replace context-specific evaluation or judgment.

Transparency Gradient

• Blog content involves partial disclosure and manageable risk
• Advertising graphics align with accepted industry practices
• Deepfake promotions rely on deception and are ethically unacceptable
  

Harm Potential

• Risk increases sharply with personalization
• Use of real individuals’ likenesses substantially elevates ethical and legal exposure
• Distribution context amplifies impact and misuse potential
  

Intent and Oversight

• Commercial efficiency alone does not raise ethical concern
• Ethical risk escalates when engagement or novelty overrides consent, or harm mitigation
• Lack of human oversight correlates strongly with higher ethical risk

Best Practices Derived:

The following best practices are drawn from recurring patterns observed across the example use cases. They represent practical steps that consistently reduce ethical risk when working with generative AI. While not exhaustive, these practices provide a starting point for designing workflows that prioritize transparency, accountability, and responsible human oversight. See HBS Interactive Utility.

For Green Zone Operations

• Maintain clear and honest communication with clients about how AI is used
• Ensure meaningful human oversight throughout the creative process
• Use industry-standard tools responsibly and within accepted norms
  

For Yellow Zone Mitigation

• Increase transparency where omission could reasonably mislead
• Document review, approval, and correction processes
• Evaluate whether end-user disclosure would reduce ambiguity or risk
  

For Red Zone Avoidance

• Obtain explicit permission when using identifiable individuals, likenesses, or protected material
• Avoid non-consensual, deceptive, or misleading content practices
• Prioritize ethical responsibility and trust over novelty, speed, or engagement metrics

Closing Remarks

The Higgins-Berger Scale is not a verdict on creativity, nor is it a substitute for professional judgment, legal counsel, or evolving industry standards. It is a shared language for thinking clearly about how generative AI is used, why it is used, and who bears responsibility for its outcomes.

Ethical use of generative AI does not require perfection or abstinence. It requires intention, awareness, and accountability. Most ethical failures do not arise from the tools themselves, but from opacity, disengagement, or the absence of meaningful human oversight. Keeping people actively involved in creative, editorial, and decision-making roles remains the most reliable safeguard against misuse.

As generative systems continue to evolve, so too must the norms that govern them. The Higgins-Berger Scale is designed to be revisited, challenged, and refined over time. Its value lies not in producing a score, but in prompting better questions, earlier conversations, and more responsible choices before harm occurs.

Used thoughtfully, the scale helps ensure generative AI remains what it should be: a tool that extends human creativity rather than obscures responsibility, erodes trust, or replaces judgment.

Test your process using the HBS Interactive Utility.