Chapter 5—AI-Augmented Research &
Engineering Ethics
5.1 Advanced Discovery: Semantic Search with Elicit
Conventional literature searches depend heavily on keyword matching, which often limits
discovery. For instance, searching for “grid stability” may fail to retrieve relevant studies using
alternative terms such as “frequency resilience.”
5.1.1 Principle of Semantic Search
Semantic search overcomes this limitation by representing text using vector embeddings, which
capture meaning rather than exact wording.
In a high-dimensional vector space, conceptually related terms—such as power factor and
reactive power compensation—are located near each other, even when they share no identical
keywords. This enables broader and more accurate research discovery.
5.1.2 Capabilities of Elicit (2026 Edition)
Modern tools such as Elicit provide advanced research support:
Contextual Extraction: Retrieves specific technical data (e.g., inverter efficiency
values) across large collections of academic PDFs.
Evidence Synthesis: Aggregates findings from hundreds to thousands of papers,
highlighting consensus, contradictions, and research gaps.
Source Verification: Links outputs directly to the exact sentences in original documents,
enabling rapid validation.
5.2. Technical Execution: AI-Powered Code Debugging
In engineering design projects, software forms a critical backbone. AI tools have evolved from
simple code completion systems into context-aware reasoning agents.
5.2.1 State-of-the-Art Tools
Examples include Cursor and Claude Code, which offer:
Repository-Level Awareness: Analyze entire projects and detect how changes in one
module (e.g., signal processing) may affect others (e.g., visualization or control modules).
Cross-File Dependency Tracking: Identify hidden interactions and potential system
failures.
1
5.2.2 Emerging Concept: Topological Verification
A key 2026 trend is topological verification, where AI evaluates code against a structural “map”
of the system architecture. This ensures that implementation aligns with intended mathematical
and logical design principles, reducing systemic errors.
5.3. Research Integrity: Hallucinations and Ethics
A hallucination refers to a confident but factually incorrect AI-generated output. In engineering
contexts, such errors—especially involving safety standards or design parameters—can lead to
serious consequences.
5.3.1 Verification Strategies
Chain-of-Verification (CoVe):
The AI generates an initial response, formulates its own verification questions, answers
them independently, and refines the original output. This improves reliability but does not
replace human validation.
Mandatory Transparency:
Leading publishers such as IEEE and Nature require clear disclosure of AI usage.
AI systems cannot be credited as authors because they lack accountability and cannot
assume ethical or legal responsibility.
5.4. Student Protocol: Responsible AI Usage
Effective use of AI requires discipline and professional judgment. The following guidelines
should be applied in theses, lab reports, and design projects.
✅ Best Practices (“Do” List)
Use AI for Structural Planning:
Generate outlines to ensure inclusion of essential sections such as assumptions,
constraints, and methodology.
Apply Inversion-Based Fact-Checking:
Challenge AI outputs by asking for possible errors or counterarguments to validate
robustness.
Leverage AI for Formatting Tasks:
Utilize AI for LaTeX formatting, table generation, and citation styling (e.g., IEEE
format).
Use AI as a Debugging Partner:
Explaining your logic to AI can clarify your own thinking and reveal hidden flaws.
2
✅ Critical Limitations (“Not-To-Do” List)
Do Not Share Proprietary Data:
Uploading unpublished research or unique designs to public AI systems may constitute
disclosure and jeopardize patent eligibility.
Do Not Trust Citations Blindly:
Always verify AI-suggested references by accessing and reviewing the original source
documents.
Do Not Delegate Final Judgement:
AI can identify patterns but cannot fully understand real-world engineering constraints.
Final decisions must remain with the engineer.
Do Not Attribute Authorship to AI:
Listing AI as a co-author violates established academic and publishing standards.
Key Takeaway
AI should be regarded as a research assistant, not a decision-maker.
As engineers and researchers, you retain full responsibility for:
The correctness of your code
The validity of your data
The integrity of your conclusions
Every equation, assumption, and citation must ultimately be verified and defended by you.