💻 computer science

Intelligent Product Conceptual Design: A Comparative Review of Paradigms, Challenges, and Future Trends

This review of 247 studies on Intelligent Product Conceptual Design reveals that while the field is shifting toward hybrid AI architectures, persistent trade-offs between interpretability, scalability, and human agency necessitate a future research agenda focused on distributed design intelligence, standardized benchmarks, and deeper industrial integration.

Original authors: ALI Mohammed Adam Jamea

Published 2026-08-14

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: ALI Mohammed Adam Jamea

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are an architect trying to design a new kind of flying car. You don't have blueprints yet; you just have a wild idea: "It needs to be fast, quiet, and fit in a garage." For decades, humans have been the only ones with the magic to turn that vague idea into a working sketch. But recently, computers have started trying to help. At first, they were like strict librarians who only knew the rules written in a giant book (if the car has wings, it must be heavy). Then, they became like memory-keepers who looked at old drawings of cars to guess what a new one should look like. Now, they are like super-fast artists who can paint thousands of pictures in a second, learning from millions of examples to guess what a "cool" car looks like. This is the world of Intelligent Product Conceptual Design. It's the very first, messy, creative stage of making things, where we decide what to build before we decide how to build it. Everyone cares about this because the decisions made in this early stage lock in most of the cost and performance of the final product. If the computer gets the early idea wrong, the whole project fails, no matter how good the engineers are later.

So, what happens when you ask a computer to be a creative partner? A new study by Ali Mohammed Adam Jamea from Quanzhou University of Information Engineering dives deep into this question. The author didn't just guess; they acted like a detective, hunting down and reading 247 different research papers published between 1990 and 2026. They wanted to see how the "brain" of these design computers has changed over time and, more importantly, whether the new, flashy AI methods are actually better than the old ones.

The study found that the story of computer design isn't a simple movie where the new hero kicks the old villain out of the room. Instead, it's more like a band where every instrument plays a different song. The old-school "rule-followers" (Knowledge-Based Systems) are great at explaining why they chose something, but they get confused if you ask them to do something they haven't seen before. The "memory-keepers" (Case-Based Reasoning) are good at copying past successes but struggle to adapt them to new situations. The "evolutionary" computers are like nature itself, breeding thousands of weird ideas and keeping the best ones, but they can be slow and expensive to run. And the new "super-learners" (Machine Learning and Deep Learning) are incredibly fast and can generate millions of ideas, but they are often "black boxes"—they give you a result without telling you how they got there, and sometimes they invent things that look cool but are physically impossible to build.

The most surprising discovery? The new "super-learners" haven't completely replaced the old methods. In fact, the study suggests that the future isn't about one perfect AI taking over. It's about hybrid teams. Imagine a design team where a rule-follower checks the physics, a memory-keeper suggests a proven shape, a super-learner generates a wild new style, and a human designer holds the pen to make the final call. The paper argues that this mix is where the magic happens.

However, the author also sounds a cautionary note. While these computers are getting smarter, they are still missing some crucial tools. The study points out that there are very few "standardized tests" to see if a computer is actually better than a human designer. In fact, only about 5.7% of the studies they looked at actually compared their AI against a human. Without these tests, it's hard to know if the AI is truly helping or just making things up. Also, most of these smart systems are still stuck in the lab; they don't talk well to the real-world software engineers use to build products. The paper concludes that while we are moving toward a future where humans and AI collaborate like dance partners, we still need to build better bridges between the computer's wild ideas and the real world's strict rules. The goal isn't to replace the human designer, but to give them a super-powered sidekick that helps them explore more ideas, faster, without losing the human touch that makes a design truly great.

Technical Summary: Intelligent Product Conceptual Design (IPCD): A Comparative Review of Paradigms, Challenges, and Future Trends

Problem Statement
Intelligent Product Conceptual Design (IPCD) is transitioning from symbolic reasoning systems to data-driven, generative, and hybrid AI architectures. However, the field lacks a coherent understanding of how these paradigms redistribute design knowledge, reasoning, automation, and human agency. Existing literature is fragmented, often focusing on specific paradigms or applications, and lacks a unified comparative framework. Furthermore, rapid advancements in deep learning and generative AI since 2018 have rendered earlier reviews incomplete. A critical gap exists in the limited use of direct human-designer baseline comparisons (fewer than 6% of studies) and the absence of standardized benchmarking protocols, which limits the generalizability of reported performance gains and weakens cross-paradigm comparisons.

Methodology
This study employs a systematic literature review following a PRISMA-informed protocol.

Corpus: The review analyzed 247 peer-reviewed studies published between 1990 and 2026 (with quantitative trend analysis based on the 2018–2025 subset).
Selection: Papers were sourced from Web of Science, Scopus, and Google Scholar. Inclusion criteria required explicit focus on conceptual design (not detailed design), use of computational intelligence, and English language. Exclusion criteria removed purely theoretical papers, detailed design focus, and non-empirical studies.
Analysis Framework: A five-dimensional comparative framework was developed to evaluate paradigms across:
1. Knowledge Representation: How design knowledge is structured (e.g., rules, cases, embeddings).
2. Reasoning Mechanism: The computational process for solution generation (e.g., rule chaining, retrieval, neural inference).
3. Automation Level: The extent of human intervention required.
4. Scalability: Capability to manage increasing complexity and data volume.
5. Human–Designer Interaction: Transparency, explainability, and controllability.
Reliability: Two independent reviewers classified papers into paradigms (Knowledge-Based Systems, Case-Based Reasoning, Evolutionary/Generative, ML/DL, Hybrid), achieving substantial inter-rater reliability (Cohen's $\kappa$ = 0.87). A 3-item quality assessment scale was applied to all papers.

Key Contributions

Unified Comparative Framework: The paper introduces a five-dimensional analytical framework that avoids single-metric bias, enabling explicit comparison of how agency is distributed between human designers and computational systems across epistemic, computational, and socio-technical layers.
Paradigm Redistribution Argument: The review argues that IPCD evolution is not a linear progression toward full automation or the replacement of earlier paradigms. Instead, it represents a progressive redistribution of design intelligence across symbolic reasoning, experiential knowledge, optimization, statistical learning, and collaborative human–AI interaction.
Identification of Structural Trade-offs: The study identifies five structural tensions inherent to cross-paradigm analysis, highlighting that no single paradigm is universally superior; performance depends on the interplay between data availability, interpretability needs, and design complexity.
Research Agenda: A prioritized agenda is proposed, emphasizing open Function–Behavior–Structure (FBS) datasets, explainable generative methods, standardized human–AI benchmarking, and industrial workflow integration.

Key Results and Findings

Paradigm Shift: There is a statistically significant shift in the field. Machine Learning/Deep Learning (ML/DL) approaches rose from ~15% of the corpus in 2018 to dominate by 2025, while Knowledge-Based Systems (KBS) and Case-Based Reasoning (CBR) declined from 45% to 18%. Hybrid systems expanded from 8% to 34%.
Paradigm Characteristics:
- KBS: Offers explicit, interpretable reasoning but suffers from knowledge acquisition bottlenecks and poor scalability.
- CBR: Effective for ill-structured problems using experiential knowledge but struggles with adaptation knowledge and scalability in high-dimensional case bases.
- Evolutionary/Generative: Excels at exploring large, non-convex design spaces but is computationally expensive and lacks global optimality guarantees.
- ML/DL: Provides high scalability and generative capacity but lacks reliable explainability and physical-feasibility guarantees (the "hallucination" problem).
Hybrid Systems: While hybrid architectures are increasingly prominent and show reported performance gains, the review finds insufficient evidence to establish their general superiority over well-designed single-paradigm baselines. Fewer than half of hybrid studies reported statistical comparisons against non-hybrid baselines, and none met criteria for both statistical testing and independent replication.
Human Comparison Gap: Only 5.7% of reviewed studies included human-designer baseline comparisons. Evidence suggests ML/DL systems can match junior designers in concept diversity but often underperform experienced experts in feasibility assessment. LLM-based systems generate structurally diverse concepts but frequently produce physically infeasible solutions.
Reproducibility and Integration: Less than 12% of studies provided executable code or full datasets. Integration with commercial CAD/PLM systems remains limited (only 6.4% of system descriptions reported integration), hindering practical industrial adoption.
Sustainability and Ethics: The literature shows a near-complete absence of sustainability considerations (only 2 of 247 papers mentioned environmental impact) and lacks standardized regulatory compliance frameworks.

Significance and Claims
The paper claims that the field is moving toward distributed design intelligence, where human judgment, symbolic knowledge, generative search, and physically grounded reasoning perform complementary roles rather than competitive ones. The significance of this work lies in:

Organizing Fragmented Evidence: It synthesizes three decades of research into a comparative framework that supports future theory-building and benchmark development.
Defining the "State-of-the-Art": It clarifies that the current state-of-the-art is not a single dominant method but a landscape of specialized tools, with hybrid approaches showing promise but requiring rigorous validation.
Addressing the Benchmarking Void: By documenting the lack of standardized human–AI benchmarks, the paper proposes a minimum benchmarking protocol (involving standard design problems and metrics for novelty, feasibility, and preference) to enable fair, evidence-based progress.
Guiding Future Research: It prioritizes the creation of open, large-scale FBS-annotated datasets as the single most critical enabling condition for the field, arguing that without them, design-specific foundation models and standardized benchmarks will remain aspirational.

The author concludes that the future of IPCD will not be determined by increasingly autonomous algorithms alone, but by the effective integration of human creativity, symbolic knowledge, and computational intelligence within collaborative design ecosystems.

Technical Summary: Intelligent Product Conceptual Design (IPCD): A Comparative Review of Paradigms, Challenges, and Future Trends

More like this