人工智能语境下的设计思维:历史比较与认知重构OA

Since Herbert Simon's seminal work on the design of artificial objects,design thinking has evolved as a core concept in design theory for five decades.In the early 21st century,Stanford's D.school packaged design thinking as a universal innovation methodology,sparking global interest.However,the essence of design thinking remains contested:is it a unique cognitive mode exclusive to designers,or a universally applicable problem-solving framework?With the rapid advancement of artificial intelligence,particularly generative AI technologies,design thinking faces unprecedented transformation and reconstruction.The emergence of AIGC technologies has created an urgent need for new theoretical frameworks to guide design education and practice,while the popularization of design thinking risks weakening designer agency and blurring professional boundaries. This research conducts a cross-disciplinary historical comparative study to examine how AI intervention impacts design thinking,providing reconstructive perspectives on both the ontology and epistemology of design thinking.Three influential traditional frameworks are analyzed:Kees Dorst's"problem space and solution space"theory,Peter Kroes'"function and structure"theory,and John Gero's FBS ontological framework focusing on transformations among functional,behavioral,and structural elements. While these traditional frameworks offer valuable insights,they face significant ontological challenges in the AI era.First,design cognition's complexity is deeply influenced by contextual factors that are difficult to separate from specific situations,yet traditional models oversimplify contextual considerations and fail to adequately reflect the complexity and dynamism of AI-era contexts.Second,design experience exists predominantly as tacit knowledge,naturally resisting formal ontological representation,making design activities difficult to fully reduce to computational logic,particularly regarding subjective intentions and value judgments.Third,traditional design thinking models implicitly assume humans as the sole cognitive agents,struggling to accommodate AI's emerging role as a collaborative cognitive partner. Recent research demonstrates that AI systems are no longer simple tools but"hybrid intelligent agents"possessing degrees of agency and autonomy.This generative AI design approach produces"hybrid reality"phenomena,where design outcomes result from mixed intelligence creation,fundamentally disrupting the traditional subject-object dualistic cognitive framework.In AI-participated design processes,human designers and AI systems form role-based collaborative creative relationships,transforming not only how design knowledge is produced but also reconstructing designers'cognitive logic and decision-making patterns. The research examines how design thinking's knowledge representation and generation mechanisms transform from individual tacit knowledge to systemic explicit knowledge,and from individual cognition to hybrid intelligence.Traditional design thinking relied heavily on designers'personal experience and tacit knowledge—Polanyi's concept of"we know more than we can tell"—which constituted design thinking's unique resource while creating barriers for AI understanding.AI intervention provides new possibilities through deep learning technologies that unify different knowledge types(logical,practical,experiential)within high-dimensional semantic spaces,bridging the gap between explicit and tacit knowledge. Design reasoning similarly transitions from individual cognition to hybrid intelligence.While traditional design practice relied on complex reasoning abilities—deduction,induction,analogy,and abduction—intertwined in ways considered uniquely designer-specific,emerging large language models demonstrate similar reasoning capabilities.Through appropriate prompting strategies,these models can simulate"chain-of-thought"reasoning processes,showing clear logic in complex problem-solving that parallels iterative design thinking processes.This development suggests design reasoning should no longer be viewed as an internal process of a single cognitive agent but as hybrid intelligent activity distributed across human-machine interaction networks. To address these theoretical challenges,this research proposes the(PBS-VC)design thinking model for human-AI collaborative environments.The model's foundation consists of three interconnected dimensions:Problem space(P),Behavior process(B),and Solution generation(S),drawing from design cognition's theoretical traditions while incorporating vector space representation for new knowledge encoding.Two critical dimensions—Value judgment(V)and Context factors(C)—are integrated to capture emergent cognitive characteristics in human-AI collaborative design. The PBS-VC model employs vector space representation for each cognitive dimension,encoding design knowledge through feature compression and mapping into multidimensional vector spaces.Problem space encompasses not only explicit functional requirements and technical constraints but also implicit emotional expectations,cultural backgrounds,and aesthetic aspirations.The Behavior dimension distinguishes between expected behavior(designers'predictions of solution performance)and actual behavior(design outcomes'real-world performance),with alignment between them determining design success.The Solution space represents concrete design outcomes as multidimensional vectors spanning from formal elegance to functional completeness. The Value dimension,derived from Keeney's"value-focused thinking"theory,pervades the design process,connecting problem definition,behavior prediction,and solution generation.In human-AI collaboration,value judgment transforms from an implicit process to a core issue requiring explicit handling—AI systems can generate numerous solutions but lack value intuition,necessitating explicit value judgment;When AI becomes a design collaborator rather than a tool,value alignment becomes a critical challenge.The Context dimension,building on Suchman's situated action theory,emphasizes design cognition's deep embedding in specific environments,transforming from external conditions to internal components of the cognitive system. The research validates the PBS-VC model through the MPGPT system,a cross-modal intelligent agent developed on Tencent's Yuanqi framework.The system addresses core challenges in current AIGC prompt engineering—semantic gaps between user intentions and generated results,insufficient tacit knowledge representation,and ambiguous value orientation.Testing with 60 design tasks from the LAION-5B dataset across various industries and complexity levels demonstrated that system-generated prompts accurately captured nuanced design intentions,particularly excelling in complex tasks requiring balanced multiple design objectives. This practical implementation proves the PBS-VC model's feasibility and effectiveness,offering a comprehensive theoretical framework for design knowledge representation in human-AI collaboration.The model not only formalizes design thinking processes but preserves design thinking's inherent complexity and creativity,potentially serving as a bridge for deeper research into design cognition and artificial intelligence integration.