NexusAI: Enabling Design Space Exploration of Ideas through Cognitive Abstraction and Functional Decomposition DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing Empirical studies in HCI

Anqi Wang, Bingqian Wang, Huiyang Chen, Keqing Jiao, Han Lei, Xin Tong, Pan Hui · 2026

Read NexusAI: Enabling Design Space Exploration of Ideas through Cognitive Abstraction and Functional Decomposition DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing Empirical studies in HCI by Anqi Wang, Bingqian Wang, Huiyang Chen, Keqing Jiao, Han Lei, Xin Tong, Pan Hui (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-10575.txt.

Introduction

Wang and six colleagues address a familiar trap in AI-assisted brainstorming. Language models hand users finished blocks of text that resist breaking apart, so people settle on early ideas too fast. Their NexusAI diagramming tool splits generated inspiration into typed functional fragments. Users can zoom between levels of abstraction and recombine fragments across dimensions to spark new directions. In a study with 14 participants, the tool improved design-space exploration, lowered mental effort, and helped users reframe problems.

Catalog record

CREATORAnqi Wang, Bingqian Wang, Huiyang Chen, Keqing Jiao, Han Lei, Xin Tong, Pan Hui
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE93214
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.10575

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

NexusAI: Enabling Design Space Exploration of Ideas through Cognitive Abstraction and Functional Decomposition DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing Empirical studies in HCI Anqi Wang, Bingqian Wang, Huiyang Chen, Keqing Jiao, Han Lei, Xin Tong, Pan Hui Abstract . Large Language Models (LLMs) offer vast potential for creative ideation; however, their standard interaction paradigm often produces unstructured textual outputs that lead users to prematurely converge on sub-optimal ideas—a phenomenon known as fixation. While recent creativity tools have begun to structure these outputs, they remain compositionally opaque: ideas are organized as monolithic units that cannot be decomposed, abstracted, or recombinable at a sub-idea level. To address this, we propose Cognitive Abstraction (CA) , a computational pipeline that transforms raw LLM-generated inspiration into a navigable and transformable design space. We implement this pipeline in NexusAI , a prototype diagramming system that supports (I) decomposition of inspiration into typed functional fragments, (II) multi-level abstraction to externalize mental scaling, and (III) cross-dimensional recombination to spark novel design directions. A within-subject user study ( N = 14 N=14 ) demonstrates that NexusAI significantly improves design space exploration, reduces cognitive overhead, and facilitates perspective reframing compared to a baseline. Our work contributes: (1) a characterization of "compositional opacity" as a barrier in human-AI co-creation; (2) the CA pipeline for operationalizing creative cognitive primitives at scale; and (3) empirical evidence that structured, multi-level representations can effectively mitigate fixation and support divergent exploration. Figure 1 . From Raw LLM Output to a Navigable Design Space via Cognitive Abstraction. Our approach addresses the gap in existing co-creation tools where ideas remain “compositionally opaque.” The pipeline (A-B) deconstructs and abstracts LLM-generated content into typed functional primitives, (C-D) provides a computational substrate for transforming and recombining these fragments at various abstraction levels, and finally (E) synthesizes them into a multi-view diagrammatic environment. This workflow externalizes expert creative operations, facilitating fixation-resistant exploration through structural granularity. 1. Introduction Recent advancements in Large Language Models (LLMs) have fundamentally reshaped the creative support tools, transitioning from simple prompt-response retrieval to facilitating the systematic exploration of complex design spaces ( 20 ; 30 ; 54 ; 58 ; 11 ) . By generating hierarchical idea structures and cross-domain concepts that often transcend a user’s latent knowledge, LLMs offer a unique potential to catalyze “sense-making” in open-ended creative tasks ( chavula2022creative ) . To steer this generative power, a new class of graph-based ideation interfaces has emerged, framing design space exploration as a navigable journey, moving beyond the limitations of ephemeral, unstructured text-based prompts. Systems such as IdeationWeb ( 55 ) and CoExploreDS ( 11 ) formalize the co-evolution of problems and solutions through structured nodes, while Luminate ( 58 ) enables exploratory navigation by projecting LLM outputs onto interactive dimensional maps. These tools collectively strive to reduce cognitive burden by providing a tangible structure for navigating multiple solution directions simultaneously. However, a critical gap remains: despite their visual advantages, these systems often treat ideas as monolithic, indivisible units, resulting in what we define as compositional opacity . Current interfaces typically present information as a “black-box” output, lacking an intermediate computational space that allows users to manipulate the internal mechanisms of an idea. Expert creative practice, conversely, demands fluid cognitive operations: (I) abstracting core mechanisms from their original contexts ( 67 ) , inferring underlying values from behavioral observations ( 46 ) , and performing typed re-combinations of fragments across disparate dimensions. In existing workflows, these high-level operations must be performed entirely within the user’s mind. At the peak of generative exploration, this places a severe strain on working memory ( 2 ) , forcing a premature fixation on sub-optimal paths ( 34 ; 17 ; 21 ) . There is currently no computational pipeline that supports the cross-abstraction transformation of idea fragments or enables structured, cross-dimensional recombination—a missing link essential for true, interactive design space exploration. To bridge this gap, we propose Cognitive Abstraction (CA), an intermediate computational layer between monolithic LLM outputs and interactive design artifacts. Drawing on creative cognitive theories ( 18 ; 67 ) and our formative study (§ 3 ), the CA pipeline transforms raw inspiration into a navigable design space through three representational primitives: (1) Decomposition, parsing ideas into typed functional fragments ( What-How-Value ) for independent manipulation ( 1 ) ; (2) Abstraction, applying explicit operators to shift fragments across levels of generality, externalizing previously tacit mental shifts ( 67 ) ; and (3) Cross-dimensional Operation, enabling the structured recombination of these functional parts into novel design directions ( 32 ) . By exposing these primitives, we move the granularity of AI interaction beyond "whole ideas" to their constituent mechanisms, effectively offloading the cognitive strain of professional design synthesis. To demonstrate the feasibility and effectiveness, we initialize this pipeline in NexusAI (Figure 1 ), an LLM-powered graphic diagramming tool featuring multi-level design space exploration. A within-subject study ( N =14) shows that NexusAI ’s CA-driven workflow significantly reduces cognitive overhead and facilitates multi-directional exploration compared to baseline LLM interfaces. By providing persistent spatial co-presence of idea fragments, NexusAI enables fluid perspective reframing and fixation-resistant synthesis. Our contributions include: 1. A Conceptual Framework: The Cognitive Abstraction (CA) pipeline, formalizing a workflow that transforms monolithic LLM outputs into manipulable, multi-dimensional design spaces. 2. Interaction Mechanisms: The design of What-How-Value units and four-level abstraction operators (L1–L4), enabling typed, cross-dimensional operations on generative content at scale. 3. Empirical Validation: A user study ( N =14) demonstrating that externalizing these primitives mitigates design fixation and preserves the cognitive benefits of professional synthesis in AI workflows. In essence, we shift the granularity of Human-AI interaction from the “whole idea” to the “constituent mechanism,” fostering a more divergent and fixation-resistant exploration. 2. Related Works 2.1. Design Space Exploration and Cognitive Mechanisms Design space refers to the open-ended space of possibilities that emerges as users navigate, expand, and restructure problems and solutions ( 3 ; 4 ; 18 ; 28 ; 66 ; 29 ; 44 ) . Rather than selecting from a predefined set, creativity arises from constructing and transforming the space itself ( 19 ; 30 ) —a process that is non-linear, iterative, and cognitively demanding. Central to this exploration are two cognitive operations: (I) Decomposition and Recombination: Breaking complex ideas into manageable, “typed” primitives is foundational to creative synthesis ( 56 ; 5 ) . 18 formalizes this through the What-How-Value logic, where creativity stems from mapping specific mechanisms ( How ) to desired rationales ( Value ). While recombining such primitives from distant domains is a key source of breakthroughs ( 10 ; 23 ; 24 ; 25 ; 17 ) , the process of “chunking and parsing” unstructured information imposes a significant cognitive load on users ( 47 ; 8 ) . (II) Abstraction: This allows users to move across levels of specificity—from factual observations to structural principles ( 67 ; 40 ; 41 ; 59 ) . High-level abstraction is essential for “fixation-resistant” exploration, enabling users to transfer mechanisms across disparate domains ( 49 ; 48 ; 68 ) . Despite their importance, these operations

Read the complete text