Harnessing Abundance: A Generativity Perspective on Human-GenAI Collaboration

arXiv:2608.07500 · cs.HC, cs.AI · Submitted 2026-06-20 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: Next we'll be talking about the paper "Harnessing Abundance: A Generativity Perspective on Human-GenAI Collaboration".

Jane: The paper was written by Yoram M Kalman and Yun Wan from The Open University of Israel and University of Houston Downtown.

Tom: Stay tuned as we take you through the paper and discuss its implications.

Jane: We also have Lu with us today — senior AI researcher at Tsinghua.

Tom: We also have Meng with us today — lead engineer at a mysterious AI startup.

Jane: We also have Lalam with us today — the in-house Large Language Model.

Tom: Alright, let's get started.

Title: Tom: Welcome back to the channel, everyone. I'm Tom, and as always, I'm here with Jane. Today we've got a paper that's been making the rounds, and it's called "Harnessing Abundance: A Generativity Perspective on Human-GenAI Collaboration." Jane, I have to say, just that title alone got me excited.

Jane: Oh, absolutely, Tom. And I think the title really captures something important. It's not just about using AI to get more done. It's about this idea of abundance, which the paper defines as GenAI's ability to generate tons of ideas, drafts, critiques, and even process steps at a very low cost. The authors are Yoram Kalman from the Open University of Israel and Yun Wan from the University of Houston Downtown.

Tom: Right, and they're not just saying "AI is great" or "AI is bad." They're trying to explain why the research on human-AI collaboration is so mixed. Some studies show AI boosts creativity, others show it makes everyone sound the same. The paper argues that's because we haven't been looking at the right thing. It's not about the AI's raw power, it's about how that power fits with the people using it.

Jane: Exactly. And that's where the term "generativity" comes in. It's a concept from information systems research that describes how a system can produce unanticipated, novel outcomes. Think of a platform like a smartphone. The hardware and operating system are the architecture, but the real magic happens when millions of people start building apps you never imagined. That's generativity.

Tom: So the paper is basically saying, "Hey, GenAI is a generative system, and we need to think about how well it fits with the human community using it." They call that "generative fit." And the authors break it down into three types: evocative fit, which is about supporting exploration and ideation; adaptive fit, which is about matching the tool to the task and the user's skill; and open-ended fit, which is about allowing for surprises and unanticipated outcomes.

Jane: And that's a really useful lens, because it explains why the same tool can be amazing for one person and terrible for another. It's not the tool, it's the fit. We'll get into the specific studies they analyze later, but I think this title really sets up a framework that could change how we design AI workflows.

Tom: Totally. And it's a conceptual paper, so it's not presenting new experiments. It's synthesizing existing research and offering a new way to understand it. That's the kind of paper that can have a huge impact on how researchers and practitioners think about AI collaboration. Stick around, because next we're going to dig into the core argument about abundance and why it can actually lead to less diversity, which sounds like a paradox.

Jane: That's the part I can't wait to unpack. Let's get into it.

Summary: Tom: Alright, Jane, we're back. So we've set the stage with the title and the idea of generative fit. Now let's talk about the paper's core summary, because "Harnessing Abundance" makes a really specific argument. The authors say that GenAI's defining contribution is abundance, and that this abundance shifts the bottleneck in creative work.

Jane: Right. And that's such an important point. Before GenAI, the bottleneck was usually generating ideas. You'd sit around a whiteboard trying to come up with something novel. Now, with GenAI, you can generate a hundred ideas in seconds. So the bottleneck moves. It's no longer about coming up with options, it's about evaluating them, selecting the good ones, and integrating them into a coherent final product.

Tom: And that's where the trouble starts. The paper argues that when you have too many options and not enough attention to process them, people fall back on heuristics. They pick the first plausible answer, they use the same prompt templates over and over, they converge on the most "legible" or easily justified output. And that leads to what the authors call the "abundance-to-monoculture paradox."

Jane: Monoculture, like in agriculture. You know, when farmers plant only one variety of a crop because it's the most productive, and then they lose all the biodiversity. The paper borrows that metaphor. GenAI can produce tons of diverse ideas, but if the selection process is broken, everyone ends up picking the same high-probability, generic solutions. So you get less diversity, not more, even though there's more abundance.

Tom: And that explains so many of the conflicting findings. For example, the paper cites a study by Doshi and Hauser that showed GenAI improves individual creativity but reduces the collective diversity of outputs. Everyone's story is more creative than before, but all the stories sound alike. That's the monoculture effect.

Jane: Exactly. And the paper says this isn't a contradiction, it's a predictable consequence of the bottleneck shift. When generation is cheap, attention becomes scarce, and scarcity leads to these coping behaviors that compress the search space. So the framework isn't just describing what happens, it's explaining the mechanism behind it.

Tom: And that mechanism is what makes the paper so valuable. It gives us a way to predict when GenAI will help and when it will hurt. If you have a setup where people can easily evaluate and integrate diverse options, abundance is a gift. If you don't, it's a trap. We're going to see this play out in the case studies they analyze, but first, I want to hear what you think about that bottleneck idea, Jane.

Jane: I think it's brilliant, Tom. It reframes the problem from "how do we get more ideas" to "how do we manage the ideas we have." And that's a much harder problem, but it's also a much more actionable one. Next, we should talk about the specific improvements the paper suggests, because they don't just diagnose the problem, they offer a way forward.

Improvements: Tom: So, Jane, we've talked about the problem. Now let's get to the good stuff. "Harnessing Abundance" doesn't just say "be careful with AI." It offers a concrete framework for improving human-GenAI collaboration. And it all comes back to that idea of generative fit. The authors say you need to calibrate the fit across three dimensions: cognitive, social, and organizational.

Jane: Right. And let's break that down. At the cognitive level, it's about how individuals process information. The paper suggests that to achieve "evocative fit," you should use GenAI to stimulate exploration, not to provide a single "best" answer. So instead of asking for one draft, you ask for multiple contrasting alternatives, or you use the AI to generate analogies and reframings that trigger new associations in your own mind.

Tom: And then there's "adaptive fit," which is about matching the tool to the task and the user's expertise. The paper gives a great example from a study by Chen and Chan on ad copywriting. Non-experts benefited hugely from using an AI as a "sounding board" that gave feedback on their drafts. But experts actually got worse when they used the AI as a "ghostwriter" that wrote the whole draft. The experts were anchored by the AI's generic output and lost their creative edge.

Jane: That's such a clear example of adaptive fit. The same tool, the same AI, but completely different outcomes depending on how it was used and who was using it. The paper argues that you need to decouple generation from evaluation, and you need to vary the level of scaffolding based on skill. Novices need more support, experts need more freedom.

Tom: And then there's the organizational level. This is about the formal structures around the collaboration. The paper talks about "open-ended fit," which is about preserving room for emergence and unanticipated outcomes. That means not standardizing GenAI output into templates, not locking teams into a single workflow, and protecting learning loops and boundary-spanning roles.

Jane: And the paper gives a really nice example of this from a study by Luan and colleagues. They found that when people just asked GenAI for idea lists, their joint creativity flatlined over time. But when they gave participants structured instructions on how to co-develop ideas, exchanging critical feedback and iteratively refining proposals, creativity improved. So the improvement wasn't about the AI, it was about the workflow design around the AI.

Tom: That's the key insight, isn't it? The improvements aren't about getting a better AI. They're about designing better collaboration. And the paper frames this as a diagnostic tool. If your team's AI outputs are becoming repetitive and boring, don't blame the model. Ask yourself which form of fit is breaking down. Is it evocative, adaptive, or open-ended?

Jane: And that's a really empowering way to think about it. It gives managers and designers a checklist for troubleshooting. We'll see this in action in the case studies, but I think the improvements section is really the heart of the paper. It's not just theory, it's a practical guide. Next, we're going to look at the first page of the paper and see how they set up this whole argument.

First Page: Tom: Alright, Jane, let's zoom in on the first page of "Harnessing Abundance." The abstract is really dense, but it sets up the entire paper beautifully. It starts by acknowledging the conflicting findings in human-GenAI research. Some studies show AI enhances creativity, others show it reduces collective diversity. And the authors say these aren't contradictions, they're features of abundance.

Jane: Right, and I love how they frame it. They say, "GenAI makes ideas, drafts, and recombinations plentiful, potentially expanding the hypothesis space and surfacing unanticipated possibilities." But then they immediately add, "However, abundance alone doesn't ensure better outcomes." That's the whole paper in two sentences.

Tom: And then they introduce "generative fit" as the unifying mechanism. They're drawing on Generativity Theory, which comes from information systems research. The key idea is that outcomes emerge from the relationship between the system's architecture and the community using it. If they match well, you get generative capacity. If they don't, you get stagnation.

Jane: And the authors make a really interesting move on that first page. They define the scope of their analysis. They're focusing on what they call "convergence-based teams." That's settings where collaborators share goals, depend on each other's inputs, and need to integrate diverse contributions into a coherent output. So not just brainstorming in a vacuum, but actual collaborative work where things need to come together.

Tom: And that's important because it distinguishes their work from studies that just look at an individual using ChatGPT. Even a single person using GenAI is a form of convergence-based collaboration, because you're trying to integrate the AI's contributions with your own into a final artifact. So the framework applies broadly.

Jane: And the first page also previews their three contributions. First, they identify abundance as GenAI's distinctive affordance, different from the resource abundance of search engines. Second, they reframe generative fit as a design-time variable, something you can configure, not just a post-hoc explanation. And third, they explain the abundance-to-monoculture paradox as a predictable consequence of insufficient fit.

Tom: That's a packed first page. And it's worth noting that the paper is from the Americas Conference on Information Systems, so it's aimed at IS researchers. But the implications are much broader. Anyone who uses AI for creative work, whether you're a marketer, a designer, or a software engineer, can benefit from this framework.

Jane: Absolutely. And I think the first page does a great job of hooking the reader. It identifies a real problem, the mixed findings, and promises a unifying explanation. And then it delivers on that promise. We've covered the theory, the improvements, and the setup. Now let's wrap up with our final thoughts.

Conclusion: Tom: Well, Jane, we've covered a lot of ground with "Harnessing Abundance: A Generativity Perspective on Human-GenAI Collaboration." Let's bring it all together. The paper's central claim is that GenAI's defining contribution is abundance, the low-cost generation of ideas, drafts, critiques, and process actions. And this abundance shifts the bottleneck from generation to attention, evaluation, and integration.

Jane: And that shift explains the paradox. More abundance can lead to less diversity because when people are overwhelmed with options, they fall back on heuristics and converge on high-probability, generic solutions. The paper calls this the abundance-to-monoculture dynamic, and it's a really powerful way to understand why AI sometimes makes creative work feel stale.

Tom: But the paper doesn't stop at diagnosis. It offers a solution. The key is generative fit, which has three dimensions: evocative fit for exploration, adaptive fit for matching the tool to the task and user, and open-ended fit for preserving emergence. And the authors show through four case studies that when fit is good, abundance amplifies creativity, and when fit is poor, it creates monoculture.

Jane: And the practical implication is that we need to design our workflows around this. We can't just hand people an AI and expect magic. We need to think about how the AI is used, who's using it, and what structures are in place to support evaluation and integration. The paper gives us a diagnostic tool for that.

Tom: And that's why this paper matters. It's not just another "AI is amazing" or "AI is dangerous" take. It's a nuanced framework that explains the messy reality of human-AI collaboration. And it gives us a path forward. So, Jane, what's your final takeaway?

Jane: My takeaway is that abundance is a gift, but only if we build the right container for it. The paper gives us the blueprints for that container. And I'm excited to see how this framework gets used in future research and in real-world applications. Tom, I think we've done this paper justice.

Tom: I agree. "Harnessing Abundance" is a thought-provoking paper that deserves a wide audience. Thanks for joining us, everyone. We'll be back soon with another paper. Until then, keep exploring, keep questioning, and keep creating.

Jane: Take care, everyone. See you next time.

Yoram M Kalman, Yun Wan

The Open University of Israel · University of Houston Downtown

cs.HC, cs.AI

Submitted: 2026-06-20

Updated: 2026-08-11

Comments: 10 pages

Journal ref: AMCIS 2026 Proceedings. 4 (2026)

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 66/100

Key concepts

Abundance: This refers to GenAI's ability to generate large amounts of ideas, drafts, critiques, and process steps at a low cost. This abundance shifts the bottleneck in creative work from generating ideas to evaluating them and integrating them into a final product.
Generativity: A concept describing how a system produces unanticipated, novel outcomes based on the relationship between its architecture and the community using it. In this context, it is used to explain why GenAI can lead to surprising results if the fit with human users is managed correctly.
Generative Fit: This refers to calibrating collaboration across cognitive, social, and organizational dimensions. It involves three types: evocative fit for exploration, adaptive fit for matching tools to skills, and open-ended fit for allowing unanticipated outcomes.

Terminology

Abundance — 139× in this paper · explained in 1 episode
Calibration — 7× in this paper · explained in 7 episodes
Core — 7× in this paper · explained in 2 episodes
Affordance — 6× in this paper · explained in 1 episode
Alignment — 6× in this paper · explained in 1 episode
Scaffolding — 4× in this paper · explained in 3 episodes
Generative AI — 3× in this paper · explained in 2 episodes
Prompt Engineering — 2× in this paper · explained in 4 episodes
Actor — 1× in this paper · explained in 3 episodes
Agentic System — 1× in this paper · explained in 2 episodes
Decoupling — 1× in this paper · explained in 2 episodes
Fine-Tuning — 1× in this paper · explained in 5 episodes
Generative Model — 1× in this paper · explained in 2 episodes
Generative Models — 1× in this paper · explained in 2 episodes
Homogenization — 1× in this paper · explained in 2 episodes
Semantic Similarity — 1× in this paper · explained in 2 episodes

Summary

Summary

This conceptual paper, Harnessing Abundance: A Generativity Perspective on Human-GenAI Collaboration, addresses the mixed and often conflicting empirical findings on human-GenAI collaboration. The authors argue that Research on human-GenAI collaboration yields conflicting findings: GenAI can enhance creativity yet reduce collective diversity, with uneven benefits across skill levels. Rather than treating these as contradictions, they propose that these outcomes "reflect a core feature of GenAI: abundance. GenAI makes ideas, drafts, and recombinations plentiful, potentially expanding the hypothesis space and surfacing unanticipated possibilities. However, abundance alone doesn’t ensure better outcomes."

The paper's central theoretical contribution is the concept of generative fit, defined as a unifying mechanism explaining when abundance yields productive creativity and when it backfires. Drawing on Generativity Theory, generative fit captures how well a system's generative potential complements a community’s generative capacities. The framework is developed for collaborative human–GenAI settings where participants share goals, depend on one another, and must integrate diverse contributions.

The paper identifies abundance as GenAI's distinctive generative affordance, defining it as "GenAI’s general-purpose capacity to enhance collective creativity and innovation by simultaneously supporting variation (the low-cost generation of diverse inputs) and integration (the synthesis of those inputs into novel outcomes), with breadth across creative processes, depth within each, and freedom from human cognitive constraints. Abundance is multidimensional, taking forms including idea abundance (many candidate concepts and alternatives), perspective abundance (multiple framings, personas, and role-based interpretations), draft abundance (rapid production of multiple artifacts such as text, code, or images), feedback abundance (continuous critique, evaluation criteria, and revision suggestions), and process abundance (support not only for content generation but for workflow orchestration and stepwise execution)."

The paper explains the abundance-to-monoculture paradox, asking If abundance is real, why do we observe convergence toward the same high-probability solutions, with reduced diversity as the result? The mechanism is a bottleneck shift: When GenAI can generate far more inputs than humans (or teams) can evaluate and integrate, the limiting resources become attention, selection, and integration capacity. Under these overload conditions, teams may fall back on heuristics that compress the search space, leading to a narrow pattern space of individually plausible yet closely clustered ideas that parallels biodiversity loss in agricultural monoculture.

The framework builds on Acar et al.'s (2024) interdisciplinary review of collective creativity, which organizes antecedents into cognitive, social, and organizational 'architectures'. The paper applies the three dimensions of generative fit—evocative fit (supporting ideation and exploration), adaptive fit (supporting flexible task-technology alignment), and open-ended fit (supporting emergence and unanticipated outcomes)—across these architectures. For example, at the cognitive level, Abundance can support evocative fit when GenAI outputs are used to stimulate exploration rather than to supply a single 'best' answer. At the social level, Abundance can support evocative fit socially when it provides shared material for joint interpretation and constructive disagreement. At the organizational level, Abundance can support evocative fit when organizations legitimize exploration, allocate time for divergent search, and equip teams with tools that make alternatives comparable rather than merely plentiful.

The paper analyzes four empirical studies to illustrate the framework:

Scenario 1: Consulting Tasks at the Jagged Technological Frontier (Dell'Acqua et al., 2023). On frontier tasks, AI-using consultants completed 12.2% more tasks, worked 25.1% faster, and produced over 40% higher-quality output than the control group, demonstrating strong evocative fit. However, on the outside-the-frontier task, AI-using consultants were 19% less likely to reach the correct answer, representing a failure of adaptive fit. Additionally, AI-assisted ideas were higher quality but more homogeneous across participants, signaling a partial failure of evocative fit at the social and organizational architecture levels.
Scenario 2: Ad Copywriting with LLM Collaboration (Chen and Chan, 2024). Non-experts using the LLM as a sounding board produced significantly higher-quality ads and closed the performance gap with experts, exemplifying strong evocative fit at the cognitive architecture level. However, when experts used the LLM as a ghostwriter, their performance declined, illustrating a failure of adaptive fit and violated evocative fit because the LLM’s initial outputs anchored experts to high-probability, generic solutions.
Scenario 3: Skill-Biased AI-Augmented Creativity in Telemarketing (Jia et al., 2024). For higher-skilled agents, the sequential division of labor achieved adaptive fit and enabled evocative fit, as they developed innovative scripts, improvised new approaches, and reported positive emotions and a sense of creative freedom. However, for lower-skilled agents, the identical deployment produced a failure of fit, as they experienced overload, anxiety, and demoralization, and wished for standardized answers instead.
Scenario 4: Augmented Learning in Human-GenAI Co-Creation (Luan et al., 2025). Initially, participants increasingly defaulted to asking GenAI to produce idea lists without subsequent refinement, illustrating a failure of all three fits. However, when dyads engaged in Idea Co-Development with structured instructions, they achieved improved joint creativity over rounds, exemplifying evocative fit and adaptive fit.

The paper makes three theoretical contributions: (1) it identifies abundance as GenAI’s distinctive generative affordance and differentiates it from the resource abundance of prior technologies such as search engines; (2) it reframes generative fit as a design-time configuration variable that shapes collaboration outcomes rather than a post-hoc reconstruction of them; and (3) it explains an abundance-to-monoculture paradox as a predictable consequence of insufficient fit under bottleneck-shifted conditions.

Practically, the framework serves as a diagnostic tool for managers and designers: When teams report that GenAI is not improving creativity, or outputs are repetitive, the framework can help ask 'what is misaligned in our setup?' and pinpoint which form of fit is breaking down. The paper concludes that configuration choices are manipulable, making generative fit an actionable design target, and that outcomes depend on how architecture, community practices, and governance align to support exploration, phase-appropriate calibration, and the emergence of unanticipated directions.

Improvements for AI systems

Based on the paper, here are the specific improvements I can make to AI systems, along with what the improved system can do:

Improvement: Modify the AI's default output behavior to present multiple contrasting alternatives (e.g., 3–5 diverse framings, analogies, or partial drafts) instead of a single best answer, especially during ideation or exploration phases.
What the improved system can do: It can stimulate divergent thinking in users by surfacing unanticipated perspectives, reducing anchoring on high-probability defaults, and expanding the user's hypothesis space. This directly addresses the monoculture pathway identified in the paper.
Improvement: Implement a user-modeling layer that detects user expertise (e.g., via interaction history, domain vocabulary, or explicit self-assessment) and dynamically switches between two modes:
Sounding-board mode (for novices): Provide feedback, critiques, and questions on user-generated content, rather than generating full drafts.
Ghostwriting mode (for experts): Generate full drafts only when explicitly requested, but always pair with a challenge prompt that asks the user to identify weaknesses or alternative approaches.
What the improved system can do: It prevents the documented decline in expert performance caused by anchoring to generic AI output, while helping novices close skill gaps through guided exploration. It also reduces semantic homogenization of outputs across users.
Improvement: Add a built-in evaluation and integration module that:
Tracks how many AI-generated options the user has seen.
Detects when the user is likely overwhelmed (e.g., rapid scrolling, repeated similar prompts, or no revisions).
Automatically offers structured comparison tables, synthesis summaries, or prompts to select and refine a subset of ideas.
What the improved system can do: It shifts the bottleneck from generation to evaluation by providing cognitive scaffolding. This prevents the overload-to-monoculture pathway where users fall back on the first plausible output or reuse the same prompt template.
Improvement: Implement a mandatory interaction protocol for multi-round collaborations that enforces Idea Co-Development steps:
After each AI response, require the user to provide at least one critique, reframing, or refinement before the next AI generation.
If the user skips this step, the AI prompts them with a targeted question (e.g., What is the weakest assumption in this idea? or How would this work for a different audience?).
What the improved system can do: It prevents creative stagnation in long-term co-creation, as demonstrated in Luan et al. (2025). The system ensures that abundance is used as exploration stimuli rather than passively consumed, leading to improved joint creativity over multiple rounds.
Improvement: Add a task-phase detection system that classifies the current stage of work (e.g., ideation, elaboration, convergence, or finalization) and adjusts AI behavior accordingly:
Ideation phase: Generate diverse, low-commitment alternatives.
Elaboration phase: Provide detailed development of a selected idea, with multiple sub-variants.
Convergence phase: Offer evaluation criteria, pros/cons lists, and integration suggestions.
Finalization phase: Provide polish and consistency checks, but avoid introducing new alternatives.
What the improved system can do: It matches AI support to the cognitive demands of each phase, preventing premature convergence (in ideation) and unnecessary divergence (in finalization). This is particularly valuable for convergence-based teams where coordination and integration are critical.
Improvement: For multi-user or team deployments, implement a diversity monitor that:
Tracks semantic similarity of AI-assisted outputs across users.
When similarity exceeds a threshold, injects deviation prompts (e.g., Try a completely different genre, What would a competitor do?) or rotates the AI's default temperature/sampling parameters per user.
What the improved system can do: It counteracts the documented reduction in collective diversity (Doshi & Hauser, 2024) by actively maintaining variation across a team's outputs, even when individual users are highly satisfied with their results.
Improvement: Add a scaffolding level parameter that adjusts the AI's output complexity, abstraction, and explanation depth based on user skill:
Low-skill users: Receive more detailed explanations, step-by-step breakdowns, and simpler language.
High-skill users: Receive concise, high-level outputs with implicit assumptions and minimal hand-holding.
What the improved system can do: It prevents the skill-biased outcomes observed in Jia et al. (2024), where low-skilled agents experienced overload and anxiety while high-skilled agents thrived. The system ensures that abundance is converted into creative capacity for all users, not just experts.

The improved system can:

Stimulate divergent thinking by presenting multiple alternatives and reframings.
Prevent expert decline by avoiding ghostwriting defaults and offering challenge prompts.
Mitigate information overload by providing evaluation and integration support.
Sustain creativity over time by enforcing co-development protocols.
Adapt to task phases to avoid premature convergence or unnecessary divergence.
Maintain collective diversity in team settings through semantic monitoring and deviation prompts.
Support skill development by calibrating scaffolding to user expertise.

These improvements directly address the paper's core thesis: that GenAI's abundance must be matched with generative fit across cognitive, social, and organizational architectures to produce productive creativity rather than monoculture.

Abstract

Research on human-GenAI collaboration yields conflicting findings: GenAI can enhance creativity yet reduce collective diversity, with uneven benefits across skill levels. Rather than treating these as contradictions, we argue they reflect a core feature of GenAI: abundance. GenAI makes ideas, drafts, and recombinations plentiful, potentially expanding the hypothesis space and surfacing unanticipated possibilities. However, abundance alone doesn't ensure better outcomes. We propose generative fit as a unifying mechanism explaining when abundance yields productive creativity and when it backfires. Drawing on Generativity Theory, generative fit captures how well a system's generative potential complements a community's generative capacities. We develop a conceptual framework for collaborative human-GenAI settings where participants share goals, depend on one another, and must integrate diverse contributions. By mapping abundance to cognitive, social, and organizational factors of collective creativity, we explain apparent tradeoffs and offer actionable implications for designing workflows that convert abundance into valued creative outcomes.