Are Semantic Networks Associated with Idea Originality in Artificial Creativity? A Comparison with Human Agents
This is a strong comparative creativity paper because it does more than compare outputs from ChatGPT-4o and students. It shows that a standard human creativity expectation about semantic-network flexibility only partly transfers to AI, which is exactly the kind of result HCI needs for more careful CST evaluation.
Axes Lens
Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.
Contribution shape
- Knowledge form
- causal knowledge typical · 31/268
- Novelty type
- empirical finding typical · 68/268
- Abstraction level
- field typical · 41/268
- Generalization target
- field argument typical · 55/268
- Validation mode
- mixed methods typical · 136/268
Evidence profile
- Evidence strength
- strong typical · 158/268
- Claim alignment
- strong typical · 231/268
- Overclaim risk
- low typical · 53/268
Review Summary
This paper is most valuable as a mechanism-oriented contribution to HCI creativity research rather than as a headline claim that an LLM is simply more creative than people. The authors take a construct from psychological creativity research—semantic network organization—and ask whether it tracks idea originality in ChatGPT-4o the way it is expected to in humans. That framing matters because much prior discussion of GenAI creativity is product-centered: if the output looks creative, the system is treated as creative. Here, the paper instead asks whether a process-related indicator travels across human and artificial agents. The answer is only partly yes. The comparison with higher creative humans fits the expected pattern: higher originality aligns with more flexible networks. But the comparison with lower creative humans breaks the simple rule, because ChatGPT-4o is reported as more original despite having a more rigid network. That asymmetry is the real contribution. It suggests that human creativity constructs can be informative for AI, but not imported naively or assumed to behave identically across agents. Empirically, the paper is persuasive within its scope: it uses standardized tests, explicit hypotheses, a human sample split by creativity level, and statistical analyses of originality and network structure. At the same time, the authors are appropriately cautious. The study is a first step, uses a single black-box model, cannot inspect hyperparameters, and is sensitive to prompting, timing, and the narrow operationalization of creativity through divergent-thinking tasks. So the paper should be read as a strong field-shaping empirical finding with bounded generality, not as a universal theory of artificial creativity.
What Changed
Canon before
Creativity in HCI has predominantly been studied using product-centric definitions and comparisons of human and machine outputs; assumptions generally hold that semantic network flexibility correlates positively with originality in human creativity, and generative AI creativity is often measured by output similarity to human creative products.
Departure from common sense
Despite a more rigid semantic network structure, ChatGPT-4o demonstrates higher idea originality than lower creative humans, breaking the assumption that semantic network flexibility directly predicts originality in both humans and AI.
Actual novelty
This paper provides empirical evidence linking semantic network organization and idea originality in a large language model compared to humans; it shows that the association depends on human creativity level and uses this contrast to motivate an artificial creativity research agenda grounded in Artificial Cognition.
Evidence
The paper tests a clear hypothesis about semantic network flexibility and originality using ChatGPT-4o and 81 psychology students, with statistical comparisons across higher- and lower-creative human groups. Results support the expected association for higher creative humans, while also showing ChatGPT-4o can be more original than lower creative humans despite a more rigid network. The study is strong on comparative evidence but bounded by a single-model, black-box AI setup and a creativity measure centered on divergent thinking.
“ To the best of our knowledge, the paper provides the first empirical evidence of an association between semantic networks (an indication of the creative process) and idea originality (an indication of the creative product).”
actual novelty · 1 Introduction · confidence 0.95
“ Consistently, ChatGPT-4o was less original than higher creative humans, who also displayed a more flexible semantic memory network. However, the expected association did not manifest in the comparison with lower creative individual. Despite having a more rigid network, ChatGPT-4o emerged as more original than them”
departure from common sense · 1 Introduction · confidence 0.97
“This paper provides a first step in linking semantic networks and artificial creativity. The novelty of the approach encompasses a number of limitations, which limit the generalisability of our results. These pertain to the data sample, the study procedure, and the complexity of the creativity construct.”
limitation · 5.2 Limitations and Future Directions · confidence 0.95
“the chat interface of GPT-4o, which reproduced a complete black box example, as no information on model hyperparameters was available. The human sample included 81 psychology students divided into higher and lower creative individuals. Results confirmed the expected association between semantic networks and originality only with respect to higher creative individuals”
validation scope · 1 Introduction · confidence 0.94
Limits
Method limits
The study used only one AI model (ChatGPT-4o) through a black-box chat interface, so hyperparameters could not be inspected or controlled. The human sample was limited to psychology students, and the creativity assessment focused on divergent-thinking/AUT-style originality rather than the full creativity construct.
Deployment limits
Findings are specific to ChatGPT-4o in the tested interface and task setting, and may not transfer to other LLMs, prompting regimes, or hyperparameter configurations. The human baseline is also population-specific, so design implications for CSTs should be treated as provisional rather than universal.
Boundary conditions
The results depend on the operationalization of originality and semantic network flexibility in verbal divergent-thinking tasks. They hold under the study's black-box GPT-4o setup and student sample, and the paper itself notes that prompting and time constraints may alter outcomes in future work.
Position in field
The paper bridges HCI and psychological creativity research by importing Artificial Cognition into an empirical comparison of human and LLM semantic networks. Its main contribution is to challenge a simple flexibility-equals-originality assumption and to reposition artificial creativity as a field-level research problem for CST design and evaluation.