Requirements elicitation can use creativity techniques to generate innovative ideas and solutions. As there is a wide variety of such techniques, it is important to be able to support analysts in choosing the most appropriate ones for a project. This position paper presents an updated version of a logical framework that synthesizes the main characteristics of creativity techniques. Aspects related to artificial intelligence and large language model systems are included in this version of the framework as a preliminary contribution to dealing with their potential.
There is a high interest in creativity techniques to support requirements elicitation (the CReaRE
workshop itself is at its 12th edition, https://creare.iese.de), however, they are not widely used by
companies, except, in a limited way, that of brainstorming [
A relevant issue for creativity in requirements elicitation is how to choose the most appropriate
creativity technique – recently also referred to also as creative triggers or design thinking
techniques – among the large number of those available (see for example, [
The problem becomes even more challenging with the proliferation of artificial intelligence (AI)
tools and, in particular, generative AI (GenAi) systems based on Large Language Models (LLMs)
[
In such a context, the aim of this position paper is to update a logical framework proposed four
years ago to help companies choose creativity techniques applicable in requirements elicitation [
The first version of the logical framework was introduced to address a specific sub-question, “If a company wants to adopt a creativity technique for requirements elicitation, are there guidelines to support the choice among the different techniques?”. This paper first summarises the main aspects of the framework and then focuses on the factors to be considered in an updated version of the framework to take into account the recent developments of LLM GenAI. This implies answering another question: “Is it possible to use LLM GenAI systems to support the application of existing creativity techniques?”.
Given the need to interact in natural language, with different stakeholders, requirements elicitation is one of the most “relational” steps of software systems design, so that the application of creativity techniques, is the one where LLM GenAI impacts are stronger.
For the aim of this paper, it is important to highlight that a critical aspect of LLM GenAI
systems is the so-called hallucinated output, which reduces their usefulness and applicability scope.
In the context of creativity, there is an interesting relationship between “out-of-the-box thinking”
and hallucinations: it is significant that the primary goal of many creativity techniques (e.g.,
brainstorming) is to generate new ideas, postponing their evaluation in a subsequent step. The
emergence of creativity via hallucination has been investigated for example in [
The transformative nature – in terms of game-changer – of LLM GenAI in requirements
engineering activities has been investigated in [
3. Third order: transform. A new way of task accomplishment or a new set of tasks [
LLM GenAI tools can automate some of the activities needed to apply a creativity technique to requirements elicitation. E.g., for a brainstorming session, supporting the creation of groups, identifying stakeholders, creating a structured report of the ideas generated. LLM GenAI tools change the roles involved in the application of the creativity techniques. E.g., thanks to its conversational nature, LLM GenAI allows also non-technical stakeholders or end users to generate requirements, without a facilitator, as an individual task that does not require groups.
LLM GenAI are used to fully automate the elicitation process. E.g., generating requirements
for different types of stakeholders, according to a given technique, describing the idea in a
more or less formalized language. The activities and the role of requirements analysts
would change in ways that so far have been investigated (almost only) for the coding
activities [
According to this classification, first-orderLLM GenAI-related impacts on the application of
creativity techniques are feasible for all techniques and do not imply new factors in the framework.
Second order changes are related to the roles required for the application of LLMs GenAI and the
relationships between them and can be addressed with appropriate prompts [
The original framework is based on two matrixes introduced to collect relevant information about
creativity techniques, and to compare them in order to identify the most suitable for a given
software system project. In the first matrix, each creativity technique is described according to 5
criteria elaborated from a classification described in a paper that was read approximately 6,000
times in 2021, reaching 10,000 times in 2024 [
Process, to specify if a given technique also suggests a creativity process, i.e. steps to be accomplished for its applications. If so, AI agents exploiting LLMs could be introduced. Group vs. individual, to indicate if the technique can be applied individually, in groups or both. For group techniques, if groups are required to represent different stakeholders or experts in different domains, LLM GenAI could cover all of them, allowing an individual application as well; it could be applied by non-technical users, thanks to its conversational interface, also mitigating group problems.
Advantages, to highlight the known positive aspects of the technique. These advantages are useful to evaluate which LLM GenAI tool could be adopted, or even if a given technique could not be supported at all (e.g., creative pause implies dealing with empathy and emotion, a feature that existing LLMs systems are not able to support). A sound description of advantages could be used to write proper prompts, e.g. “force to consider different viewpoints” suggests asking the LLMs GenAI to play different roles.
Disadvantages, to indicate the critical aspects of the technique. Symmetrically to the previous factor, e.g., to translate into a prompt “Requires knowledge of stakeholders’ viewpoints”.
Sources, to allow an analyst to have more information on the technique and its application. In this case the factor is split in two columns: ‘Technique sources’ and ‘AI application sources’, where the second indicates if a given technique has already been implemented with LLMs GenAI.
The second matrix (Table 2) includes a set of parameters identified in requirements elicitation and
project management best practices and guidelines [
Both Both
Well known High number of new ideas Simply to apply Force to consider different viewpoints Performs better than brainstormi ng
Disregarded principles Requires high abstraction skills Requires knowledge on stakeholders Highlighted in bold those where existing LLM GenAI systems can be helpful: documentation can be translated into many languages, also adapting to different levels of technical language; LLMs can be trained on a company’s documents to gather knowledge about the required domains (e.g., if there are no experts in a domain relevant to the application of the creativity technique); the role of facilitator could be played by the GenAI system if necessary. ‘Tool support’ is split into three new parameters to specify, respectively, if the technique could be applied almost straightfully, i.e. translating the guidelines or principles of a given technique using prompts; AI agents are needed if more activities have to be supported in a coordinated way; prompt library, if there are already prompts for the technique. Low No Yes Yes
No Examples of creativity techniques that could be applied with a simple translation of the suggestions into a prompt for LLMs GenAI systems are ‘Synapses’ (“Seeking stimuli in fields far from the one where the problem arises”), or “Forced relations” (“looking for forced relations between usually uncorrelated ideas”) [21]. Further investigations are needed to create a prompt library.
The Research Agenda for GenAI for software engineering [
This paper is a preliminary contribution to address this gap by updating the framework
proposed in [
Could we use the factors in the matrixes to ask GenAI systems to choose the most suitable technique? And how to apply it? How will GenAI change analysts work and software engineers work? Will it be possible to write programs that invent requirements? And how could a human-AI collaboration be established to be creative in requirements elicitation? How can we update curricula, certifications and in general education to prepare the future analysts for a world where AI and LLM GenAI is dominant, so that they will be able to use creativity in requirements elicitation [22]? What are the risks of using (evolving) LLM GenAI for requirements elicitation? How to address copyright (documents used for training) and explainability problems [23]. Finally, a more disruptive question is: Are creativity techniques still necessary? Or will there be AIbased invention systems that are able to find new ideas and requirements giving companies a competitive advantage? A preliminary answer to this question could be that creativity techniques are necessary to design such general-purpose invention systems. In fact, an AI-assisted invention method has been developed by Iprova, a company that actually applies one of the most classical creativity principles, i.e., “adapting a given solution or idea to a different area”. Another inventionproducing AI focused on biological problems is the one used by BioMedIt for AlphaFold [24].
The author has not employed any Generative AI tools. [19] A. Rachmann, Six thinking chatbots: a creativity technique deployed via a Large Language Model, in: Proc. REFSQ 2024, Workshops, WS-CEUR, 3672, 2024. URL: https://ceur-ws.org/Vol3672/CreaRE-paper1.pdf [20] L. Mich, C. Anesi, D. M. Berry, Applying a pragmatics-based creativity-fostering technique to requirements elicitation. Requirements Engineering 10 (2005) 262-275. [21] J. W. Young, A technique for producing ideas. Thinking classics, 1st ed., McGraw-Hill
Education, 2003. [22] E. Shein, The impact of AI on Computer Science education, Commun. ACM 67, 9 (2024) 13–15.
doi:10.1145/3673428. [23] I. Ozkaya, Application of Large Language Models to Software Engineering Tasks: Opportunities, risks, and implications, IEEE Software 40, 3 (2023) 4–8. doi:10.1109/MS.2023.3248401. [24] L. Laursen, Artificial Intelligence: Can we automate eureka moments? IEEE Spectrum 61, 1 (2024) 24-27. doi:10.1109/MSPEC.2024.10749728.