This is chapter 8 of the book "Heuristic Enquires" edited by Professor Welby Ings and Professor Keith Tudor

https://www.taylorfrancis.com/books/edit/10.4324/9781003507758/heuristic-enquiries-welby-ings-keith-tudor

Written by: Akbar Ghobakhlou and Hossein Najafi

Introduction

This chapter is written by a computer scientist and an artistic practitioner who embarked on a discussion about the nature of heuristics in their respective fields of inquiry. Although seemingly distant, we recognised significant con- nections between heuristic inquiry in artificial intelligence and in artistic research. The application of heuristics is crucial in navigating the complexi- ties of problem-solving, particularly in fields that demand creative and flexi- ble thinking. To begin the discourse on the necessity of heuristic approaches to problem-solving, a simple yet illustrative example may serve as an effective starting point.

Imagine you are a salesperson with the task of visiting a number of cities, selling goods, and then returning home. To do this, you must find the most efficient route between the cities to save both time and fuel. Let’s begin by assuming that there are only three cities to visit. In this scenario, determining the shortest route is relatively straightforward because there are only a few possible paths to compare. For cities A, B, and C, you could travel A-B-C-A or A-C-B-A, and it would be easy to determine which journey will be shorter. However, as you add more cities, the complexity of the problem increases dramatically. With each additional city, the number of potential routes grows exponentially, making it very challenging to find the shortest route. For exam- ple, with ten cities, you face over 3 million possible routes, and visiting 60 cities, the number of possible routes becomes more than the estimated num- ber of atoms in the observable universe. This phenomenon is known as com- binatorial explosion (Schuster, 2000).

This famous example is called the Traveling Salesman Problem, a conun- drum which, in computational complexity theory falls within a class of non- deterministic polynomial (NP) time, and is considered NP-hard (Jünger et al., 1995). NP-hard, in computer science, refers to problems that are notoriously difficult to solve because the time required to solve them increases exponen- tially with the size of the problem. In essence, as the number of cities grows, finding the absolute shortest route becomes impractical. To tackle this, we utilise heuristics. Heuristics trades off completeness, accuracy, and precision for speed. For example, for the Traveling Salesman Problem, heuristic meth- ods provide good enough solutions. These might not provide the absolute shortest route, but they come close in a fraction of the time it would take to calculate an exact solution. Techniques like the nearest neighbour, where you always visit the closest unvisited city next, or the genetic algorithm, which mimics natural selection to find progressively better routes, are examples of heuristic approaches (Jünger et al., 1995). In a completely different kind of challenge, when an artist is navigating an immense number of unknown possibilities inside a complex project, they also employ heuristic methods (Ings, 2011). However, this application is more qualitative than quantitative. This said, both approaches elevate broad estima- tion or a sense of effectiveness over the pursuit of certainty or absolute accuracy.

Heuristics in artificial intelligence

Artificial intelligence (AI) is a generic term that encompasses a broad range of computational techniques designed to mimic or replicate human-like intelli- gence. In this chapter, AI is discussed on multiple levels, from that of a generic concept to artificial neural networks (Figure 8.1). AI systems are capable of performing tasks that typically require human intelligence, such as visual per- ception, speech recognition, decision-making, and language translation (Russell & Norvig, 2010). Within the domain of AI, machine learning (ML) emerges as a crucial subset. It focuses on the development of processes and statistical models that enable computers to learn and make predictions or decisions based on data, rather than being explicitly programmed (Alpaydin, 2020). ML includes different methods including a specialised approach known as deep learning (DL), which involves training artificial neural net- works to work with large volumes of data. DL models can automatically dis- cover the representations needed for feature detection or classification from raw data, so this makes it a powerful tool for many AI tasks (LeCun et al., 2015). A pivotal element of DL is artificial neural networks (ANNs), which are inspired by the biological neural networks of animal brains. ANNs are com- posed of interconnected groups of nodes, resembling the vast networks of neurons in the biological brain. These are used for pattern recognition and cognitive tasks (Goodfellow et al., 2016).

FIGURE 8.1  Mapping the scheme of artificial intelligence, machine learning, deep learning, and artificial neural networks.

Heuristics in AI refers to strategies or methods of problem-solving when encountering complex and difficult problems.1 They operate quickly at the expense of accuracy, so they can potentially lead to errors or suboptimal solu- tions. Heuristics are essential to AI because if we put the need for optimality aside, they allow a system to make rapid progress when solving complex problems. They are essential tools for moving about the wide, frequently dark spaces that exist in many AI problems. Given the distinctive nature and use of heuristics in AI, the next section provides an overview of evolutions in the field.

A brief history of artificial intelligence

The term ‘artificial intelligence’ was first coined by John McCarthy during a seminal workshop at Dartmouth College in 1955, and this use marked the formal birth of the field (Flasinśki, 2016). Having convened leading research- ers from diverse disciplines, McCarthy’s workshop sought to create machines capable of mimicking human cognitive functions. However, despite their ini- tial enthusiasm, early researchers in AI faced formidable challenges, such as the complexity of natural language and the constraints of computational resources available at the time (Flasinśki, 2016).

Between the 1950s and the 1980s, AI research predominantly focused on symbolic AI (an approach in which systems manipulated symbols to perform tasks, akin to human reasoning). This era saw the rise of expert systems,2 which operated based on predefined rules and knowledge bases, thereby enabling such systems to emulate the decision-making abilities of human experts in specific domains (Jackson, 1986). However, this approach was lim- ited because of its reliance on extensive domain-specific knowledge and its inability to handle uncertainties and learn from data.

The late 20th century witnessed periods known as ‘AI winters’ (Harguess & Ward, 2022, p. 1) that were characterised by reduced funding and a decline in interest in AI research. These winters were largely a result of the limitations of symbolic AI, particularly its ineffectiveness in dealing with real-world uncertainties and learning challenges, which, in turn, led to a more sceptical public perception of AI (Crevier, 1993).

A resurgence of interest in AI emerged in the 1990s with the advent of machine learning (ML). Unlike its predecessors, ML algorithms empowered systems to recognise patterns from data, and they marked a significant shift from rule-based methodologies to data-driven approaches. Techniques such as decision trees, support vector machines, and clustering became instrumen- tal in this new era, enabling machines to learn and make predictions or deci- sions based on data (Alpaydin, 2020).

The 2000s saw the rise of artificial neural networks (ANNs), inspired by the structure and function of the human brain. A notable subset of ML, known as deep learning, involved the use of neural networks with multiple layers. This approach achieved remarkable success, including image recognition and nat- ural language processing (Goodfellow et al., 2016). The growth of deep learn- ing may be attributed to several factors including the availability of large volumes of labelled data, significant advances in hardware and software, and the development of new architectures and algorithms, including heuristics (Schmidhuber, 2015).

Machine learning

As a subset of AI, machine learning has gained significant traction and appli- cation across various fields including computer vision, cybersecurity, health, and language processing. Driven by data, ML algorithms recognise patterns and statistical relationships, thereby enabling learning and adaptation, and encompass a variety of approaches such as regression, random forests, sup- port vector machines, and artificial neural networks, each with unique capa- bilities to address diverse problems (Cioffi et al., 2020).

ML algorithms are commonly categorised into three main types:

• Supervised learning, which is an approach that involves training algorithms and processes on labelled data, where the relationship between input and desired output is The algorithm learns to map input to output, identifying patterns within the data. It is further divided into two main categories: classification and predic- tion. Classification involves the process of categorising data into predefined groups or classes, while prediction involves forecasting continuous or discrete values (Dridi, 2021). Applications range from email filtering to medical diagnosis and price prediction.
• Unsupervised learning, whereby algorithms learn from unla- belled data, discovering hidden patterns without guidance, which results in being able to cluster similar data points and reduce data Examples of unsupervised learning include busi- ness use of clustering for customer segmentation, and association rules in market analysis (Hruschka, 2021).
• Reinforcement learning, in which using a trial-and-error approach an agent learns optimal actions through rewards or This type of learning is a growing area in autonomous systems such as self-driving cars, where the vehicle learns and improves decision- making over time.

(Moussaoui & Benslimane, 2023)

ML turns data into useful predictions and decisions. To do this well, ML fol- lows six steps, which help us progress from collecting data to ensuring that the ML model works properly (Figure 8.2). Each step plays a key role in making sure we obtain reliable results from models, while the process as a whole forms the backbone of a successful ML project.

FIGURE 8.2  The six-step process in machine learning.

Data gathering involves collecting relevant information from various sources to create a comprehensive dataset (Provost & Fawcett, 2013). Data preparation is used to clean and structure the data, for instance, addressing inaccuracies, and ensuring it’s in the right format (Mansingh et al., 2016). Feature engineering focuses on selecting and modifying key data elements to enhance model learning. This requires domain expertise (Zheng & Casari, 2018). Model selection entails choosing the appropriate ML model based on problem type and data characteristics, while model training involves adjust- ing parameters and avoiding overfitting or underfitting (James et al., 2021). Finally, model evaluation assesses the model’s performance using appropriate metrics on a separate testing dataset (Kuhn & Johnson, 2013).

Kuhn and Johnson (2013) note that the iterative nature of this process often requires revisiting and refining steps based on model performance, and this underscores the dynamic and adaptive essence of ML. James et al. (2021) argue that this not only enhances the predictive accuracy of ML models but also reinforces their applicability across diverse domains. Deep learning, an advanced subset of machine learning, involves comput- ers learning from examples, not explicit instructions, allowing them to under- take complex tasks like speech recognition and object identification in images (Goodfellow et al., 2016; Schmidhuber, 2015). Central to this are artificial neural networks (ANNs), inspired by the human brain (Figure 8.3), which use interconnected artificial neurons to learn data patterns, enabling them to rec- ognise intricate details like handwritten digits (LeCun et al., 2015).

FIGURE 8.3  Deep neural networks. Source: (Merenda et al. (2020). Note: Used under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International.

Deep inside ANNs lies the latent space, an abstract, multidimensional arena where data patterns are simplified and compressed. This space, crucial for tasks like facial recognition, is often incomprehensible to humans, leading to the black box problem, where AI decisions are not easily interpretable (Guidotti et al., 2020). Current research is addressing this by developing tech- niques to make AI more interpretable, ensuring its decisions are more aligned with human understanding (Doshi-Velez & Kim, 2017).

Examples

As noted earlier, the Travelling Salesman Problem is generally considered an NP-hard problem because producing an optimal solution even for a small or moderate-sized dataset is a challenging task (Jünger et al., 1995). This means that there is no known efficient algorithm that can solve the problem in poly- nomial time, and the time required to find the best solution grows exponen- tially with the number of cities. An example of a heuristic method used for the Travelling Salesman Problem is ‘the greedy algorithm’. This works by choosing the closest city to the current location and repeating this process until all cit- ies are visited. This method does not guarantee the optimal solution, but it is fast and easy to implement. The Travelling Salesman Problem is a combinato- rial optimisation problem that has multiple applications in our daily lives. This includes airport networks, goods delivery, logistics (planning and scheduling), the maritime industry, public transportation networks in top-tier cities, vehi- cle routing, and so on (Matai et al., 2010).

In another example, in search optimisation problems, the A* (A-star) algo- rithm stands out for its effective use of heuristics to find the most efficient path between two points. A* combines the principles of breadth-first and greedy best-first searches. Breadth-first search examines the search space comprehensively by exploring all nodes at a certain depth before moving deeper, ensuring a thorough search but often at the cost of efficiency. Greedy best-first search, in contrast, focuses on paths that appear to lead directly to the target, guided by a heuristic that estimates the closest path. A* merges these strategies by using a heuristic that considers both the travelled distance (breadth-first aspect) and an estimated distance to the goal (greedy aspect). This integrated approach allows A* to prioritise paths based on the total esti- mated cost, effectively balancing thoroughness and efficiency. Widely used in fields like robotics and navigation, A* optimises routes to reach the desti- nation, demonstrating its effectiveness in complex problem-solving scenarios (Wilt et al., 2010).

Heuristics are also crucial in practical applications like antivirus software. They enable these programs to effectively detect and isolate various forms of malware by identifying code or behaviour patterns indicative of viruses. Such heuristic approaches are particularly adept at identifying new or mutating threats, showcasing their adaptability and providing foresight in cybersecurity (Dube et al., 2012).

Another significant application of heuristics in AI is in decision-making processes where situations are marked by uncertainty and ambiguity. Fuzzy logic, proposed by Zadeh (1988), employs a heuristic approach to model the imprecision inherent in real-world scenarios. By allowing for gradations or degrees of truth rather than a strict binary approach, fuzzy logic facilitates more nuanced and human-like problem-solving in AI applications, ranging from control systems to natural language processing (Sarker, 2022).

Finally, we find heuristics involved in what is sometimes described as the Knapsack problem. Imagine a thief sneaking into a bank vault filled with treasures. The thief has only one knapsack and there are countless valuable items available. The thief must decide which articles to take to maximise the heist without overloading the bag. In the Knapsack problem the objective is to select the most valuable combination of items without exceeding a weight limit. To tackle this, AI often employs a heuristic method that mirrors the thief’s strategy (Cacchiani et al., 2022). An evaluation of each item will be based on its value-to-weight ratio and the thief will grab the most valuable items first, while ensuring that the knapsack is not too heavy to carry. This assessment allows the thief to make the most out of a risky venture. In real- world applications, such methods help in optimising resource allocation and decision-making across various industries, and are applied across the fields of logistics and asset management to network optimisation and energy efficiency (Cacchiani et al., 2022). Having now considered the nature of heuristics in the development of AI, we turn to research in the field of artistic inquiry, by which we refer to research where artistic knowledge is pursued and embodied in products of art (Klein, 2010; Lesage, 2009).

Heuristics in artistic research

Heuristic inquiry is increasingly recognised and embraced by creative researchers working in practice-based and practice-led artistic studies (Hoyle, 2023; Peacock, 2023). Rooted in humanistic and phenomenological tradi- tions, heuristic inquiry in such creative contexts emphasises the role of imag- ination, intuition, and subjective experience in the process of knowledge creation. It offers a framework that not only acknowledges but also actively incorporates the researcher’s personal experience, emotional engagement, and creative intuition in the research process (Douglass & Moustakas, 1985). This methodology contrasts with more conventional research approaches that often prioritise objectivity and detachment. In practice-led artistic research, the self cannot be separated from the inquiry, because the creator is central to the creation, evaluation, and development of their practice (Griffiths, 2010). When developing an artistic inquiry, Klein states that research engages sen- sory and emotional perception ‘through artistic experience, from which it cannot be separated’ (2010, p. 6). He argues that, ‘whether silent or verbal, declarative or procedural, implicit or explicit… artistic knowledge is sensual and physical, “embodied knowledge”. The knowledge that artistic research strives for, is a felt knowledge’ (2010, p. 6).

However, there is always a tension between the self within the arena of practice, and the self as an objective reviewer of context (Hamilton, 2011). The researcher must negotiate this balance without tipping too far towards the solipsistic (or confessional), or too far towards the objectified (Sela-Smith, 2002). This is because to function authentically as an artistic researcher, one must be positioned as a subjective, critical insider (Duncan, 2004).

Heuristic inquiry emphasises the role of the researcher as a whole person, integrating their cognitive, emotional, and creative capacities. This holistic view, suggested by scholars like Clark Moustakas, underscores the impor- tance of creativity and artistry in research. Heuristic inquiry encourages researchers to embark on a personal creative journey. Such an approach is particularly effective where innovation and personal interpretation are key elements, allowing researchers to uncover new understandings (Scrivener, 2000) and new meanings (Ings, 2011). Ultimately, heuristic inquiry has prac- tical implications in bridging the gap between research and real-world appli- cations, especially in humanistic and artistic fields. By focusing on embodied experiences and fostering a profound understanding of the self and the sur- roundings, heuristic inquiry enhances the relevance and impact of research. As a system of inquiry, it is particularly beneficial in domains where human experience and personal narratives are central, because it provides a bridge between academic research and practical, humanistic applications (Snyder, 2012).

Phases of heuristic research

In 1990 Clark Moustakas posited a flexible framework for heuristic inquiry contained by (as he experienced it) six phases, each of which he saw as inte- gral to facilitating a journey from initial curiosity to profound understanding and creative synthesis. The phases were: initial engagement, immersion, incu- bation, illumination, explication, and creative synthesis.

However, these phases rarely play out as a chronological sequence of engagements in artistic research. If we trace the earliest applications of heu- ristic inquiry in postgraduate theses in art and design, Moustakas’ six-phase framework appears to have emerged in a dissertation in 2006 when Welby Ings employed it in his practice-led doctoral thesis Talking Pictures. In this study he drew on the thinking of John Wood who, in 2000, had suggested that because heuristics is more concerned with discovery than proof, it is a useful method for designers because it enables them to move away from the pursuit of linear, finite questioning.

Since Ings’ thesis, heuristic inquiry has been employed in a range of research projects by candidates working in visual communication design, film, and photography. Among these, it has generally been adopted as a con- ceptual framework that avoids following a ‘how to do’ template that Douglass and Moustakas (1985) were so wary of proposing. In artistic research one normally encounters a broad heuristic approach, within which specific artis- tic methods are brought to data gathering and creative synthesis, with empha- sis being placed on ‘a subjective process of reflecting, exploring, sifting, and elucidating the nature of the phenomenon under investigation’ (Douglass & Moustakas, 1985, p. 40).

In gender studies, scholars have approached the study of non-binary gen- der identity with a nuanced understanding of contemporary queer discourses. Halberstam, in their seminal work, conceptualises these processes as a ‘con- tinuous building and unbuilding’ (2018, p. 3), encapsulating a dynamic and evolving nature of gender identity. This perspective is further expanded by Ball, who articulates the concept as an ‘iterative and intimate self-making’ (2021, p. 12), highlighting the personal and recursive aspects of gender iden- tity formation. These theoretical frameworks have been instrumental in informing Ezra Baldwin’s 2023 thesis, which delves into the complexities and lived experiences of non-binary individuals, grounding their research within these established discursive constructs.

Conversely, the photographer Leanne Miller, whose study dealt with the subjective experience of home invasion, defined the phases of her inquiry as ‘recollecting, relating, and creating’ (2013, p. 33). The film poet David Sinfield described the phases he navigated when creating a typographical voice for an abandoned Freezing Works as ‘dwelling within a site, recording, and creative synthesis’ (2020, p. 80).

Rene Burton (2015) initially posited, followed by Sinfield (2020), that phases in an artistic heuristic inquiry ‘operate in relation to each other [and the researcher] moves fluidly between them’ (Sinfield, 2020, p. 83). This per- spective was further developed by Igelese Ete (2021) and then most recently by Toiroa Williams (2024).

Visual communication designers Tavares (2011), Chooi (2017), and Panaita (2018), who have written and illustrated graphic novels as thesis submissions, describe immersive ways of heuristically creating stories that contain varying degrees of autobiographical material. None of these candidates employed Moustakas’ six phases of heuristic research, although they all identified broader approaches like ‘immersion’ (Moustakas, 1990), or the concept of ‘indwelling’ (Conlan, 2000), as important when accessing, then composing, narrative elements. Elena Panaita described her heuristic trajectory as ‘deeply subjective because I am sensing my way through memories and creating a process that may help to understand my personal experience, the conditions that gave rise to it, and the communicative potentials of my artistic orchestra- tion of elements’ (2018, p. 34).

Both Ventling (2018) and Najafi (2023) have employed heuristic inquiry in artistic research journeys based on theories of practice and enlightenment. Ventling employed Bonaventure’s 13th-century spiritual theory that illumina- tion is a transformational aspect in a cognitive journey in which light is sub- jectively experienced through engagement with materials and processes of making. Najafi’s (2023) study interwove heuristic inquiry and a Persian illumi- nationist approach to investigate the phenomena of acculturation, displace- ment, self-continuity, and identity transition experienced by refugees. In Persian illuminationism, heuristic inquiry is transcended in the form of a jour- ney, where knowledge is gained through praxis, or metaphorically through a journey where the artist and practice unify.

Effective questioning

Heuristic inquiries, including artistic ones, are heavily question-driven (Ings, 2011; Ventling, 2017, 2018). As Moustakas notes:

Heuristic inquiry is a process that begins with a question or problem which the researcher seeks to illuminate or answer. The question is one that has been a personal challenge and puzzlement in the search to understand oneself and the world in which one lives. (Moustakas, 1990, p. 15)

Moustakas proposes that the heuristic research journey cannot be hurried or timed by a clock or calendar (for a discussion of which, see Cousins et al., 2024 [Chapter 9 in this volume]). It demands the researcher’s complete pres- ence. Although a heuristic inquiry is largely autobiographic, he suggests that ‘with virtually every question that matters personally, there is also a social, and perhaps universal, significance’ (Moustakas, 1990, p. 15). In the pursuit of a research question, he says that the researcher enters into dialogue with the phenomenon, allowing the phenomenon to speak directly to one’s own experience, to be questioned by it. In this way, one is able to encounter and examine it, to engage in a rhythmic flow with it – back and forth, again and again – until one has uncovered its multiple meanings. Kleining and Witt suggest that although heuristic research procedures are non-linear, they are dialectical. They recommend that the heuristic researcher asks their material open questions ‘in a similar way one may ask a person, receiving “answers” and questioning again’ (2000, para. 14).

According to Moustakas (1990), effective heuristic questions are character- ised by a desire to reveal the essence of human experiences qualitatively, and they are elucidated through creative means. An effective heuristic question, he suggests, will lead the researcher to a deeper understanding and connec- tion with the subject, creating a thirst for discovery and clarification. Although Moustakas maintains that the researcher’s openness in clarifying the question will lay the foundation for discovery, in much artistic inquiry one is dealing with questioning that is neither singular nor fixed. This is because questions must respond dialogically to expressions of ideas as artefacts take conceptual and physical form. Thus, while a question about an experience may initiate an inquiry, as a project progresses questions will morph as the researcher engages with artistic data processing and critical reflection (Baldwin, 2023; Sinfield, 2020; Ventling 2018). Significantly, as a project progresses into creative syn- thesis, questions may adopt a seemingly poetic nature. Thus, one might ask ‘What is the weight of the colour of this emotion?’ or ‘What is the rhythm of this pause?’ or ‘Is the depth of this event best served with monologue or a series of images?’

Comparing heuristics in AI and artistic research

This chapter explores the intersections of heuristics within the fields of artifi- cial intelligence (AI) and artistic research. Despite their distinct disciplinary backgrounds, both fields are heavily informed by heuristic inquiry proce- dures. The following discussion is organised into seven key themes, each highlighting a specific aspect of heuristic inquiry as it manifests in AI and artistic research. The selection and order of these themes are designed to reflect the progression from abstract concepts, such as embracing the unknown, to more concrete applications, like problem-solving and pattern recognition, culminating in a discussion on certainty. This progression aims to provide a comprehensive understanding of how heuristic principles operate similarly and differently across these fields.

Being at ease with the unknown

The rise of AI indicates a notable shift in scientific exploration as it moves towards a greater acceptance of the unknown and an openness to engage with perplexing data. This transition represents a departure from rule-based and algorithmic systems, including the older symbolic AI. Historically, scien- tists and researchers sought to decipher the logic of the natural world through deterministic and linear methods. However, the complexity of real-world phenomena often baffled these approaches. Increasingly, ambiguity and uncertainty are not only acknowledged but also embraced as integral to understanding and exploration. This shift is evident in the work of pioneering computer scientists and cognitive psychologists, who recognised the limita- tions of rigid algorithmic methods and advocated for more flexible, data- driven approaches (Gigerenzer & Gaissmaier, 2011; Newell & Simon, 1972; Russell & Norvig, 2010). AI procedures aim to be inherently adaptive and capable of learning from unstructured data (LeCun et al., 2015) and utilise probabilistic models that accommodate uncertainty and variability.

Embracing the unknown and navigating one’s way through its potential is fundamental to heuristic inquiry in artistic research (Ings, 2014; Paora, 2023; Ventling, 2018). This process involves delving into the unknown features of an experience in intuitive ways (Bach, 2002), which can be related to what Keats calls ‘negative capability’, a state ‘when a man is capable of being in uncertainties, mysteries, doubts, without any irritable reaching after fact or reason’ (1895, p. 193). In the context of heuristic research, this capability manifests as a readiness to explore and engage with the unknown, allowing the flow of novel insights, putting aside apprehension, fear, and the pursuit of anxious or rapid closure (Eisold, 2000).

This embracing of the unknown in heuristic inquiry may be aligned with the concept of latent spaces or the ‘black box’ in AI, where complex data are transformed into abstract, less tangible forms. Just as heuristic researchers navigate the unknown to gain insights, AI systems traverse through latent spaces, deciphering patterns and deriving meaning from data that is not immediately clear or structured.

Iterative processes

In AI, particularly within machine learning, iterative processes are integral to the development and optimisation of procedures. Heuristic models in AI engage in a cycle of generating, modifying, and restructuring data or sym- bols, where each iteration refines the approach based on feedback (Moussaoui & Benslimane, 2023). This involves assessing the effectiveness of each step in reaching the desired outcome, learning which paths are more promising, and eliminating less effective ones. For example, heuristic rules in antivirus soft- ware iteratively improve their detection and isolation of malware, enhancing accuracy and efficiency with each cycle (Ye et al., 2018). In artistic research, the iterative process is generally more introspective and exploratory, and involves a recursive journey of exploration, reflection, and synthesis (Najafi, 2022). Broadly, researchers engage with subject matter and enter into dia- logue with their subjective experience as they begin to synthesise insights, in an attempt to deepen their understanding of the research topic. While in AI the focus is on efficiency and accuracy, in artistic research emphasis is placed on gaining a nuanced understanding and personal insight that leads to communicative artefacts. Nevertheless, both fields utilise an iterative process as researchers move from initial concepts or models towards increasingly refined and resolved outcomes.

Intuition

Intuition is a multifaceted concept with diverse definitions. Simon describes it as ‘bounded rationality’ (1979, p. 501), a simplified reasoning sufficient for practicality. Alternatively, Myers defines intuition as ‘our capacity for direct knowledge, for immediate insight without observation or reason’ (2002, p. 1) and according to Epstein, ‘Intuition involves a sense of knowing without knowing how one knows’ (2010, p. 296). Similarly, in heuristic artistic inquiry, intuition draws on tacit and explicit knowledge, enabling researchers to gain deep insights into their subject matter (Moustakas, 1990). This process involves a personal and subjective application of intuition, where artists engage deeply with their experiences uncovering new meanings and under- standings (Ings, 2015).

In AI, intuition is often subtly embedded in heuristic algorithms, serving as a form of artificial intuition (Johnny et al., 2020). These procedures enable AI systems to make educated guesses. This involves an intuitive understanding of which paths to pursue and which to avoid, where the heuristic model itera- tively refines its approach to align closely with the desired solution (Wu et al., 2021). Whether deeply personal of artificial, in both artistic research and AI intuition is employed to uncover new insights.

Problem-solving

Both AI and artistic research utilise heuristics for problem-solving, although in different contexts and applications. In AI, heuristics are instrumental in sim- plifying complex decision-making processes, enabling systems to provide feasible solutions swiftly, especially in scenarios where exact solutions are not computationally practical (NP-hard). This concept is discussed by Simon and Newell (1958), who note that computers employ heuristic problem- solving in instances lacking specific algorithms, and this closely parallels human problem-solving methods. Additionally, heuristics in AI play a crucial role in reducing overwhelming complexities into more manageable forms, thus accelerating the problem-solving process (Simon, 1955).

On the other hand, in artistic research, heuristics guide the discovery pro- cess, aiding researchers in making intuitive decisions in complex situations where definite answers are out of reach (Yilmaz et al., 2016). Simon’s (1979) principle of satisficing (which advocates for satisfactory rather than optimal decision-making) is particularly applicable in creative contexts where decisions often have to be made under time, resource, or information constraints (Yilmaz & Seifert, 2011).

Pattern recognition

ML models, integral to AI, rely heavily on data to identify patterns and statisti- cal relationships, facilitating the learning and adaptive capabilities of AI sys- tems (Flasinśki, 2016). This approach marks an evolution from earlier rule-based methods to more dynamic, data-driven models. Techniques such as decision trees, support vector machines, and k-means clustering3 have been instrumental in this paradigm shift, with deep learning excelling in tasks such as image recognition and natural language processing (Russell & Norvig, 2010). These advancements owe much to sophisticated pattern recognition algorithms that enable these systems to interpret complex datasets effectively. In the domain of artistic research, pattern recognition plays a similarly inte- gral role that involves the identification of resonances, themes, or recurring elements (Ventling, 2017). Kleining and Witt (2000) argue that one of the significant skills in heuristic inquiries is the ability to discover similarities, patterns, analogies, or homologies within the diverse data that are being col- lected and processed, and they suggest that the success of pattern recognition as a procedure can be measured by the richness of the result, its cohesive patterns, and inter-subject validity. In heuristic artistic inquiry, this skill has been discussed in relation to artefact development by Ings (2004, 2018), Panaita (2018), Mortessen Steagall (2019), Tavares (2019), and Najafi (2023).

Embracing experience

In AI, data fundamentally represents experiential knowledge, encompassing a wide spectrum of human experiences, interactions, and observations (Jordan & Mitchell, 2015). This experiential nature of data is pivotal because it pro- vides the raw material from which AI systems learn and develop. These often vast and multifaceted datasets capture the intricacies of human behaviour, societal trends, and natural phenomena, making them integral to the learning processes in AI. Utilising statistical and neural procedures, AI translates these iterations of human experiences into computational models capable of addressing complex real-world problems. This translation process is not merely a mechanical replication but involves the infusion of human cognitive patterns into the algorithms. Russell and Norvig (2010) note how AI systems, especially in machine learning (ML), incorporate these cognitive residues in their procedures, effectively blending human-like intuition with computa- tional efficiency. In artistic research, heuristic inquiry enables a transformative process, where personal knowledge and experience lead to universal insights. This mirrors how AI, through its learning procedures, transforms experiential data into generalisable knowledge, capable of addressing broad, real-world issues. The process is not purely analytical: it also involves a creative synthesis of data, paralleling the intuitive and deeply reflective nature of heuristic inquiry in artistic research. Moreover, just as heuristic inquiry in artistic research embraces the complexity and ambiguity inherent in human experiences, AI’s use of experiential data embraces the multifaceted nature of real-world data, weaving it into sophisticated models that reflect nuanced human experiences and cognitive patterns. This similarity underscores a shared underlying assumption that deep understanding and innovation emerge out of engaging intimately and creatively with complex, experiential data.

Certainty and uncertainty

In our previous discussion on uncertainty, we explored how the rise of AI and artistic research has led to a greater acceptance of ambiguity and openness to engage with perplexing data. AI approaches differ from traditional algorithmic methods in their handling of ambiguity and uncertainty by acknowledging and embracing these elements as integral parts of the problem-solving process. Certainty in AI and in artistic research can be discussed through the phenomena of underfitting and overfitting.

Underfitting in AI occurs when a model is too simple and fails to capture the complexity or patterns in the data. This results in poor performance on both training and new datasets (Russell & Norvig, 2010). This may be likened to conducting insufficient research in a creative endeavour, where a lack of depth and exploration leads to superficial or incomplete understandings and outcomes. It may also be associated in artistic research, with instances when the researcher becomes too self-centred. If they avoid external feedback, they might be underfitting the data critical to their research. In an artistic inquiry the researcher must negotiate a balance of the self and the arena of practice without tipping too far towards solipsism (Barrett & Bolt, 2014), or towards the overly objectified (Sela-Smith, 2002).

Conversely, overfitting in AI refers to a model that is excessively complex, fitting too closely to the training data, and capturing noise along with the underlying pattern. This leads to a model that performs well on training data but poorly on new, unseen data, essentially because it is too sure of its train- ing data to generalise effectively (Goodfellow et al., 2016). In artistic research, overfitting is akin to being overly convergent and certain about one’s findings or creative direction, to the extent that there is excessive focus on specific details or a single perspective. In discussing this phenomenon, Polanyi sug- gests that the researcher must linger in ambiguity to overcome ‘sharp outlines of certainty, only to dissolve again in the light of second thoughts or of further experimental observations’ (1966, p. 30).

Conclusion

This chapter has considered the nature of heuristics in the realms of AI and artistic research. The comparison reveals a nuanced understanding of how heuristics function within seemingly disparate domains, each with its unique demands and characteristics, yet both connected by underlying concepts.

In the field of AI, heuristic methods have proven essential for tackling com- plex, data-rich environments. These methods, including algorithmic approaches to neural network-based systems, embody a form of artificial intuition, allowing AI systems to navigate vast datasets and uncover patterns within them (Goodfellow et al., 2016; Russell & Norvig, 2010). This aspect of AI highlights the integration of human cognitive patterns into computational models, that enable machines to perform tasks with a degree of human-like intelligence and intuition.

Heuristic inquiry as it is employed in artistic research draws some of its distinctiveness from the work of Clark Moustakas (Douglass & Moustakas, 1985; Moustakas, 1967, 1990) and the thinking of John Wood (2000). These theorists from the Social Sciences and Design emphasise a comparatively introspective and subjective heuristic encounter, where value is placed on personal experiences, intuition, flexibility, and emotional engagement. This approach underlines the importance of the self in the research process, where the artistic scholar is both a participant and an observer who navi- gates the tension between subjective experience and objective analysis (Ings, 2015).

The comparative analysis of heuristic inquiry in these fields reveals several similarities. Both domains engage deeply with uncertainty and ambiguity, recognising these elements as sources of innovation and understanding. This engagement is evident in the iterative nature of the processes involved, whether refining algorithms in AI based on feedback loops or in deepening understanding through reflection and synthesis in artistic research (Moussaoui & Benslimane, 2023; Najafi, 2022).

Moreover, both AI and artistic research demonstrate a preference for func- tioning outcomes over absolute precision. In AI, this is reflected in the devel- opment of heuristic models that prioritise efficiency and effectiveness, while in artistic research, it is manifested in a reliance on intuition and educated guesses in the pursuit of elegant solutions rather than absolute truth (Wood, 2000).

The chapter underscores the significance of human cognitive abilities, whether manifested through artificial intuition in AI or in the intuitive insight of a researcher. It highlights the importance of flexible, approximate, problem- solving approaches in two very different disciplines, where researchers embrace uncertainty, value iterative learning, and recognise the positive role of intuition and creativity. In shedding light on similarities and differences in how heuristic approaches are employed, the chapter anticipates continued interdisciplinary research between science and the arts, and an increasing understanding such that shared insights from diverse fields might inform and enrich each other, and result in fostering a deeper appreciation of both.

Notes

1 The difficulty of solving a particular problem is related to the computational com- plexity. This means time and memory required for the execution of a certain oper- ation. Computational complexity depends on the number of elements and is often expressed with respect to estimated upper bounds employing the Big-O notations introduced by Bachmann (1894). The point of these notations was to provide some indication of how an algorithm would function in the worst possible scenario as input sizes grow.

2 Expert systems – computer programs that mimic human expert decision-making in specific fields – use a knowledge base and heuristics to offer practical solutions (Jackson, 1986). Like humans estimating travel times with approximate guesses, these systems simplify complex problems for quicker, effective resolutions (Russell & Norvig, 2010).

3 K-means clustering is a data partitioning method that divides a dataset into k clus- ters by minimising the variance within each cluster. It was originally proposed by Stuart Lloyd in 1957 and later named ‘k-means’ by James MacQueen in 1967.

References

Alpaydin, E. (2020). Introduction to machine learning. MIT Press.

Bach, L. (2002). Heuristic scholar: Heuristic inquiry and the heuristic scholar.

Counterpoints, 183, 91–102. https://www.jstor.org/stable/42976833

Bachmann, P. (1894). Zahlentheorie: Die analytische Zahlentheorie [Number theory: The analytical number theory] (Vol. 2). Teubner.

Baldwin, E. (2023). Transcendence: An exploration of collage as a voice for non-binary identity [Master’s thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/16581

Ball, J. (2021). Collage as queer methodology: The pleasures and politics and queer photographic representations [Doctoral dissertation, Curtin University]. espace. https://hdl.handle.net/20.500.11937/83186

Barrett, E., & Bolt, B. (2014). Practice as research: Approaches to creative arts enquiry. Bloomsbury Publishing.

Binar, K., Imah, E., Maureen, I., Rakhmawati, L., & Saphira, H. (2023). Trend and vis- ualisation of artificial intelligence research in the last 10 years. TEM Journal, 12(2), 918–927. https://doi.org/10.18421/TEM122-38

Burton, R. (2015). A creative consideration of climate adaptation as a social and eco- logical palimpsest [Master’s thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/9138

Cacchiani, V., Iori, M., Locatelli, A., & Martello, S. (2022). Knapsack problems – An overview of recent advances. Part II: Multiple, multidimensional, and quadratic knapsack problems. Computers & Operations Research, 143, 105693. https://doi. org/10.1016/j.cor.2021.105693

Chooi, D. (2017). ‘Homebound’: The illustrated graphic novel as an autobiographic voice for an immigrant Asian gay male in New Zealand [Master’s thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https:// hdl.handle.net/10292/10551

Cioffi, R., Travaglioni, M., Piscitelli, G., Petrillo, A., & De Felice, F. (2020). Artificial intelligence and machine learning applications in smart production: Progress, trends, and directions. Sustainability, 12(2), 492. https://doi.org/10.3390/su12020492

Conlan, D. (2000). Heuristic research: With thanks and apologies to Clark Moustakas. In P. Willis, R. Smith, & E. Collins (Eds.), Being, seeking, telling: Expressive approaches to qualitative adult education research (pp. 12–31). Post Pressed.

Cousins, G., Humphris, A., Kingi, A., Oram, L., & Tudor, K. (2024). Heuristic research in a time-limited context. In W. Ings & K. Tudor (Eds.), Heuristic enquiries: Research across disciplines and professions (pp. xx–xx). Routledge.

Crevier, D. (1993). AI: The tumultuous history of the search for artificial intelligence. Basic Books. https://doi.org/10.5555/151188

Doshi-Velez, F., & Kim, B. (2017). Towards a rigorous science of interpretable machine learning (arXiv:1702.08608). arXiv. https://arxiv.org/abs/1702.08608

Douglass, B. G., & Moustakas, C. (1985). Heuristic inquiry: The internal search to know. Journal of Humanistic Psychology, 25(3), 39–55. https://doi.org/10.1177/ 0022167885253004

Dridi, S. (2021). Supervised learning – A systematic literature review. https://files.osf. io/v1/resources/tysr4/providers/osfstorage/624a442a7a7d9e04500608ce?action=d ownload&direct&version=3

Dube, T., Raines, R., Peterson, G., Bauer, K., Grimaila, M., & Rogers, S. (2012). Malware target recognition via static heuristics. Computers & Security, 31(1), 137, https://doi.org/10.1016/j.cose.2011.09.002

Duncan, M. (2004). Autoethnography: Critical appreciation of an emerging art. International Journal of Qualitative Methods, 3(4), 28–39. https://doi.org/10.1177/ 160940690400300403

Eisold, K. (2000). The rediscovery of the unknown: An inquiry into psychoanalytic praxis. Contemporary Psychoanalysis. 36(1), 57–75. https://doi.org/10.1080/00107530.2000.10747045

Epstein, S. (2010). Demystifying intuition: What it is, what it does, and how it does it.

Psychological Inquiry, 21(4), 295–312. https://doi.org/10.1080/1047840X.2010. 523875

Ete, I. (2021). Naatapuitea: An artistic interpretation of traditional and contemporary Samoan musical structures, instrumentation and koniseti [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https:// hdl.handle.net/10292/14792

Flasinśki, M. (2016). Introduction to artificial intelligence. Springer. https://doi. org/10.1007/978-3-319-40022-8

Gigerenzer, G., & Gaissmaier, W. (2011). Heuristic decision making. Annual Review of Psychology, 62(1), 451–482. https://doi.org/10.1146/annurev-psych-120709-145346

Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press. Griffiths, M. (2010). Research and the self. In M. Biggs & H. Karlsson (Eds.), The Routledge companion to research in the arts (pp. 167–185). Routledge.

Guidotti, R., Monreale, A., Matwin, S., & Pedreschi, D. (2020). Black box explanation by learning image exemplars in the latent feature space. In U. Brefeld, E. Fromont,

Hotho, A. Knobbe, M. Maathuis, & C. Robardet (Eds.), Machine learning and knowledge discovery in databases (Vol. 11906, pp. 189–205). Springer. https://doi.org/10.1007/9783-030-46150-8_12

Halberstam, J. (2018, October). Unbuilding gender. Places Journal. https://doi.org/10.22269/181003

Hamilton, J. (2011). The voices of the exegesis. In F. L. Justice & K. Friedman (Eds.), Pre-Conference Proceedings of Practice, Knowledge, Vision: Doctoral Education in Design Conference (pp. 340–343). School of Design, Hong Kong Polytechnic University. https://eprints.edu.au/41832/

Harguess, J., & Ward, C. M. (2022). Is the next winter coming for AI? Elements of mak- ing secure and robust AI. 2022 IEEE Applied Imagery Pattern Recognition Workshop (AIPR), 1–7. https://ieeexplore.ieee.org/abstract/document/10092230/

Hoyle, E. (2023). Untitled grief: An application of heuristic inquiry. Rangahau Aranga: AUT Graduate Review, 2(3). https://doi.org/10.24135/rangahau-aranga.v2i3.208

Hruschka, H. (2021). Comparing unsupervised probabilistic machine learning methods for market basket analysis. Review of Managerial Science, 15(2), 497–527. https://doi.org/10.1007/s11846-019-00349-0

Ings, W. (2006). Talking pictures: The creative utilisation of structural and aesthetic profiles from narrative music videos and television commercials in non-spoken film texts [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/346

Ings, W. (2011). Managing heuristics as a method of inquiry in autobiographical graphic design theses. International Journal of Art & Design Education, 30(2), 226-242, https://doi.org/10.1111/j.1476-8070.2011.01699.x

Ings, W. (2014). Embodied drawing. A case study in narrative design. Artifact: Journal of Design Practice, 3(2), 2.1–2.10. https://intellectdiscover.com/content/journals/10.14434/artifact.v3i2.3983/art.3.2.2.1_

Ings, W. (2015). The authored voice: Emerging approaches to exegesis design in crea- tive practice PhDs. Educational Philosophy and Theory, 47(12), 1277–1290. https://doi.org/10.1080/00131857.2014.974017

Ings, W. (2018). Heuristic inquiry, land and the artistic researcher. In M. Sierra & Wise (Eds.), Transformative pedagogies and the environment: Creative agency through contemporary art and design (pp. 55–80). Common Ground Research Networks.

Jackson, P. (1986). Introduction to expert systems. Addison-Wesley.

James, G., Witten, D., Hastie, T., & Tibshirani, R. (2021). An introduction to statistical learning: With applications in R. Springer. https://doi.org/10.1007/978-1-0716-1418-1

Johnny, O., Trovati, M., & Ray, J. (2020). Towards a computational model of artificial intuition and decision making. In L. Barolli, H. Nishino, & H. Miwa (Eds.), Advances in intelligent networking and collaborative systems (Vol. 1035, pp. 463–472). Springer. https://doi.org/10.1007/978-3-030-29035-1_45

Jordan, M. I., & Mitchell, T. M. (2015). Machine learning: Trends, perspectives, and prospects. Science, 349(6245), 255–260. https://doi.org/10.1126/science.aaa8415 Jünger, M., Reinelt, G., & Rinaldi, G. (1995). The traveling salesman problem.

Handbooks in Operations Research and Management Science, 7, 225–330.

Comparative study of heuristic inquiry in AI and artistic research  155

Keats, J. (1895). The letters of John Keats. Reeves & Turner.

Klein, J. (2010). What is artistic research? https://www.researchcatalogue.net/ view/15292/15293

Kleining, G. & Witt, H. (2000). The qualitative heuristic approach: A methodology for discovery in psychology and the social sciences. Rediscovering the method of introspection as an example. Forum Qualitative Sozialforschung/Forum: Qualitative Social Research, 1(1). https://doi.org/10.17169/fqs-1.1.1123

Kuhn, M., & Johnson, K. (2013). Applied predictive modeling. Springer. https://doi. org/10.1007/978-1-4614-6849-3

LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444 https://doi.org/10.1038/nature14539

Lesage, D. (2009). Who’s afraid of artistic research? Art & Research, 2(2), 1–10.

Mansingh, G., Osei-Bryson, K.-M., Rao, L., & McNaughton, M. (2016). Data preparation: Art or science? 2016 International Conference on Data Science and Engineering (ICDSE), 1–6. https://ieeexplore.ieee.org/abstract/document/7823936/

Matai, R., Singh, S. P., & Mittal, M. L. (2010). Traveling salesman problem: An over- view of applications, formulations, and solution approaches. In D. Davendra (Ed.), Traveling Salesman Problem, Theory and Applications, 1(1), 1–25. https://doi.org/10.5772/12909

Merenda, M., Porcaro, C., & Iero, D. (2020). Edge machine learning for AI-enabled IOT devices: A review. Sensors, 20(9), 2533. https://doi.org/10.3390/s20092533

Miller, L. (2013). The visceral metaphor: A contemplation on the invaded self [Master’s thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/6731

Mortessen Steagall, M. (2019). The process of immersive photography: Beyond the cognitive and the physical [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/12251 Moussaoui,  H.,  &  Benslimane,  M.  (2023).  Reinforcement  learning:  A  review.

International Journal of Computing and Digital Systems, 13(1), 1465–1483. https:// doi.org/10.12785/ijcds/1301118

Moustakas, C. (1967). Heuristic research. In J. Bugental (Ed.), Challenges in humanistic psychology (pp. 100–107). McGraw-Hill.

Moustakas, C. (1990). Heuristic research: Design, methodology, and applications. Sage.

Myers, D. G. (2002). Intuition: Its powers and perils. Yale University Press.

Najafi, H. (2022). Persian illuminationism as a heuristic methodology in creative practice-led research (M. Mortensen Steagall, Trans.). Revista GEMInIS, 13(2), 36–45 https://doi.org/10.53450/2179-1465.RG.2022v13i2p36-45

Najafi, H. (2023). Displacement of self-continuity: An illuminative, heuristic inquiry into identity transition in an animated short film [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/15784

Newell, A., & Simon, H. A. (1972). Human problem solving (Vol. 104). Prentice-Hall.

Panaita, E. (2018). An exegetical consideration of upgraded gods [Master’s thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/11510

Paora, T. I. (2023). Takatap̄ui – Beyond marginalisation: Exploring Maōri gender, iden- tity and performance [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/16962

Peacock, F. (2023). A heuristic inquiry into my use of Theraplay® with children expe- riencing the impact of relational and developmental trauma. European Journal for Qualitative Research in Psychotherapy, 13. https://www.ejqrp.org/index.php/ejqrp/article/view/200

Polanyi, M. (1966). The tacit dimension. Garden City, NY: Doubleday.

Provost, F., & Fawcett, T. (2013). Data science for business: What you need to know about data mining and data-analytic thinking. O’Reilly Media.

Russell, S. J., & Norvig, P. (2010). Artificial intelligence: A modern approach. London.

Sarker, I. H. (2022). AI-based modeling: Techniques, applications and research issues towards automation, intelligent and smart systems. SN Computer Science, 3(2) https://doi.org/10.1007/s42979-022-01043-x

Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural Networks, 61, 85–117. https://doi.org/10.1016/j.neunet.2014.09.003

Schuster, P. (2000). Taming combinatorial explosion. Proceedings of the National Academy of Sciences, 97(14), 7678–7680. https://doi.org/10.1073/pnas.150237097

Scrivener, S. (2000). Reflection in and on action and practice in creative production doctoral projects in art and design. Working Papers in Art and Design 1. https://www.herts.ac.uk/ data/assets/pdf_file/0014/12281/WPIAAD_vol1_scrivener.pdf

Sela-Smith, S. (2002). Heuristic research: A review and critique of Moustakas’s method.

Journal of Humanistic Psychology, 42(3), 53–88. https://doi.org/10.1177/00267802042003004

Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 99–118. https://doi.org/10.2307/1884852

Simon, H. A. (1979). Rational decision making in business organisations. The American Economic Review, 69(4), 493–513. https://www.jstor.org/stable/1808698

Simon, H. A., & Newell, A. (1958). Heuristic problem solving: The next advance in operations research. Operations Research, 6(1), 1–10. https://doi.org/10.1287/opre.6.1.1

Sinfield, D. (2020). Typographical voices: Poetic reflections on the Pat̄ea Freezing Works [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/13601

Snyder, L. J. (2012). Qualitative research ethics: An heuristic inquiry exploring the meaning and application of ethics in qualitative research [Doctoral thesis, University of Calgary]. PRISM. https://hdl.handle.net/1880/51588

Tavares, T. (2011). Carnival land: A performance of metaphors [Master’s thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/1318

Tavares, T. (2019). Paradoxical realities: A creative consideration of realismo maravil- hoso in an interactive digital narrative [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle. net/10292/12958

Ventling, F. D. (2017). Illuminativa – The resonance of the unseen [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/7037

Ventling, F. D. (2018). Heuristics – A framework to clarify practice-led research. DAT Journal, 3(2), 122–156. https://doi.org/10.29147/dat.v3i2.88

Williams, T. (2024). Tangohia mai te taura – Take This Rope: Exploring Maōri documentary-making approaches to elevate whan̄au narratives [Doctoral thesis, Auckland University of Technology]. Tuwhera Open Access Theses & Dissertations. https://hdl.handle.net/10292/17323

Wilt, C., Thayer, J., & Ruml, W. (2010). A comparison of greedy search algorithms. Proceedings of the International Symposium on Combinatorial Search, 1(1), 129–136 https://doi.org/10.1609/socs.v1i1.18182

Wood, J. (2000). The culture of academic rigour: Does design research really need it?

The Design Journal, 3(1), 44–57. https://doi.org/10.2752/146069200789393599 Wu, Y., Song, W., Cao, Z., Zhang, J., & Lim, A. (2021). Learning improvement heuristics for solving routing problems. IEEE Transactions on Neural Networks and Learning  Systems,33(9),  5057–5069.  https://doi.org/10.1109/TNNLS.2021.3068828

Ye, Y., Li, T., Adjeroh, D., & Iyengar, S. S. (2018). A survey on malware detection using data mining techniques. ACM Computing Surveys, 50(3), 1–40. https://doi.org/10.1145/3073559

Yilmaz, S., Daly, S. R., Seifert, C. M., & Gonzalez, R. (2016). Evidence-based design heuristics for idea generation. Design Studies, 46, 95–124. https://doi.org/10.1016/j.destud.2016.05.001

Yilmaz, S., & Seifert, C. M. (2011). Creativity through design heuristics: A case study of expert product design. Design Studies, 32(4), 384–415. https://doi.org/10.1016/j.destud.2011.01.003

Zadeh, L. A. (1988). Fuzzy logic. Computer, 21(4), 83–93. https://doi.org/10.1109/2.53 Zheng, A., & Casari, A. (2018). Feature engineering for machine learning: Principles and techniques for data scientists. O’Reilly Media.