On the (limited) use of touchscreen-based behavioural and cognitive research with dogs: potential causes and future directions

Abstract

Dogs (Canis familiaris) are an increasingly popular model in comparative cognition research. Traditional research paradigms often rely on heavy experimenter involvement, which can introduce biases and inconsistencies. Touchscreen-based automated systems offer a solution by enhancing standardization, reducing experimenter effects, and improving data quality, and have been widely adopted for research with diverse species. However, their adoption in dog research has remained underexplored. This systematic review quantified dog behaviour and cognition literature in relation to touchscreen-based methodologies and evaluated existing touchscreen-based approaches. Our search confirmed that while the field of dog cognition continues to grow, touchscreen-based studies are exceptionally rare, with only fourteen such publications identified for the entire indexed record. We explored the potential reasons for this limited adoption by categorizing them into three main barriers: technical, such as the need for interdisciplinary skills; practical, including high costs and lengthy training requirements and Umwelt, relating to the dogs’ species-specific constraints. Finally, we propose strategies to overcome these obstacles, including the use of open-source solutions, establishing multi-lab collaborations, and designing interfaces better aligned with dogs’ Umwelt. We conclude that despite being underutilized, touchscreen-based methods hold significant promise for advancing canine research. By addressing current challenges through collaborative, open, and dog-centred practices, the field can integrate touchscreen-based methods more widely, enhance reproducibility, accelerate discovery, and obtain objective indicators of dogs’ behavioural and cognitive capabilities.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10071-026-02055-3.

Introduction

Dogs (Canis familiaris) have become a favoured model in comparative research (Aria et al. 2021; Bensky et al. 2013). A bibliometric analysis showed an increase in publication of dog behavioural and cognitive articles since 2000, with an accelerating trend from 2010 onwards that was considerably steeper than the growth observed in animal cognition and behaviour research in general (Aria et al. 2021). Several factors might have contributed to this increase, including dogs’ accessibility as companion animals, their trainability, and the relatively low cost of recruiting pet dogs compared to maintaining laboratory colonies, making them an attractive and practical model for cognitive and behavioural research (Elliott et al. 2017; L. S. Hall and Boxall 2024).

Traditionally, research on dog behaviour and cognition has relied heavily on paradigms that require extensive involvement from a human experimenter. In these tasks, the experimenter presents stimuli, initiates trials, records the dogs’ choices and then dispenses (or withholds) rewards. For example, working-memory paradigms, require the experimenters to place a treat, impose a delay, and later invite the dog to engage (Craig et al. 2012; Fiset et al. 2003; Head et al. 1995; Krichbaum et al. 2021; H. C. Miller et al. 2009). Similarly, discrimination tasks depend on people to arrange and present various physical or sensory cues (e.g., colours, shapes, quantities, odours, or sounds), and record dogs’ choices by hand, either live through an external observer or retrospectively via video (Miletto Petrazzini and Wynne 2016; Milgram 2003; Milgram et al. 1994; Mongillo et al. 2017; Rivas-Blanco et al. 2024; Tanaka et al. 2000; Tapp et al. 2003). Reward contingencies (e.g., reinforcing correct responses and withholding or changing rewards for incorrect responses) are also frequently manually controlled (Ashton and De Lillo 2011; Handley et al. 2023; Milgram 2003; Yin et al. 2008).

Reliance on human operators, however, introduces potential inconsistencies and biases. Inconsistencies in stimulus placement, inter-trial intervals or reward delivery delays occur across trials, days, and experimenters, and can potentially impact dog performance. Measurement and judgement biases can arise when observers code behaviour differently across sessions or individuals, expectancy bias can lead experimenters to (consciously or unconsciously) signal desired responses, observer biases may cause selective observation or misinterpretation of behaviour, and inter-rater reliability problems can further inflate variance (Clark et al. 2020; Holzbach 1978; Johnen et al. 2017; Reynolds et al. 2021; Schmidt and Hunter 1996; Tuyttens et al. 2014).

Recognising these issues, researchers have launched multi-lab collaborations across a variety of model species, such as the ‘ManyDogs’ project (ManyDogset al. 2021) aiming to reduce variation across different research sites by coordinating methodologies while increasing statistical power through larger combined sample sizes (Alessandroni et al. 2025). These initiatives mark an important progress in addressing replication concerns by increasing the number of subjects and replications, which can help buffer site-specific deviations and can even out local inconsistencies. However, such large-scale collaborations are not feasible for all research questions or lab infrastructures. For example, studies requiring specialised, high-cost equipment, such as fMRI (Karl et al. 2020; Prichard et al. 2021), are not accessible or affordable for many laboratories.

Alternatively, many researchers have begun adopting technology-enhanced approaches to mitigate methodological inconsistencies and potential biases. For example, markerless video-based tools such as DeepLabCut or SLEAP have been used to automate pose extraction and behaviour annotation, therefore reducing the need for manual scoring and enhancing consistency across sessions and observers (Lauer et al. 2022; Mathis et al. 2018; Nath et al. 2019; Pereira et al. 2022). App-based platforms like ZooMonitor allow digitalised data-entry, enabling automated reliability tests to check observer consistency (Wark et al. 2019). There have also been monitor-based studies where presentation of the stimuli is standardised, sometimes combined with eye-tracking technologies (e.g., Barber et al. 2016; Mongillo et al. 2021; Somppi et al. 2012; Völter et al. 2020). These tools provide more standardisation over manual procedures, and each have distinct strengths, but in terms of automation, they typically target individual components of the study (standardisation of trial contingency, presentation of stimulus, detecting or recording of behaviour) rather than integrating the entire process in a closed-loop fashion.

Touchscreen-based systems, however, represent a more comprehensive approach: they unify stimulus control, standardised task control, direct response logging, automated reward delivery, and trial-wise data logging inside a single apparatus, supporting cross-site standardisation and rapid parameterisation of tasks (Dumont et al. 2020; Egelkamp and Ross 2019; Kangas and Bergman 2017; Orphanides and Nam 2017; Seitz et al. 2021). The closed loop between the stimuli presentation and animals’ responses enables animals to receive immediate feedback based on their interactions with the touchscreen, which in turn guides their behaviour (Ajuwon et al. 2023, 2024a, b; Kane et al. 2020; Lopes and Monteiro 2021; Sun et al. 2025). These systems also support real-time dynamic adjustments to stimuli and task events; automated reward systems deliver rewards precisely within (and across) trials; and scripted (whether fixed, dynamic, or random) inter-trial intervals remove potential timing errors that human experimenters inevitably introduce (Degrande et al. 2022; Lee and Spence 2008; Marquardt et al. 2017; Odland et al. 2021). Their use allows for standardised procedures and reduces experimenter variability in studies where tasks can be translated into screen-based interactions. For certain studies, especially those that require high temporal or spatial precision in task control, automated, bias-resistant methods, such as touchscreen-based setups, offer a practical alternative to reduce operator-dependent variabilities. We acknowledge that full automation is not always necessary or beneficial for every experimental design or research questions, and can still introduce design-related biases, such as those related to stimulus presentation, interface configuration, or reinforcement schedules, which must be carefully controlled, but in the wider context of the reproducibility crisis challenging behavioural science (Johnen et al. 2017; Open Science Collaboration 2015; Spanagel 2022; Wilson et al. 2023), such methods can contribute to enhanced control, consistency, and standardisation.

More specifically, direct and real-time data collected by touchscreen-based systems improves data quality and throughput. Every screen touch is recorded instantly, producing datasets with higher accuracy and lower susceptibility to bias, while reducing delays, expectancy effects, and measurement errors associated with manual procedures (Holman et al. 2015; Hoyt 2000; Michelson et al. 1985; L. E. Miller and Stewart 2011; Rosenthal 1994). Since task progression is determined solely by the subjects’ responses, animals proceed through trials at their own pace without waiting for human input. This design minimises the testing time that would otherwise be lost to operator-related delays, such as manually changing stimuli or re-engaging inattentive subjects. Fully programmed protocols also allow the apparatus to run without continuous human supervision, even inside animals’ home environments, supporting flexible testing while reducing demands on researchers and subjects (Butler and Kennerley 2019; Cabrera-Moreno et al. 2022; Fizet et al. 2017; Huskisson et al. 2021; Nasrini and Hampton 2024).

Integrated timestamping enables synchronisation across behavioural and physiological data streams. We note that precise temporal data are not unique to touchscreen systems. Most automated closed-loop systems as well as many automated computer vision and machine learning approaches can also generate high resolution timestamped data (e.g., Mathis et al. 2018; Pereira et al. 2022; Romero-Ferrero et al. 2019). Within animal behaviour and cognition research, however, many tasks still rely on manual control and coding; in this context, touchscreen paradigms offer a practical and standardised way to automatically log a structured set of task events (e.g., stimulus onset/offset, touches, latencies, inter-response intervals, reward delivery, timeouts) with minimal dependence on real time human annotation. These machine readable events can be used as basis for aligning concurrent video, audio, motion tracking, heart rate, EEG, or other data streams, yielding rich multimodal records that extend beyond aggregate accuracy measures and support modern pattern-mining and modelling approaches (e.g., Gupta 2024; Li et al. 2011; Menaker et al. 2021; Mluba et al. 2024; Saad Saoud et al. 2024; Ye et al. 2023). Collectively, the advantages of touchscreen-based methods respond to current calls for greater rigour and transparency in animal behaviour research (Spanagel 2022; Wilson et al. 2023).

Given these practical advantages of efficiency and flexibility, it is not surprising that touchscreen-based methodologies have been adopted across diverse species, such as gorillas (e.g., Hopper et al. 2021; Truax and Vonk 2023; Vonk 2024; Vonk et al. 2022), chimpanzees (e.g., Gao and Adachi 2024; Huskisson et al. 2020; McEwen et al. 2025; Muramatsu and Matsuzawa 2023; Sato et al. 2023), orangutans (e.g., Gazes et al. 2017; Perdue et al. 2012; Renner et al. 2016; Scheel 2018), capuchin monkeys (e.g., Malassis and Seed 2023; Mendes et al. 2024; Renner et al. 2021), bears (e.g., Bernstein-Kurtycz et al. 2024; Perdue 2016), rats (e.g., Crijns & Op Beeck, 2019), mice (Attalla et al. 2024; Chasse et al. 2023; Palmer et al. 2021), goats (e.g., Gao et al. 2025; Langbein et al. 2023), carrion crows (e.g., O’Hara et al. 2017), pigeons (Blaisdell et al. 2018; Huber et al. 2005; Spetch et al. 1992; Toegel et al. 2021; Wasserman et al. 2013), chickens (e.g., Nasrini and Hampton 2024), and tortoises (e.g., Mueller-Paul et al. 2014). Importantly, their use has not only spanned diverse taxa but has also been reported to be increasingly prevalent in rodents (Dumont et al. 2021) and in nonhuman primates (Egelkamp and Ross 2019) over the past few decades, reflecting both the method’s adaptability and its growing role in comparative cognition (Seitz et al. 2021).

Despite this broad adoption, it remains unclear whether similar growth is occurring in dog research. To address this, the present review had four aims: (1) to investigate whether the upward trajectory of dog behaviour and cognition research continues; (2) to assess whether touchscreen use in dogs is increasing in parallel with both the general dog literature and with touchscreen-based literature using other model species; and (3) to summarise and critically evaluate the main methodological approaches reported in previous touchscreen-based dog studies; and finally, (4) to speculate and discuss the challenges and species-specific considerations related to touchscreen-based research with dogs based on existing evidence.

Methods

We conducted two literature searches in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to ensure that our search and article selection process were transparent and replicable (Liberati et al. 2009; Group et al. 2015). The first search, designed to extend the bibliometric trends reported by Aria et al. (2021), targeted general dog behaviour and cognition literature published between 2019 and 2024, thereby updating overall publication trajectories beyond their coverage period. The second search specifically targeted touchscreen-based studies involving dogs across the full available record. The protocol included the following steps: identification (defining search strategies and exclusion criterion), screening and eligibility (selection process), and final inclusion.

Recent dog cognitive and behavioural studies search

Search strategy

We defined the eligibility criteria following the protocol established by Aria and colleagues (2021) to identify dog behaviour and cognition studies, extending the timeframe of the original study beyond 2018 to include all relevant articles published up to (and including) December 2024. The search was conducted on Scopus abstract and citation database (Elsevier; https://www.scopus.com; search date: 5th January 2025; search strategy: (((TITLE-ABS-KEY (((dog OR dogs) AND cogniti) OR (canis AND familiaris AND cogniti))) OR (TITLE-ABS-KEY (((dog OR dogs) AND communicat) OR (canis AND familiaris AND communicat))) OR (TITLE-ABS-KEY (((dog OR dogs) AND behav*) OR (canis AND familiaris AND behav*))))) AND PUBYEAR > 2018 AND PUBYEAR < 2025 AND (LIMIT-TO (LANGUAGE, “English”)) AND (LIMIT-TO (SRCTYPE, “j”)) AND (LIMIT-TO (PUBSTAGE, “final”)) AND (LIMIT-TO (DOCTYPE, “ar”)); retrieved record: 4673 articles).

Aria and colleagues (2021) defined cognitive and behavioural research domains in dogs, and we adopted these domains as our baseline eligibility criteria. To ensure comprehensive coverage, we also interpreted their existing topics and introduced additional topics where necessary, which are marked with an asterisk (*) in Table 1. This step was essential because our search strategy retrieved not only cognitive and behavioural studies but also medical, physiological, and clinical research without direct relevance to cognition or behaviour. Therefore, the inclusion topics outlined below served as a content-based criterion for classifying eligible studies. We included original, peer-reviewed journal articles in English reporting experimental studies on domestic dogs (Canis familiaris).

Table 1.

Topics included in general dog cognitive and behavioural studies (in alphabetical order). * denotes topics newly added or interpreted by us.

(adapted from Aria et al. 2021).

Aging (outcomes) as a natural cognitive decline of behavioral

Attention

Behavioral ecology (predation, scavenger behavior, roaming etc.)*

Cranial MRI studies*

Development of new methodology (when using behavioural or cognitive data; such as machine learning algorithms to detect behaviours)*

Domestication (when they deal with cognitive evolution)

Drug treatments or physical therapy (in the case untreated controls are involved; e.g., oxytocin)

Emotions

fMRI studies

Lateralisation

Learning

Memory

Numerical abilities

Perception

Reasoning and problem solving

Spatial cognition

Studies with questionnaires (unless only aimed to study aggression-related behavioural problems, such as biting other dogs)*

Temperament and personality

Theory of mind

Training and working dogs (including assistant and therapy dogs, unless the target is the human benefit)

Welfare and stress in general

Exclusion criteria

We excluded review articles and conference proceedings. We excluded studies that focused on other canid species (e.g., African wild dogs, Lycaon pictus).

Selection process and data collection

The resulting retrieved records were exported in BibTeX format and transferred into Microsoft Excel for further screening to check for inclusion to the dataset. During this stage, articles were divided and reviewed independently by three selectors according to established eligibility criteria for dog behaviour and cognition studies. Any uncertainties or disagreements among selectors were flagged, discussed, and resolved through consensus. Articles meeting the criteria were included in the final dataset (See Supplementary Information and Fig. 1 for a visual depiction of the protocol and the number of articles included/excluded in each stage. Reasons for exclusion for each article can be found in Supplementary Information. Finally, we counted the number of articles in the final dataset per year using basic Excel functions.

Fig. 1.

PRISMA flow diagram showing articles on dog behaviour and cognition included and excluded from this review. The diagram summarises the selection procedure for this review following PRISMA guidelines (Group et al. 2015). Identification: A total of 4,673 records were identified through Scopus search. Screening: 3,540 records were excluded. Included: 1,133 studies were included in the count-per-year analysis

Dog touchscreen-based studies

Search strategy

We conducted a second search targeting touchscreen-based dog studies. The search was also conducted on Scopus abstract and citation database (Elsevier; https://www.scopus.com; search date: 5th January 2025; search strategy: TITLE-ABS-KEY(((dog OR dogs) AND (“*screen” OR screen OR (touch W/2 screen))).

OR ((canis AND familiaris) AND (“*screen” OR screen OR (touch W/2 screen)))); retrieved record: 8 articles). As in the previous search, we included original, peer-reviewed journal articles in English reporting experimental studies on domestic dogs (Canis familiaris).

In contrast to our search for recent dog cognitive and behavioural studies, the Scopus search for touchscreen-based studies using domestic dogs as subject species only returned eight articles. Aiming to capture the breadth of this relatively new and narrowly focused topic, we conducted backward (by checking the reference list of retrieved studies) and forward citation chasing (by checking newer studies that have cited the retrieved articles), also referred to as snowball sampling, to identify additional touchscreen studies (Badampudi et al. 2015; Booth 2016; Higgins and Thomas 2019).

Exclusion criteria

We excluded review articles and conference proceedings. We excluded studies that focused on other canid species (e.g., African wild dogs, Lycaon pictus). We also excluded any publications that used alternative controllers (e.g., joysticks, buttons, handles) rather than direct screen touches.

Selection process

The resulting retrieved records were exported in BibTeX format and transferred into Microsoft Excel for further screening to check for inclusion to the dataset. Articles meeting the criteria were included in the final dataset. The protocol used for this search and articles included in each stage were illustrated in Fig. 2.

Fig. 2.

PRISMA flow diagram showing articles using touchscreen-based methods with dogs included and excluded from this review. The diagram summarises the selection procedure for this review following PRISMA guidelines (Group et al. 2015). Identification: A total of eight records were retrieved from database searches. Screening: four were excluded, and ten additional records were identified through snowballing. Included: fourteen studies met the inclusion criteria and were included in the final data extraction

Data collection

We counted the number of articles in the final dataset per year using basic Excel functions. Methodological details and results from the final dataset were manually extracted: authors; year of publication; type of touchscreen; recruited subject number; included subject number; dropped-out subject number (i.e., dogs lost for reasons outside experimental control, such as owner unavailability, dog illness, or death); excluded subject number (i.e., dogs that failed to learn the task, showed persistently low motivation, or disengaged during training); type of stimulus; training schedule; training frequency; training duration; pre-training required (session number); training required (session number); subject population (pet or lab dog); and previous touchscreen experience. We also calculated meta-results: inclusion rate (included subject number/recruited subject number), and pre-training and training time (sessions).

Results and discussion

The comparison between the number of publications on dog behaviour and cognition and touchscreen-based dog studies highlights the disparity between the two (Fig. 3). While the general field of dog behaviour and cognition has shown a continuous growth, with hundreds of articles being published every year, touchscreen-based dog studies remain very limited in number — there have only been fourteen touchscreen-based articles since 2007 (Fig. 3, inset and Table 2). Importantly, our search strategy for touchscreen dog studies was deliberately more inclusive than the general literature count, extending beyond database search to citation chasing, because the initial database search returned very few records. This asymmetry in method supports our observation that even after applying broader and more comprehensive retrieval procedures designed to increase coverage, the touchscreen dog literature remains limited relative to the rapidly expanding general dog behaviour and cognition field.

Fig. 3.

Number of dog cognitive and behavioural studies show a sustained and increasing trend. Grey and black lines depict the data reported in Aria et al. (2021; data covering 1985–2018) and from our search (covering 2019–2024), respectively. 1133 articles were included in the final dataset for recent dog cognitive and behavioural studies. The complete list of included articles can be found in Supplementary Information. Inset shows dog studies that have used touchscreen-based methods. Created using Matlab 2024b

Table 2.

Methodological and training details for dog touchscreen-based studies

* denotes data calculated by us based on original raw data. Pre-training generally refers to familiarising dogs with the experimental apparatus or basic task demands (e.g., touching a given stimuli or touching the screen). Training typically involves teaching dogs to distinguish between two stimuli (e.g., shapes, faces, or quantities), with correct choices reinforced. Terminology varies between studies. We contacted corresponding authors for additional data not reported in the original manuscripts. Cells marked as N/A indicate information that was either no longer accessible or for which no reply was received by the time of submission

These results show that the number of publications in dog behaviour and cognition continues to rise, reflecting the growing interest in using dogs as a model for these lines of research, consistent with what has been reported previously (Aria et al. 2021), supporting our first hypothesis. However, contrary to our expectations, our search revealed only a very limited number of studies that used touchscreen-based methodologies with dogs. This finding indicates that, despite the increasing number of dog studies using other methods, touchscreen-based approaches remain underutilised.

In the sections that follow, we discuss a few insights from the fourteen touchscreen-based dog studies included in our review. We then speculate on potential barriers to the wider adoption of touchscreen-based methods in dog research and consider their methodological limitations. Finally, we propose strategies to mitigate these challenges and offer recommendations for future studies intending to incorporate touchscreen-based approaches.

Existing touchscreen-based dog studies

Among the fourteen publications of touchscreen-based studies using domestic dogs, several overarching principles emerged. To start, multi-stage training is universal. The majority of studies described incremental training programmes in which dogs first learned to orient toward the apparatus, then to touch a single stimulus, and only subsequently asked to discriminate among alternative options. Training schedules ranged from a few sessions (Aust et al. 2008; Pitteri et al. 2014), to multiple visits per week (e.g., Laude et al. 2016), to extended training regimes exceeding more than one hundred sessions (Lonardo et al. 2022), suggesting that reliable screen interaction is rarely achieved in a short time frame (Table 2). The fact that most studies required lengthy multi-stage training highlights both the critical role of training for experimental success and the substantial investment of time and personnel it demands. These demands become even more pronounced when the subjects are pet dogs, who live in private homes with owners, and it is simply impractical to test them with the same frequency as lab animals, which we will further discuss below.

Apparatuses’ set-ups fall into two general categories: open-design and constrained-design (Fig. 4). In open-design apparatuses, the entire touchscreen is accessible to the dog (e.g. Siniscalchi et al. 2023; Wallis et al. 2016) allowing flexible placement and real-time relocation of multiple stimuli, which facilitates paradigms that require more than two response options or spatially defined metrics (Fig. 4a). In constrained design apparatuses the touchscreen is placed behind physical barriers such as metal frames or wooden boards (Fig. 4b), exposing a pair of response windows (Müller et al. 2015; Pitteri et al. 2014; Rivas-Blanco et al. 2020). By limiting access to task-relevant areas, this configuration inherently prevents interactions with irrelevant screen areas, thereby simplifying the task into binary response structure and reducing accidental irrelevant touches. Therefore, the design of the apparatus should be selected based on the research question under inquiry. Researchers interested in multi-choice arrangements, dynamically shifting stimuli (e.g., moving patterns), or flexible spatial analysis should benefit from an open design. In contrast, projects that only require binary discriminations or choices could employ the constrained design.

Fig. 4.

Comparison of open-design and constrained-design touchscreen apparatuses used in dog studies. (a) Open-design set-up in which the full touchscreen is accessible to the dogs (adapted from Wallis et al. 2016); (b) Constrained-set-up in which the touchscreen is placed behind physical barriers with only defined response windows presented to the dogs (adapted from Pitteri et al. 2014). Images used (cropped for layout) under the Creative Commons Attribution license (CC BY 4.0; https://creativecommons.org/licenses/by/4.0/)

The stimuli used across studies varied widely. Roughly half of the studies surveyed used geometric shapes, while the remaining employed colour photographs, human faces, or mixed icon sets (Table 2). While this diversity highlights the flexibility of touchscreen-based platforms, using different stimulus types could impact dog engagement, learning rates, and ultimately differently affect the underlying cognitive processes (Hirskyj-Douglas et al. 2017).

Studies also varied in terms of the level of human involvement. In some, the handler stood beside the dogs, holding them back during the inter-trial interval and releasing them when choice became available (Müller et al. 2015; see later for a discussion of potential reasons for requiring a human experimenter). While this practice can prevent undesired screen touches, it reintroduces potential experimenter cues. Even when the handler cannot see the screen or the stimuli, their physical interaction with the dog, such as the timing, firmness, or subtle hesitation in restraint and release, may inadvertently signal information that influences the animal’s choice. For example, if a dog anticipates that the experimenter’s brief hold or delayed release indicates an incorrect approach, it might adjust its behaviour accordingly, leading to choices guided by perceived social feedback rather than independent discrimination. Such unintentional cues compromise the objectivity of the data, invalidate measures like response latency, and may shift the dog’s attentional focus toward the human rather than the task itself. These factors complicate cross-site replication and limit the validity of results, as performance in such setups may no longer accurately reflect autonomous decision-making. In other studies, fully autonomous apparatus operate with no experimenter present (Siniscalchi et al. 2023), thereby minimising social influence, but requiring more robust hardware and potentially more training with the dogs to shape and maintain adequate behaviours.

Participants populations were mostly pet dogs, with one study that tested pack dogs kept in enclosures (Dale et al. 2019). Notably, more than half of the articles reported that at least some of their subjects had prior experience with touchscreen experiments. Yet only one had specific information about it (Aust et al. 2008 tested dogs that also participated in Range et al., 2007). The observation that several articles mentioned previous touchscreen experiences without citing corresponding published studies prompted us to speculate that some touchscreen-based experiments may have been conducted but never reported. This potential ‘file-drawer’ effect will be discussed further in following section.

Training schedules varied considerably across the studies surveyed. Some laboratories conducted intensive daily sessions (e.g., Dale et al. 2019; Table 2), while others scheduled training sessions more flexibly (e.g., Keep et al. 2018; Table 2). The frequency and total duration of training have been suggested to influence the rate at which animals acquire and retain learned material (Demant et al. 2011; Heinrich et al. 2020; Meyer and Ladewig 2008; Smith et al. 2025). For instance, studies have reported that for acquisition and memory in conventional obedience work, training twice per week would be optimal (Demant et al. 2011); and two short sessions separated by a five-minute break can optimise performance in touchscreen discrimination tasks (Range et al. 2008). As these recommendations are based on specific study design and limited datasets, they are suggestive rather than prescriptive. Where consistency across testing sessions is impractical, detailed reporting of inter-session intervals will enable subsequent meta-analyses to investigate how schedule structure influences results and how it interacts with other control factors (e.g., age, breed and motivational state).

Interestingly, during our search, we came across a subset of studies that used semi-automated approaches. For instance, Byosiere and colleagues tested visual perception in dogs by displaying stimuli on a monitor and automatically recording nose-touches, but task control such as sample presentations and trial contingencies remained under human control (Byosiere et al. 2017, 2018a, b, 2019a). These setups reduce some experimenter influence by standardising response logging, yet the remaining manual procedures can still introduce subtle biases and imprecisions. Nonetheless, semi-automation may be a pragmatic solution for laboratories lacking the resources or expertise to implement full touchscreen automation. Despite the constant reduction in the price of hardware components, the initial investment might still be unaffordable for many facilities, and the design choices evident in these studies offer valuable insight into incremental paths toward reduced experimenter involvement.

In addition to the journal articles included in our dataset, we note two relevant conference proceedings. Zeagler and colleagues (2014) reported that dogs’ touchscreen inputs often resemble swipes or dragging rather than discrete taps, and that infrared design can be reliable to dogs’ wet noses contacts, whereas some common alternatives (e.g., capacitive approaches) can be problematic in dog use. Byrne and colleagues (2018) demonstrated the practical feasibility of in-home touchscreen deployment by training assistance dogs to execute an emergency action on a touchscreen. They showed that dogs can learn and perform a multi-step icon sequence with high reliability in a naturalistic home setting. Taken together, these studies illustrate an incremental pathway toward reliable, dog-centred systems that can be used to study various questions, thereby informing the recommendations we outline below.

Barriers and prospective routes to wider adoption of touchscreen-based methods in dog research

The studies reviewed here provide a basis for identifying why touchscreen-based methodologies remain underutilised in dogs and for extracting actionable design and implementation guidance. Below, we organise the discussion into key barriers and potential solutions, drawing not only on the dog touchscreen literature but also on relevant work from broader dog cognition, animal–computer interaction/human-computer interaction, and touchscreen-based research in other model species.

Technical difficulties

Implementing a dog-proof touchscreen apparatus requires a combination of specialised know-how across multiple domains, including knowledge of or experience with working with the target animal species, proficiency in programming languages for task control, and some degree of engineering and/or technical skills for hardware development and troubleshooting (Ajuwon et al. 2024a, b; Kravitz and Laubach 2024). From the interface side, researchers must design dog-appropriate systems that respond effectively, to facilitate learning, which can be difficult as dogs interact with the screen with their noses and/or paws. If the apparatus fails to respond when needed, it makes task contingencies unreliable, which may cause frustration, disengagement, or slow learning. Hardware choices also involve trade-offs. For example, projected capacitive screens are susceptible to moisture, producing false or inaccurate touches from wet noses or when in contact with saliva. Protective overlays or water-resistant designs can reduce these problems, but they add cost and weight, which is relevant for laboratories that value portability (e.g., mobile or home-deployable setups).

Solutions for technical difficulties

Open-source hardware and software solutions provide affordable options for laboratories wishing to incorporate touchscreen-based tasks without reinventing apparatus design from scratch. Numerous studies have demonstrated that self-built computer-controlled apparatuses, created from combinations of custom-made (e.g., laser cut and/or 3D printed parts) and off-the-shelf components, can be successfully adapted to a variety of species (Ajuwon et al. 2024a, b; Arce and Stevens 2022; Buscher et al. 2020; Butler and Kennerley 2019; Devarakonda et al. 2015; O’Leary et al. 2018; Pineño 2013). Furthermore, the ongoing reduction in the cost of digital devices and hardware fabrication (Pearce and Qian 2022; Rayna and Striukova 2021) means that the costs associated with building touchscreen-based setups are gradually decreasing (Oxley et al. 2022). Repositories such as GitHub (https://github.com) and OpenBehavior (https://edspace.american.edu/openbehavior/) also reduce the technical learning demands for scientists with limited engineering or programming experience by providing task-control scripts (e.g., stimulus presentation and response logging), circuit diagrams, and design schematics that are re-usable across projects.

Dog researchers can draw insights and inspiration from the human-computer interaction literature. Human-computer interaction has a long history of research on how users perceive, remember, and respond to screen-based stimuli (e.g., Hutchins et al. 1985; Norman 1999; Shneiderman and Plaisant 2005). These principles stress the importance of feedback, visibility, and intuitive mapping between action and outcome, considerations that are just as relevant when dogs are the ‘users.’ For example, studies of pointing and movement (Bi et al. 2013; Fitts 1954; Parhi et al. 2006) demonstrate that target size and spacing affect the accuracy and speed of selections, suggesting that using larger, well-spaced, and visually distinct stimuli is likely to facilitate dogs’ interaction with stimuli on the screen. Similarly, research on gesture recognition and error prevention has proposed strategies that emphasis accessibility of technology and how it should and could be designed to make interactions easier for the users (e.g., Khan and Khusro 2019; Wigdor and Wixon 2011; Wobbrock et al. 2011; also see such investigation in dogs: Hirskyj-Douglas et al. 2017; Zeagler et al. 2016 [note: Zeagler et al. examined touchscreens, while Hirskyj-Douglas et al. investigated interactions towards screens in general]). Applying these insights from human-computer interaction studies to dog-specific contexts could potentially improve reliability, reduce frustration, and assist learning.

We also suggest establishing collaborative frameworks across laboratories for touchscreen-based studies in dogs to standardise and verify touchscreen design decision. Multi-lab initiatives have demonstrated that shared protocols, open materials, and transparent reporting can accelerate method refinement and improve reproducibility (Alessandroni et al. 2024; Altschul et al. 2021; Byers-Heinlein et al. 2020; Coles et al. 2022; De Moor et al. 2025; International Brain Laboratory & International Brain Laboratory 2017; Lambert et al. 2022; Lucca et al. 2025; ManyDogs et al. 2021; ManyGoats 2024; ManyPrimates et al. 2020; Rance 2024). By investing in similar multi-lab collaborative models specifically tailored to dog touchscreen-based research, the community could systematically refine experimental procedures, validate protocols within and across dog populations, and develop benchmarks for training, testing, data sharing and analyses. Such collaborative efforts could foster a culture of open data and produce shareable ‘survival guides’ that include step-by-step software and hardware build instructions and troubleshooting manuals, and can substantially reduce study timeframes, while making a more substantial contribution to the cumulative scientific progress within dog cognition research.

Practical constraints of costs, labour, and publication bias

Technical demands described above translate into significant financial and personnel costs. As successfully adopting touchscreen-based methods require interdisciplinary skills, it might require extensive training, effective collaboration among researchers from diverse fields, or attracting individuals who already possess this multifaceted skill set; potentially challenging given the increasing trend of talent migration from academia to industry roles, driven by better compensation and life quality (Shao et al. 2024; Torrisi and Pernagallo 2020; Woolston 2020).

A touchscreen-based apparatus requires initial investment. Building such apparatus normally involves (at minimum) a touch-capable monitor, an automated feeder, and a computer. Based on commercial material prices and published studies, the minimal costs for a complete set-up would be around 800 Euros, and depending on the system’s design, brand, and customization, the cost can exceed 10,000 Euros (e.g., Horner et al. 2013; Steurer et al. 2012). While this investment can be amortised across many studies (reducing cost per session over time), it is substantially higher than that of many manual experiments, which may require little more than basic materials such as pen, paper, or simple objects.

Extensive training, sometimes more than a hundred sessions (e.g., Lonardo et al. 2022), requires substantial labour not only from researchers but also time investment and availability from owners accompanying their pet dogs to the lab. Moreover, current dog touchscreen protocols still require supervision to manage the session and monitor animal welfare, limiting the degree to which training can be fully automated.

Finally, these costs interact with publication bias, as slow training trajectories, unsuccessful pilots, or failures do not reach publication, generating a file-drawer effect (Greenwald 1975; Rosenthal 1994; Simonsohn et al. 2014). Consequently, the literature likely over-represents successful implementations and under-represents the true distribution of training timelines, feasibility constraints, and effect sizes, potentially discouraging adoption through unrealistic expectations or repeated reinvention of unproductive designs.

Practical solutions

Survival guides discussed above would also help to reduce the financial and personnel demands. Detailed shopping lists (i.e., build of materials, BOM), assembly protocols and checklists, would allow laboratories with limited engineering capacity to assemble and deploy touchscreen-based methodologies and at reduced price points.

Standardised touchscreen-based training schedules and stimuli set for dogs can also streamline training, reducing the time spent designing and piloting protocols. Crucially, a shared repository should invite uploads of all outcomes (including null or negative results), enabling cumulative learning about what fails, under what conditions, and at what cost. Such a shared repository directly addresses the file-drawer problem and reduces duplicated effort across laboratories.

When such open-source designs are made portable or autonomous, testing can shift closer to dogs rather than requiring dogs to travel to labs, reducing owner burden and expand sampling. Home-based apparatuses would enable owner-assisted sessions, expanding geographic reach (although this may require more experimental units), and their practical feasibility has been demonstrated in an in-home touchscreen deployment with assistance dogs (Byrne et al. 2018). Even though participant recruitment remains voluntary, it could support a new form of citizen science, where owners engage in cognitive testing with their dogs, contributing to larger datasets while enriching their animals’ daily routines. Conversely, installing apparatuses in kennels or facilities with captive dogs would create continuous data streams from larger cohorts under controlled conditions (Griggs et al. 2021; Harrison et al. 2023; Martin et al. 2022; Schmitt 2018). In both scenarios the logistical burden shifts away from lab personnel, shortens start-up time and diversifies sampling beyond the typical ‘owners who can drive to campus’ demographic, mitigating prevalent recruitment biases (e.g., Ganguli et al. 2015; Oswald et al. 2013; Rosnow and Rosenthal 1976).

Umwelt concerns

Beyond technical and practical barriers, touchscreen-based research with dogs must also consider the alignment between the experimental interface and dogs’ perceptual systems and motor capabilities -- their Umwelt (Uexküll 2001; Uexküll and Kriszat 2013). Here we use the concept of Umwelt to refer not only to an animal’s sensory abilities, but also to the bodily and motor capacities through which it interacts with its environment, as well as the ecological and social contexts that shape their daily life. For dogs, this includes their reliance on other sensory cues (e.g., olfaction) in addition to vision, the physical ergonomics associated with using a nose on a flat, vertical, screen, and their social role as human companions that constrains how and when they can be tested in laboratory settings. Cognitive characteristics, such as dogs’ varying levels of inhibitory control, also form part of this species-specific Umwelt and influence how easily individuals can adapt to touchscreen-based tasks.

Screen-based paradigms, including touchscreens, share general display constraints due to dogs’ lower acuity (Byosiere et al. 2018a, b; Lind et al. 2017; P. E. Miller and Murphy 1995), dichromatic vision (Siniscalchi et al. 2017), and a higher flicker-fusion threshold (Coile et al. 1989; Hirskyj-Douglas et al. 2017; Sheldon et al. 2024). Empirical data on canine luminance and colour sensitivity also remain limited (but see (Byosiere et al. 2018a, b, 2019b; P. E. Miller and Murphy 1995; Neitz et al. 1989). These considerations are not unique to touchscreens, but they limit what any monitor-presented stimulus can effectively test.

Motorically, dogs are not naturally inclined to interact with vertical, two-dimensional surfaces using their noses. Many behaviourally relevant object properties are absent or only symbolically represented on a display: a nose-touch on a screen lacks the olfactory, haptic, and proprioceptive cues that normally guide exploratory or social behaviours (e.g., texture, temperature, compliance, and odour). Performance on 2D representations can vary widely across individuals even when discrimination is possible (e.g., Autier-Dérian et al. 2013; Mongillo et al. 2021; Pongrácz et al. 2003; Range et al. 2008). Additionally, although no studies have concluded sufficiently large or diverse samples to robustly test for breed differences in touchscreen interactions, it is likely that these perceptual and behavioural factors vary across breeds due to physiological differences. Ergonomics also matter, as the spatial offset between eyes and nose tip means that the attended location may not match the contact point of the nose on screen, and this offset varies with morphology (e.g., short-headed compared to medium- or long-headed breeds (Canori et al. 2025; Ichikawa et al. 2024), with some individuals or breeds might need to learn awkward or uncomfortable positions during repeated trials. Additionally, it has been reported that some dogs interact with the screen by sliding or swiping nose from one target to another rather than making clean taps (Zeagler et al. 2016). This observation questions whether the default tap-only interaction, derived from human interface design, is the most intuitive or efficient input method for dogs.

Furthermore, pet dog testing contexts constrain exposure frequency. Unlike laboratory-maintained models that can complete hundreds of trials per day, 5 to 7 days a week (e.g., Alsiö et al. 2019; Fagot and Paleressompoulle 2009; Horner et al. 2013),, pet dogs often attend only a few sessions per week, limiting training throughput. This interacts with the lengthy training often required for touchscreen use mentioned before, two predictable outcomes follow. First, researchers might limit themselves to research questions that can be answered quickly or with minimal training, narrowing the scope of the studies that are deemed feasible to be conducted with dogs. Second, studies that turn out to be slower or less productive within the expected timeframe might stall before publication, contributing to the publication bias described earlier.

Touchscreen-based paradigms may introduce an additional layer of selection of subjects. Because participation requires criterion-based training, individuals who fail to acquire the task (or who disengage during training) are often excluded from subsequent testing. This exclusion creates the possibility that the final tested sample is systematically biased toward individuals with particular physiological, behavioural or cognitive characteristics (e.g., higher trainability, persistence, lower neophobia, stronger food motivation), thereby limiting generalisability to the broader population (e.g., Barnard et al. 2018; Lazzaroni et al. 2019). This issue aligns conceptually with the gifted word learner subpopulation in word-learning research, where only a small subset of dogs shows exceptional acquisition and retention of object labels (Dror et al. 2024, 2026; Fugazza et al. 2021). In the dog touchscreen studies included in our review, we did not find systematic examinations of learner-non-learner differences or analyses of how training shapes the resulting sample, likely reflecting the small sample sizes that characterise this literature.

Additionally, laboratory studies typically benefit from consistent rearing history of animals and controlled schedules for feeding, activity, and testing; such standardisation, however, is not feasible with pet dogs living in private households. Researchers have no control over their day-to-day routines, recent exercise, or food intake, which can vary widely across dogs. Touchscreen-based studies rely primarily on food as the reinforcer, and researchers can only request that owners withhold food for a certain period before a testing session; compliance and exact timing inevitably vary. This variability reduces experimental control over motivation, inflates within- and between- subject noise, and can lengthen training.

Another potential problem is that the inhibitory control of dogs is context-dependent, varying depending on the task, social context, and testing paradigm (Bray et al. 2014; Brucks et al. 2017a, b; Fagnani et al. 2016a; Marshall-Pescini et al. 2015; Vernouillet et al. 2018). For example, in a large comparative across 36 species, dogs were ranked in the mid-range overall, but their performance differed markedly between inhibitory control paradigms, with higher impulsivity scores in cylinder tasks compared to in A-not-B tasks (MacLean et al. 2014). This variation supports that inhibitory control in dogs is not a stable trait but can depend strongly on the specific task structure. Social contexts matter too, as dogs performed better in a social version of an inhibition task with humans present than in a non-social version (Fagnani et al. 2016b; but also see Bray et al. 2015). Touchscreen-based studies, however, make the opposite demand: once the training is complete, the dog must work without human help, choosing whether and when to engage or withhold a nose-touch in front of a screen. It is likely that dogs have difficulties withholding impulsive nose-touches once they know how to touch the screen, human handlers therefore end up restraining the dog or physically restricting the screen to prevent unwanted touches during the inter-trial intervals or before stimulus presentation. Currently it remains unclear whether the apparent ‘impulsivity tax’ associated with touchscreen use reflects a genuine limitation in dogs’ inhibitory control, or whether it arises from task design factors, such as the removal of human social cues, that normally support inhibition during manual testing.

A related and critical challenge is ensuring the dogs’ attention is directed at the stimulus itself. It can be difficult to confirm that the dog is not just interacting with the apparatus but is actively processing the visual or auditory cues presented. This lack of attention can prolong learning, as a subject might touch the screen impulsively without looking at the stimulus, yet still receive reinforcement for a technically correct, yet random, response.

Umwelt considerations

Aligning task design with dogs’ perceptual capacities begins with using stimuli that dogs can reliably see and discriminate. Because dogs perceive colour through a dichromatic (blue-yellow) system, stimuli differing strongly along this spectrum (e.g., deep blue vs. vibrant yellow, or black vs. yellow) are likely to be more easily distinguishable than contrasts that rely primarily on human red-green distinctions (P. E. Miller and Murphy 1995; Pongrácz et al. 2017; Siniscalchi et al. 2017). Discriminability also depends on luminance, contrast and saturation, so colour manipulations should be piloted with different visual parameters controlled where possible (Siniscalchi et al. 2017; see also Byosiere et al. 2019a). Notably, dogs can sometimes succeed on red-green discriminations under specific stimulus configurations, particularly when luminance is carefully controlled, suggesting that performance depends on precise stimulus properties and task design (Byosiere et al. 2019b; P. E. Miller and Murphy 1995; Neitz et al. 1989; Pongrácz et al. 2017; Siniscalchi et al. 2017).

For stimuli with motion, display refresh rates should be chosen with dogs’ temporal resolution in mind (Abdai 2025; Lõoke et al. 2020; Lorenzi and Vallortigara 2021; Scholl and Tremoulet 2000; Schultz and Frith 2022; but also see Kanizsár et al. 2017). Screens that appear stable to humans may shimmer to dogs if refresh rates are too low, potentially degrading stimulus fidelity (Coile et al. 1989). Motion may increase stimulus salience against static backgrounds (e.g., Waldin et al. 2017), although this remains to be empirically tested in dogs. Before committing to a final stimulus set, pilot studies that vary size, contrast, colour composition, and motion should ensure that dogs can, in fact, detect the different experimental contingencies.

To provide the screen with more sensory feedback that typically guides a dog’s exploratory behaviour, visual events are often paired with secondary feedbacks such as brief auditory tones or inter-trial signals in touchscreen-based dog studies (e.g., Laude et al. 2016; Rivas-Blanco et al. 2020). Different forms of secondary reinforcement have been studied in dog training, including the use of auditory cues such as clickers, verbal signals, and different reinforcement schedules aimed at bridging the temporal gap between a correct response and the delivery of food (e.g., Cimarelli et al. 2021; Feng et al. 2016; N. J. Hall et al. 2013; Lazarowski et al. 2025). Empirical findings have been mixed regarding the use of clickers, with some studies found little or no advantage of clicker training compared to conventional food-only reinforcement (e.g., Burton 2020; Gilchrist et al. 2021), whereas Lazarowski et al. (2025) reported clear benefits of clicker in detection dog training. They attributed this difference to methodological and contextual factors, specifically the temporal gap between behaviour and reward delivery, task complexity, and trainer expertise. Markers appear most advantageous in conditions requiring precise timing, when the task is difficult, or when immediate primary reinforcement might be delayed, supporting its function as an effective secondary reinforcer in touchscreen studies. It is also possible to connect the touchscreen with external devices such as buttons or capacitive paw pads (e.g., Kenawell et al. 2025), aligning the visual presentation of stimuli with the motor act associated with natural exploratory movements. Related work has likewise developed touchscreen for medical alert and wearable interfaces for working dogs, emphasising designs that remain usable under constraints and that leverage behaviours dogs can perform comfortably and reliably (Jackson et al. 2018).

Practical interface should include ergonomic accommodation and design for nose-based input, such as generous target sizes, wider spacing, and tolerance for sliding contacts, rather than assuming discrete tap events (Zeagler et al. 2016). Where feasible, software should treat contact trajectories as meaningful data (not merely noise), both for error analysis and for understanding dogs’ motor capabilities.

To address potential sampling bias, studies should compare baseline characteristics of dogs that reached final testing stage versus excluded dogs. This strategy aligns with broader calls to improve external validity in behavioural science (e.g., STRANGE considerations; Webster and Rutz 2020), and would allow the field to quantify whether touchscreen paradigms favour particular dog profiles.

Regarding inhibitory control, studies can incorporate mechanisms such as limited hold responding, variable inter-trial-intervals, and timeouts for impulsive touches (responses before stimuli presentation), drawing conceptually from rodent touchscreen protocols (e.g., Beraldo et al. 2019; Kim et al. 2015). These measures can both reduce impulsive touches and generate quantifiable metrics of inhibitory control or attentional failures.

A direct way to test the social support hypothesis is to implement paired conditions in which dogs perform the same touchscreen task with and without human presence. This comparison would clarify whether inhibitory control problems are intrinsic to touchscreen use or are partially a byproduct of removing social cues that facilitate response inhibition.

To ensure attention to the stimulus, and to handle cases where the dog makes incidental touches or does not respond at all, touchscreen paradigms are typically implemented as closed-loop systems, in which the animal’s ongoing behaviour (e.g., response initiation, touch timing, location, and trajectory) determines whether a response is accepted and whether the trial advances. Actionable design suggestions to further improve the touchscreen-based task design have been proposed in other species’ literature. In classic touchscreen research, selection can be defined at initial contact, at the first contact with a target, or on release, with lift-off selection allowing continuous feedback and correction before committing a response (Potter et al. 1988). These distinctions are especially relevant for dogs, where sliding on screen is common (Zeagler et al. 2014). A study with parrots also reported frequent multi-contact and recommended intervention by using larger targets and filtering excessive touches with software to improve selection reliability (Kleinberger et al. 2024). In macaques, protocols require responses on the restricted areas where stimuli are presented and penalize touches on blank areas (e.g., Loyant et al. 2022). Similarly, chickens are required to peck the same stimulus multiple times to confirm their selection and distinguish deliberate choices from accidental or clumsy touches (e.g., Nasrini and Hampton 2024). We acknowledge that not all design choices can be directly transferrable to dog. Precision requirements, such as responding to a narrowly defined touchscreen area, may need adjustment to accommodate a dog’s nose touches.

In conclusion, touchscreen research with dogs is rare not because it lacks value, but because it faces technical, practical, and species-specific challenges. Potential solutions are proposed in this review. By adopting open-source designs, sharing detailed building and training guides, and developing common reporting standards, researchers can make touchscreen studies more accessible and comparable across labs. Multi-lab collaborations and home-based setups could further expand participation and data diversity. With collective effort and dog-centred design, touchscreen methods can become a standard, reliable tool for studying canine behaviour and cognition, improving reproducibility and accelerating discovery.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

S.Y.F. conceptualization, data curation, formal analysis, investigation, methodology, writing— original draft, writing—review and editing; C.M. data curation, formal analysis, writing—review and editing; C.C. data curation, formal analysis, writing—review and editing; D.R.B. data curation, formal analysis, writing—review and editing; O.G. data curation, formal analysis, writing—review and editing; F.R. supervision, funding acquisition, writing—review and editing; T.M. conceptualization, supervision, formal analysis, investigation, methodology, writing—review and editing.

Funding

Open access funding provided by University of Veterinary Medicine Vienna. S.Y., C.M., D.R.B., F. R., and T.M. were supported by the Austrian Science Fund (FWF) Grant DOI https://doi.org/10.55776/P37052. D.R.B. was funded by a Marietta Blau Grant from the Austrian Agency for International Cooperation in Education, Science and Research (MMC-2023-07030). C.C. was funded by a doctoral grant from the University of Parma (cycle XXXVIII). The FCT (Fundação para a Ciência e Tecnologia, I.P.; https://www.fct.pt/en/) supported this work, through a grant to T.M. (CDL-CTTRI-249-SGRH/2022), and multiannual funding to the WJCR in the context of the R&D Unit: UID/04810/2025.

Data availability

All data is available as supplementary material.

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical approval

This article does not contain original research data to which ethical considerations would apply.

AI statement

We have not used AI-assisted technologies in creating this article.