Can Artificial Intelligence be used to improve productivity by automating elements of the User Experience design processes?

1.0 ABSTRACT Artificial Intelligence (AI) has many advantages over humans in that it can detect incredibly subtle patterns within large quantities of data. This study suggests using AI algorithms in user research tasks for mining variables ranging from the tone of voice, image banks, historical records, and product use to determine where brands sit and where there could be competitive advantages. This study reviews the current design processes of User Experience (UX) designers and the advancements in Deep Learning (DL). It identifies potential areas of automation in the Preparation, Incubation and Illumination stages of the design process. It recognises the possibility of using Long Short-Term Memory (LSTM) models to automate design feedback and the creation of alternate designs with Generative Adversarial Networks (GANs).

in his book The Art of Thought (Wallas 2014) . Wallas theories are that all creatives follow a four-stage creative process, Preparation 1 , Incubation 2 , Illumination 3 and Verification 4 .
The UK Design Council has also made great strides to analyse creative processes. Formalising a framework for innovation in (Design Council 2015 with the Double Diamond . The Double Diamond represents a process for exploring issues more widely or deeply (divergent thinking) and then taking focused action (convergent thinking). The framework Discover, Define, Develop and Deliver adapts the divergence-convergence model proposed in Béla H. Bánáthy's book Designing Social Systems in a Changing World (Banathy 1997) .
However, all too often, our collective understanding of the design process is the assumption that ' If you do this, then this will happen . It may do, yet it may not. This is something to be cautious about if we are to suggest automating elements of the UX design process as the design process is agile 5 in nature.
Often projects require frequent changes because the result is rarely known. Defined as ' The solution-first strategy' (Cooley 2000;Carroll 2002;Vicente 2013) , interim design solutions are generated as a method for gathering feedback. These feedback cycles are important to product optimisation and truly user-centred design. It can be assumed from the research that the AI would need to adapt based on new information at various stages of the process to be useful and productive.
Further research uncovers an Illumination or 'eureka' moment as a defining aspect of the design creation phase. Numerous design scholars (Lawson 2005;Cross 2011;Dorst 2011) have sought to understand the cognitive processes designers use-leading to the belief that most ideas come from the subconscious. In recent decades, scientific approaches to inform our understanding of how designers think have been explored by a range of authorities (Flaherty 2005;Carson 2010) . Each scholar found that a combination of distraction and dopamine was required to improve the creativity and ideation process. Triggers were found within events that made participants feel great and relaxed (releasing dopamine), allowing the subconscious mind to solve the problems that the conscious mind failed to consider.
If this is true, that ideas come from dreams and the subconscious. For our algorithm to be useful it would need to ideate autonomously. But can we create algorithms that dream, let their thoughts wander and begin to innovate? Research suggests this can be answered in many ways, but must begin with the assessment of intelligence.
The Turing Test (Turing 1950) , for example, is claimed by many to be a way to test a machine's ability to exhibit intelligent behaviour. Turing proposed that a human judge evaluate natural language conversations between a person and a machine designed to generate human-like responses. Unfortunately, attempts to 5 Agile workflow is an iterative method of delivering a project. In Agile, multiple individual teams work on particular tasks for a certain duration of time termed as 'Sprints'. Agile Workflow can be defined as the set of stages involved in developing an application, from ideation to sprints completion. build computational systems able to pass the Turing Test have devolved into manipulation games designed to trick participants into thinking they are interacting with a human (Bringsjord 2013) , falling short of emulating real intelligence. Despite this, the Turing Test is still highly influential and remains an important concept in assessing Artificial Intelligence (Pinar Saygin et al. 2000) . Many scholars use it as the basis to design other tests such as the Lovelace Test (Bringsjord et al. 2003: 215-239) .
Named after the 19th-century mathematician Ada Lovelace, the Lovelace Test has the notion that if you want to replicate human-like abilities, AI must have the ability to create new things. Demonstrating Intelligence and original thought. Thus the level of creative ability becomes a proxy for the measurement of Intelligence. This could create a huge problem for this study. As in theory, ideas and creativity come from dreams and the subconscious. As found in the research, the subconscious is an area too challenging to measure with a level of confidence.
The measurement of consciousness has seen several proposals (Arrabales et al. 2010;Brazdău and Mihai 2011) . However, this has led to much debate. Some scholars argue that consciousness could be considered a quantitative property (Baum 2004: 362) rather than a qualitative one, and others see consciousness as a multidimensional feature (Fazekas and Overgaard 2016) . Whether consciousness can be measured or developed with any accuracy, I would consider it too difficult to resolve within this paper, so I will attempt to resolve this feature in a future study.

-PRIMARY RESEARCH OF CREATIVE PROCESSES
Our secondary research theories that all creatives follow a four-stage creative process: Preparation, Incubation, Illumination and Verification. However, secondary research is often outdated and may no longer be accurate. The study invited design professionals to participate in a survey to validate the secondary research. Again, to ensure our model is trained accurately enough to automate components.

-Participants
Twenty digital graphic designers were recruited by personal invites on Linkedin to complete a ten-question survey. The data set comprised 4 participants, aged 25-34, 12 participants aged 35-44 and 4 participants aged 45-54. The participants were similar in that they all had a fairly high level of education, either completing an Undergraduate degree (13 participants) or a Postgraduate degree (7 participants).

-Research questions
A range of questions was asked around the origins of briefs, core skills, design process, tools, techniques and project success. The primary research identified divergent thinking, identification and mitigation of blind spots and variant idea generation as key requirements for designers. The findings also suggest Sprint 2.0 6 and Design Thinking 7 ideation techniques would be the recommended processes to teach an AI in addition to Abstractive summarisation 8 . RQ1: Participants indicated that most professional design works responded to either Horizontal Integration 9 or Vertical integration 10 strategies. Horizontal integration is where a client has acquired 10 Vertical integration can refer to something not working or a need to reduce costs by improving operational efficiencies. This can be anything but commonly starts with root cause analysis. 9 Horizontal integration can refer to acquiring someone or creation of a new product and service. This can usually mean a redesign or integration of a new process. 8 Abstractive text summarisation is the task of generating a headline or short summary of a few sentences that capture the salient ideas of an article or a passage of text. 7 Design Thinking is an iterative process in which we seek to understand the user, challenge assumptions, and redefine problems to identify alternative strategies and solutions that might not be instantly apparent with our initial level of understanding. 6 A variation of the Google Ventures Design Sprint that is more refined and condensed and aims to solve Design Challenges over four days. someone or created a new product and service-usually requiring a redesign or integration of a new process. Whereas Vertical integration is when something isn't working, a client needs to reduce costs by improving operational efficiencies. This can be anything but commonly starts with root cause analysis.
RQ2: Our participants identified ten soft skills graphic designers must have. They identified Analytical thinking, Collaboration, Communication, Composition, Creativity, Cultural awareness, Curiosity, Empathy, Problem-solving and Understanding as the most desirable traits and behaviours.
RQ4: A common sentiment in our sample is that working alone allowed focus but was rare in the modern-day workplace. Collaboration is also identified as a way to improve the work and innovate. For example, one responder explained: "Working with other people takes you to new places, and that's exciting. It's also reassuring when you then realign, and it's a much more neurotic process alone. Due to the size of the business, it's very rare for an individual to design completely in isolation." RQ5: A broad range of ideation tools and techniques were used across our sample size. Sprint 2.0 and Design Thinking Ideation techniques were popular, with 18 participants mentioning some practice elements, such as Sketching, Experience mapping, Crazy 8s, Mood boarding and Mind mapping.
RQ6: Our participants identified several research techniques. The most popular techniques identified were: Online desk research, Data analysis, Books and library research. Competitor analysis, Process analysis, Experience mapping, Primary research, Consumer trend and market research reports investigation were also seen.
RQ7: Project management, creation process, Research scope, Speed and breadth of idea generation, validation, and sign-off were identifiable trends. For example, one responder goes on to say: "The distilling and picking one path forward. So many fears around "is this the right way?" and "there are so many good nuggets of ideas here, how can I leave this behind?". RQ8 and RQ9: Metrics, feedback loops and serendipitous discovery were cited as the main measurements when designers felt the work was really good and successful.
RQ10: Participants showed that AI could optimise the Graphic Design ideation processes in several ways. For instance, the QA and Testing process in digital product creation. Management of repetitive and time-consuming tasks. Analysis of processes that had to be repeated, such as Design optimisation, creation of assets within design systems. Exploration of ideas, iteration and abstractions was cited by a majority of participants (15), with divergent thinking being a key requirement. One participant went on to say, "It would be amazing if AI could help generate many variations of an overall design approach and do some testing to see which version(s) perform the best. A quick way to help distil possible approaches."

-AI LANDSCAPE RESEARCH
Our primary research aligns with our secondary research. It was surprising to observe that the answers to The research also identifies areas where AI could support a project's preliminary stages and determines divergent thinking, identification and mitigation of blind spots and variant idea generation as key areas designers would appreciate automation. The research indicates that designers would use AI algorithms for research tasks, such as mining tone of voice in documents. The sorting and classification of the image banks, sentiment identification in historical designs, and insight extraction from product analytics to determine where design optimisations could be made.
An exploration of the AI landscape is required to explore the solutions that could potentially be trained to automate the aspects identified by designers.
AI could be one of the most transformative technologies in human history and has seen use across multiple industries (Amazon Web Services 2020; Caulfield 2021) . PricewaterhouseCoopers (2016)  how AI undoubtedly would take over routine tasks and liberate us to do more stimulating work. With the AI landscape growing so rapidly, the challenge was to explore the ecosystem to evaluate current uses, how each technique works and what methods could be repurposed for the benefit of designers.
One of the most well-known architectures of DL is Convolutional neural network (CNN). It is typically used in image processing applications (Valueva et al. 2020) . CNN has seen use in identifying user intention (Ding et al. 2015) , which suggests great potential for our model if integrated.
Another is Recurrent neural networks (RNN), typically designed to recognise sequences and patterns in speech, handwriting and text applications (Mandic et al. 2001) . RNN research suggests (Zhou et al. 2019) it could aid conversion in complex user journeys, guide creative design (text and images), and lead to higher product effectiveness which demonstrates great potential for our model.
An extension of AutoEncoder (AE), Denoising AutoEncoder (DAE) is an asymmetrical neural network for learning features from noisy datasets (Kramer 1991) . DEA has seen use in image classification (Druzhkov and Kustikova 2016) and speaker verification. This model shows great potential for mining and sorting variables from various datasets, significantly reducing the time in manual processes currently used within my organisation.
Typically employed for high dimensional manifold learning (Bengio 2009) , Deep belief networks (DBNs) is one of the most reliable deep learning methods with high accuracy and computational efficiency (Bengio et al. 2007) . Its application has been seen in many engineering and scientific problems-specifically, human emotion detection (Movahedi et al. 2018) , which would be an exciting feature of our model. whether generated examples are real, from the domain, or fake, generated by the generator model. Its application has seen use in Art generation, Astronomy, Video games (Schawinski et al. 2017;Wang et al. 2018;Yu et al. 2018) , and the generation of Deepfake (Singh et al. 2020) images, which would serve us well to generate design variants based on user feedback. These discoveries allow us to hypothesise that AI could aid designers in all four stages of the creative process. Using a combination of technologies, AI could mine variables ranging from the tone of voice, image banks, historical records, and product use during the Preparation stage. Deliver insights from the gathered data, map them to business priorities and features during the Incubation and Illumination stages.

An RNN method that benefits from feedback connections is
And automate the feedback process during the Verification stage.
An AI design in response to my research findings will be developed in the form of an artefact. However, the creation of such a model is beyond the skill-set of this author and would require a larger team.

-FURTHER STUDY
This work is part of my ongoing research into the efficacy of business operations for myself and my teams.
Inevitably I see further alignment in my practice with advanced AI systems. The authors of 'Understanding Consumer Journey Using Attention Based Recurrent Neural Networks' (Zhou et al. 2019) have made great steps in creating models that aid conversion in complex user journeys. They provide testing criteria that could be used to develop the findings in this paper, given the technical knowledge and resources.
Additionally, the measurement of consciousness and whether it can be developed with confidence seems too intriguing to ignore. Fundamentally, it isn't easy to measure consciousness itself as it is believed to be a constellation of features, mental abilities, and thought patterns. Although some people may experience their own consciousness as a unified whole, we assume that consciousness is a multidimensional set of attributes present to differing degrees in a given mind. i.e. you might give love a higher value than me and so on. So whether consciousness can be measured or developed with any accuracy would be interesting to study.

Background and Context
Artificial intelligence (AI) could be one of the most transformative technologies in human history. It is a technology that has seen use in Healthcare (Amazon Web Services 2020) , Manufacturing (Caulfield 2021) , and Education (Carnegie Learning 2020) . With PricewaterhouseCoopers predicting a $15.7 trillion investment in AI by 2030 (PricewaterhouseCoopers 2016) .

Relevance and Importance of the Research
Contributions presented at the International Conference on Human Interaction and Emerging Technologies: Future Applications (Ahram et al. 2020) have demonstrated how AI undoubtedly would take over routine tasks and liberate us to do more stimulating work. However, there is little research on whether AI can become intelligent enough to contribute to creativity, critical thinking, and original thought.

Problem Statement
In this study, I will look at how humans use technology to enhance tasks in Design and whether Design inevitably can be automated.

Research Questions
My central proposition is that if Design is a field of study where 'Good' Design is taught, why can't a machine fully perform these practices?

Research design
I will explore AI, ML and DL methods to gain insights into which methods can potentially be trained in the Graphic Design ideation process. I will then explore design pedagogy to identify the most efficient, repeatable, scalable and consistent processes that an AI could be trained in. To validate the secondary research, I will invite a sample size of 15-20 graphic design professionals to participate in a survey.

Methods and Sources
Digital graphic designers will be invited to participate in a study via personal invites on Linkedin to complete over five days. The survey will be designed to question participants aged between 18-54+ and will not discriminate against individuals with non-formal education training.

Practical Considerations
The study will not create a working AI but suggest how an AI could potentially be built if technological resources were available.