See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/378285485

Weaving Fashion with A.I.

Article in Fashion Highlight · February 2024

DOI: 10.36253/fh-2500

CITATIONS READS

0 36

1 author:

Patrizia Marti
University of Siena
192 PUBLICATIONS 2,643 CITATIONS

SEE PROFILE

All content following this page was uploaded by Patrizia Marti on 28 November 2024.

The user has requested enhancement of the downloaded file.

WEAVING FASHION
WITH A.I.

PATRIZIA MARTI
University of Siena
[email protected]

Copyright: © Author(s). This is an open access, peer-reviewed article published by Firenze University Press and distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability Statement: All relevant data are within the paper and its Supporting Information files.
Competing Interests: The Author(s) declare(s) no conflict of interest DOI: https://doi.org/10.36253/fh-2500

Fashion Highlight - No. 02/2023 ISSN: 2975-0466 [Online] 52

Abstract

Artificial Intelligence (AI) has revolutionized the fashion industry, impacting various stages of the
fashion lifecycle and offering promises of enhanced operational efficiency, personalized con-
sumer experiences, and contributions to sustainability and innovation. This paper outlines the
main trends, implications, constraints, and potential risks associated with AI in fashion design
and production. In particular, the paper delves into biases in AI-powered systems, emphasizing
data, algorithmic, and transparency gaps that can perpetuate discrimination.
Bringing ethics and pluralism into AI-powered design, the paper introduces intersectional femi-
nism and feminist design principles, which can help envisioning a more inclusive, equitable, and
unbiased fashion discourse.
In conclusion, the paper contends that feminist design principles can inspire the development of
more inclusive and ethical AI-powered fashion. By challenging norms, promoting diverse re-
presentations, and incorporating user perspectives, feminist design can contribute to a socially
responsible and equitable future for fashion technology.

Keywords: AI-Powered Fashion, Biases, Intersectional Feminism, Feminist design, Ethics

Introduction
Personalised shopping experience
In the realm of fashion, Artificial Intelligence (AI)
has been transformative, impacting various aspects In the age of data-driven insights, personaliza-
of the fashion lifecycle. Several applications have tion acts as the entry point to numerous opportu-
entered the market and the manufacturing process nities, fostering deeper connections between
with the promise that the integration of AI in retailers and their customers. One of the most
fashion cannot only enhance operational efficien- popular applications of AI in fashion is the creation
cy but also offer a more personalized and engaging of a personalized shopping experience with a
experience for consumers while contributing to variety of modes and contexts of use. Through
the industry’s evolution towards sustainability and sophisticated AI algorithms, customer preferences,
innovation. purchase history, and browsing behavior are
In what follows, we first briefly portray the main analysed to provide tailored product recommen-
trends in the use of AI in fashion, and later dations. This not only can enhance the overall
discuss implications, constraints and potential shopping experience but also ensure that customers
risks associated with design and production of are presented with items that resonate with their
AI-powered fashion. individual tastes.

53 No. 02/2023 - Fashion Highlight

AI-powered styling tools offer personalized style quality is controlled by AI-powered computer
and outfit recommendations. These tools analyze vision systems during the manufacturing process
individual preferences, body types, and fashion to identify defects or inconsistencies in textiles and
trends to suggest complete outfits. Whether garments. Image recognition is also used to ensure
integrated into online shopping platforms or the authenticity of luxury goods, protecting brands
utilized by fashion influencers and bloggers, these and consumers alike.
tools aim to enhance the overall fashion experience Retailers can also benefit from AI algorithms for
for consumers. dynamic pricing which optimise pricing strategies
Personalised shopping does not limit to receiving based on market conditions, competitor pricing,
suggestions of purchasing. The development of and customer behavior allowing them to stay
virtual try-on solutions powered by AI enables competitive and enhance overall profitability.
users to visualize how clothing items will look At a more systemic level, AI can also be actively
on them without physically trying them on. For employed to address the sustainability challenge,
example, Walmart has launched the “Be Your which is a growing concern in the fashion industry,
Own Model” feature, enabling users to virtual- by sourcing materials and monitoring the environ-
ly try on clothing by utilizing their own photos mental impact of production processes.
and augmented reality (AR) technology. Amazon
Fashion collaborated with Snap, offering Snapchat AI-generated fashion
users the capability to virtually showcase branded
glasses and sunglasses through AR filters. In the creative processes of fashion design, the use
Leveraging AR, these virtual fitting rooms create of AI is controversial due to contrasting viewpoints
realistic representations, revolutionizing the way on its impact. Some designers consider generative
consumers interact with fashion in the digital AI as a threat, some others as a valuable ally. These
space. The virtual try-on App Auglio (2023) offers divergent views stem from a complex interplay of
an easy to integrate plug-in to quickly try eyewear, technological advancements, industry dynamics,
cosmetics and accessories. The company advertises and individual perspectives within the fashion
a 20% average increase in sales from the use of AR. design community.
The shopping experience can also be enhanced Risks are associated to potential job displace-
using AI-driven chatbots which provide instant ment, loss of human touch, and a shift away from
and personalized support, assisting customers traditional craftsmanship. A main concern is
with product inquiries, sizing questions, and related to the lack of creativity since generative AI
order tracking. This serves the purpose to not only relies on patterns and data from existing designs,
improve efficiency in customer service but also which may lead to the generation of designs that
present prospects for brands to explore hyper-per- lack true originality producing designs that closely
sonalization and refine their offerings to align with mimic existing trends without introducing novel
the preferences of their customers. concepts.
From an opposite viewpoint, some designers
Supply chain optimization and quality control consider AI as a valuable ally, appreciating its
ability to enhance efficiency, provide innovative
AI also plays a pivotal role in optimizing the design suggestions, and contribute to a faster and
fashion supply chain. By predicting demand, more data-informed creative process. By providing
automating inventory management, and refining inspiration, generating design concepts, and even
production planning, AI mitigates overstock and assisting in the creation of patterns, AI has the
understock situations, leading to more streamlined potential to accelerate the design process and
and efficient operations within the industry. encourage creativity.
In the dynamic and fast-paced world of fashion, Whatever the personal opinions, there is no doubt
predictive trend analysis is crucial. AI algorithms that the integration of AI into the fashion industry
sift through vast amounts of data from social has ushered in a new era of innovation, efficiency,
media, fashion blogs, and historical sales to forecast and personalization, fundamentally reshaping the
emerging trends. This can empower designers to way fashion is designed, produced, and consumed.
stay ahead of the curve, responding promptly to However, beside some benefits, AI-generated
shifting consumer preferences, but also retailers fashion design poses challenges and ethical
who can rely on high-quality end products whose considerations. Questions related to intellectual
Fashion Highlight - No. 02/2023 ISSN: 2975-0466 [Online] 54
property, cultural sensitivity, and the potential loss crafted by a team entirely composed of cisgender
of human craftsmanship need to be addressed as designers might overlook the needs of non-binary
this technology becomes more prevalent. individuals. The lack of diversity in design teams
Furthermore, AI-based technologies may harbor can also create blind spots, hindering the identifi-
biases, arising from incorrect assumptions in the cation of potential issues or opportunities. Without
machine learning process. If the training data is varied perspectives, important details may be
generated by a stochastic process or an ill-defined overlooked, leading to services that are inefficient,
problem, or if it poorly represents discriminat- ineffective, or even harmful.
ing features, creating an effective prediction model Transparency Gap: The lack of transparency is a
becomes challenging. For instance, training a critical factor contributing to discrimination in
model to predict a specific trend in fashion using AI-based systems. As algorithms become more
only gender information limits its predictive power complex and autonomous, understanding the
due to the absence of other significant factors like inference mechanisms adopted by the machine
height, age, geographic area, social and cultural can become challenging. This opacity in the
background. decision-making process can result in biased
While an algorithm representing every point in outcomes that are difficult to scrutinize or rectify.
the input distribution would minimize errors on
training data, it may also capture random variations Bringing ethics and pluralism to AI-powered
and noise, leading to poor predictive ability and design
high generalization error. Therefore, selecting
the right dataset and training model is crucial for D’Ignazio and Klein (2019) advocate for a fresh
generating reliable predictions, as it ensures the perspective on data science and data ethics,
inclusion of relevant factors and minimizes biases proposing the adoption of Data Feminism as a
introduced by inadequate or skewed data. framework. This approach seeks to comprehend
how factors such as race, class, sexuality, ability,
Biases in AI-powered systems age, religion, and geography profoundly influence
an individual’s experiences and opportunities
Jamnik and Cheng (2021) posit that biases in in the world. Data Feminism extends beyond
AI-based systems can stem from three distinct gender considerations; it embraces pluralism and
sources: diverse voices, acknowledging that knowledge
Data Bias: Biases may emerge from datasets creation is inherently situated, embodied, and
that lack representation, especially concerning context-dependent.
minority groups. This deficiency can arise when This perspective aligns with “intersection-
minority groups lack internet access, hindering al feminism”, a term popularized by the legal
the generation of suitable data for processing. The scholar and advocate Kimberlé Crenshaw (1991),
term “Missing Data” (D’Ignazio & Klein, 2020) is which emphasizes the significance of recognizing
coined for data not intentionally created or used for how various social identities intersect and
training, influenced by factors like prejudice and a interact to shape an individual’s experiences of
lack of societal will to include them. The absence privilege and oppression. Intersectional feminism
of representation perpetuates existing prejudices acknowledges that individuals can simultaneously
against minority groups and further contributes to embody multiple marginalized identities, such as
their exclusion from society. race, gender, class, sexuality, disability, and more.
Algorithmic Bias: Professionals (develop- This feminist approach challenges the limitations
ers, designers, and engineers) shaping AI-based of single-issue feminism, which focuses solely
systems may not encompass the diversity present on gender-related issues without acknowledging
in the wider world. This lack of diversity in the the intricate ways other factors intersect to shape
design process results from the natural tendency individuals’ lives.
to work and reason within one’s worldview, shaped In the context of feminist design, intersectional
by personal beliefs and ideas that get transmitted theory and critique are considered essential
to machines. The absence of diverse perspectives components. They provide a more inclusive and
may inadvertently lead to biased services favoring comprehensive understanding of gender issues
a specific user group, while neglecting the needs and their intersection with other social identities.
and viewpoints of others. For instance, a t-shirt Feminist design actively critiques and challenges
55 No. 02/2023 - Fashion Highlight
gender stereotypes and biases that may be on gender commonalities and differences, human
perpetuated through design choices. Designers are sexuality, pleasure and desire, experience, and
encouraged to steer clear of reinforcing harmful emotion.
gender norms and, instead, promote positive and
empowering representations. Feminist design as a way to AI-powered fashion
This critique has been expanded within a feminist
approach aiming to encourage the integration of a We believe that feminist design can significantly
norm-critical, gendered lens in the design of new contribute to the development of better AI-powered
technologies. fashion. Taking a feminist design orientation in
Bardzell (2010) defined qualities of feminist design fashion can challenge and avoid the reinforcement
which can help deepen the knowledge on how of gender stereotypes. By questioning traditional
different gender identities can shape both the use of norms and biases, AI-powered fashion systems can
technologies and their design. be designed to offer a more diverse and representa-
The first quality is pluralism. It encourages tive range of styles that cater to a broad spectrum of
the designers to take multiple points of view gender identities.
prioritising the diverse experiences and Feminist design emphasizes inclusive represen-
perspectives of users, particularly those from tation, acknowledging the diversity of body
marginalized communities, in the design process. shapes, sizes, and identities. AI-powered fashion
This helps to ensure that technology is inclusive can benefit from this approach by ensuring that
and accessible to all users. recommendations, designs, and representations
The quality of participation refers to valuing are inclusive and celebrate a variety of styles and
participatory processes to facilitate the emergence expressions.
of empathic relations between the designer and In the following we try to envision how AI-powered
the target people. Feminist interaction design fashion could concretely incorporate feminist
places the needs and experiences of people at the design principles toward a more inclusive, equitable
forefront of the design process. This helps to create and unbiased fashion discourse.
technology that is responsive to the unique needs of
different communities. Pluralism
The quality of advocacy prompts the designer to AI systems in fashion often involve sizing
behave ethically during the design process, valuing recommendations. Feminist design principles
the contribution of people involved, offering encourage the consideration of size and body
progressive solutions, and trying to be open diversity, ensuring that fashion AI takes into
minded without imposing values which might account a wide range of body types and offers
offend or harm on users and other stakeholders. personalized recommendations that empower
Ethical considerations such as privacy and security people.
must be incorporated into the design process Levi’s recently announced its initiative to explore
to ensure that technology serves the needs and computer-generated models designed to represent
interests of users. individuals from marginalized communities in
The quality of self-disclosure refers to the the digital realm. Developed in collaboration
possibility for people to be aware about the with LaLaLand.ai, these digital renderings are not
assumptions that a software program is taking intended to replace human models entirely but
about him/her and creates opportunities for rather serve as a complementary tool. The goal is
expressing preferences through implicit or explicit to enhance the company’s representation of diverse
behaviours. sizes, skin tones, and ages. Embracing AI in this
The quality of ecology encourages interaction manner opens up possibilities for brands to combat
designers to adopt a comprehensive and structural biases by contributing data to program algorithms
approach, contemplating how design artifacts with inclusivity at the forefront. Moreover, it offers
impact all stakeholders and their interrelationships an opportunity for models, especially those from
within the system ecology they operate in. underrepresented backgrounds, to be compensated
In contrast, the embodiment quality challenges for the use of their digital likenesses.
the disembodied perspective of the user associated The intention of Levi’s in using the AI clothing
with a mentalistic view of human information models resulting from their partnership with
processing. Instead, embodied design places value LaLaLand.ai is to foster greater body inclusivi-
Fashion Highlight - No. 02/2023 ISSN: 2975-0466 [Online] 56
ty allowing consumers to visualize how a garment Feminist design places a premium on user
would appear on a variety of models, encompassing empowerment and agency. In the realm of
diverse body types, ages, sizes, and skin tones. The AI-powered fashion, this translates to systems that
concept holds the potential to assist individuals allow users to control and customize their fashion
who struggle with the common issue of not seeing experiences, making choices that align with their
items modeled on a body that aligns with their preferences and values.
own. However, Levi’s application has also not been For example, in 2019, Carlings, a Scandina-
free from criticism as some have highlighted the vian fashion brand, collaborated with creative
fact that the use of digital models reduces that of agency Virtue to launch a digital collection called
real models, and this can generate the undesirable “Neo-Ex.” This collection involved customers
result of job loss. submitting a photo, and designers used AI
technology to create a unique digital outfit that
Participation appeared in the photo. The initiative aimed to
Feminist design values participatory processes, reduce the environmental impact of fast fashion
ensuring that the voices of diverse users are heard and offer a new, digital approach to personal style.
in the design of AI-powered fashion systems.
Involving users in the design process helps create Ecology
technologies that better meet their needs and Feminist design often incorporates considerations
preferences. of environmental and social sustainability.
For example, Quietude is a project funded within AI-powered fashion systems can be designed to
the H2020 EU Research Programme, WEAR prioritize sustainable practices, ethical sourcing,
Sustain (2023), which focuses on raising awareness and responsible manufacturing, aligning with
of ethical considerations related to the collection feminist values of social and environmental justice.
of users’ personal (physiological) data by wearable For example, Adidas’ Speedfactory initiative marks
technologies. The project developed a jewellery a groundbreaking move by the brand to incorporate
system addressing a complex entanglement of artificial intelligence into the realms of sustainable
needs of deaf people from being aware of environ- fashion marketing and production. As a pioneer-
mental sounds to personalisation to express an ing approach to reduce its carbon footprint, Adidas
individual sense of style and identity (Marti, 2021). utilized these Speedfactories not only for manufac-
The jewels can detect, recognize and filter sounds turing but also for direct product distribution from
of interest like an alarm, a car horn, the doorbell a single location. This strategy led to a substantial
etc. and notify the wearer through lights patterns, decrease in long-distance freight and associated
vibrations and shape change of the jewels. emissions, making a notable contribution to a more
The jewellery system was developed by a team sustainable supply chain (Rathore, 2017).
comprising deaf women with diverse professional
backgrounds and varying degrees of deafness, along In conclusion, we believe that feminist design
with ethicists, makers, designers, and technology principles can inspire the creation of more
experts. This diverse group has been intentionally inclusive, ethical, and empowering AI-powered
assembled to create aesthetically sophisticated and fashion. By challenging traditional norms,
desirable objects—iconic pieces inspired by the promoting diverse representations, and
personal experiences of deaf women, the inherent incorporating people’s perspectives, feminist
sensuality of the body, and the capacity of accesso- design can imbue fashion technology with values
ries to embody desires. Recognizing the intricate to contribute to a more equitable and socially
connection between personal expression, visibility, responsible future.
and discretion, Quietude aims to bring these
products to market, acknowledging the profound References
and complex meaning they hold. Importantly, the Auglio, Retrieved December 20, 2023 from: https://auglio.
project values the expertise of deaf contributors com.
Bardzell, S. (2010). Feminist HCI: Taking stock and
alongside other team members, viewing them as outlining an agenda for design. In Proceedings of the SIGCHI
equal contributors shaping the final outcome. Conference on Human Factors in Computing Systems.
New York: ACM, New York, 1301–1310. doi: https://doi.
Advocacy and self-disclosure org/10.1145/1753326.1753521
Crenshaw, K. (1991). Mapping the Margins:

57 No. 02/2023 - Fashion Highlight

Intersectionality, Identity Politics, and Violence against
Women of Color. Stanford Law Review, Vol. 43, No. 6, pp.
1241-1299.
D’Ignazio C., & Klein L. F. (2020). Data Feminism,
Massachusetts. The MIT Press. ISBN 0262044005.
Jamnik, M. & Cheng, P.C.H. (2021). Endowing machines
with the expert human ability to select representations: why
and how. In Muggleton, S. and Chater, N., (eds.), Human-Like
Machine Intelligence, chapter 18, pages 355-378. Oxford
University Press.
Marti P. (2021) Framing diversity: designing hearing aids
from a deaf culture perspective. Design Culture(s). Cumulus
Conference Proceedings Roma 2021, Loredana Di Lucchio,
Lorenzo Imbesi, Angela Giambattista, Viktor Malakuczi
(Eds.), 2021. Volume #2 ISBN 978-952-64-9004-5. Cumulus
Conference Proceedings Series, N°7, ISSN 2490-046X, pp
1958-1978.
Rathore, B., 2017. Aligning Profitability and Environmental
Responsibility: A Study on Sustainable Marketing Strategies.
Eduzone: International Peer Reviewed/Refereed Multidisci-
plinary Journal, 6(2), pp.7-15.
WEAR (2023) Wearable technologists Engage with Artists
for Responsible innovation. Retrieved December 20, 2023
from: wearsustain.eu/about

Fashion Highlight - No. 02/2023 ISSN: 2975-0466 [Online] 58

View publication stats