《The Interline：2024人工智能對時尚領域的變革性影響研究報告（英文版）（186頁）.pdf》由會員分享，可在線閱讀，更多相關《The Interline：2024人工智能對時尚領域的變革性影響研究報告（英文版）（186頁）.pdf（186頁珍藏版）》請在三個皮匠報告上搜索。

1、1AIREPORT#2024CoCon nc ceptuaeptual al art by Bart by Baris Genris Genc ce el l2WELCOME TO THE BEGINNING OF OUR JOURNEY TO CHART THE EVOLUTION OF ARTIFICIAL INTELLIGENCE,AND UNDERSTAND ITS PLACE IN FASHION IN 2024 AND BEYOND.BY BEN HANSON,EDITOR-IN-CHIEF,THE INTERLINEAs technology hype waves go,ther

2、es only one in recent memory that equals the fervour thats currently swirling around arti昀槨cial intelligence.And thats the dot com rush and the rise of the internet.Whether a direct comparison between AI and the web is accurate or useful is honestly beside the point:as a business you simply need an

3、AI strategy in 2024 the same way you needed a web strategy in the late 1990s.Which is to say that it feels,today,like everyone is demanding an immediate answer from you,to an evolving question where the end state is essentially impossible to predict.Try to think about it like this:without the bene昀槨

4、t of the hindsight we all now have,would you have been able to forecast when,where,and how deeply the 昀氁edgling internet was going to change our personal and professional lives?Would I?Could anyone have really seen,looking at a loose cohort of scrappy startups and scientists,how a synergy of softwar

5、e,hardware,protocol and infrastructure would unlock everything from remote working and streaming media to the gig economy and progressive web applications?With any ill-de昀槨ned question with a short timeline for responding,its extremely di昀cult to tell,in the current moment,whether your developing an

6、swer is heading in the right direction or not.Are you doing the right thing with AI here and now,considering how important it could eventually become?Is anyone?And are they doing it by design or by luck?All the same,just as the word on everyones lips in my late teens was“online,”right now investors,

7、analysts,partners,users,communities,media,lawyers-everyone wants to know where you,as an organisation,stand on AI.And like the frenzy around the web,“waiting and seeing”feels analogous to letting a potentially epoch-de昀槨ning moment pass you by.Whatever you do,the market says,you should be working to

8、 put AI in it.Not in the future.Today.INTRODUCING THE AGE OF AIThe front cover artwork for this publication was designed by Baris Gencel,an AI creator who now serves as Group Director for Digital Transformation&Innovation at Lanvin Group.Throughout this report you will 昀槨nd a mixture of traditional

9、photography,art,and illustration,as well as AI-generated elements across all those categories.The Interline will continue to commission artists for all of our downloadable reports,and we endeavour,wherever possible,to use AI tools that are trained on licensed content for which original artists are c

10、ompensated in some way-although the setup of AI services and the uncertain provenance of AI training data makes this di昀cult to guarantee.Images&art3And there are,to be clear,a lot of reasons to be amenable to that attitude.Behind the scenes,in academia,science,and R&D,deep learning,neural networks,

11、narrow models and other techniques that 昀槨t under the umbrella of AI have been achieving remarkable things for years.Theyve beaten the best human minds at even the most combinatorially di昀cult games and challenges.Theyve helped decipher burnt and otherwise unreadable ancient scrolls.Theyve uncovered

12、 novel classes of antibiotics.And yes,theyve both supplemented and supercharged human workers and,in some cases,they have straight-up replaced peoples jobs.Even before we get to the 2022-onwards explosion of interest in AI,the overall trajectory of machine intelligence as a revolutionary engine of d

13、iscovery,automation,e昀ciency and scale was on a quiet but consistent upward curve.There might not have been so many consumer-facing businesses being built on top of it,but AI was doing anything but standing still.Then generative AI blew the lid o昀昁 the entire industry-technically and commercially.Af

14、ter an apparent 100 million people signed up for ChatGPT(a Generative chatbot built on top of a Pre-trained Transformer architecture,hence the acronym)in record time,transformers and di昀昁usion models(the systems behind popular generative image and video models,that create order from noise)quickly be

15、came synonymous with“AI”for most people.Rapidly,both trade and general media were consumed with stories about this new class of AI sparking a revolution in the way we live,work,and create.There had,of course,been chatbots before,but it didnt take long for most of the world to coalesce around the rea

16、lisation that this time was going to be di昀昁erent.And equally quickly it became apparent that there was a lot of money to be made from this new category of AI tools-especially for the companies that prove themselves able to successfully read the room,sidestep cultural and ethical pitfalls,and pare d

17、own the wild possibility frontiers of large,general-purpose AI models to create focused solutions and applications that will become foundational to peoples personal and professional lives.This is why we have seen-in the space of just 18 months-an entirely new crowd of consumer and enterprise applica

18、tions being built on top of these new,always-evolving generative AI models.And even established vendors of enterprise platforms and applications have made rapid moves to integrate AI capabilities into their existing solutions where they see opportunities to either change the end user experience or t

19、o o昀昁er new ways to manipulate,interact with,and surface insights from data.So the race is on for AI toolmakers in fashion-at a scale and at a pace that re昀氁ects a similar transition the wider world is undergoing,whereby AI is breaking free of the con昀槨nes of dedicated apps and becoming integrated i

20、nto a wide range of applications,services,devices,and appliances.If we want any further evidence,from beyond the walls of fashion,that AI is now seen as the next wave of technology as a whole,we neednt look any further than the CEOs of Microsoft and Google describing it as a galvanic,species-level e

21、vent on the same order as the taming of electricity or 昀槨re.Its also certainly noteworthy that 2024 could be the last year where well have the option to buy computers,phones,and tablets that are not also billed as“AI devices”.And thats without considering the 昀氁edgling new category of wearable and a

22、mbient AI hardware designed to run small models on-device and then tap into larger ones on the cloud.This is more than just a theoretical milestone when we think about how fashion uses its top-end computing hardware.Today,the most acute demand for power comes from 3D and digital product creation tea

23、ms,who need dedicated GPUs(graphics processing units)for local simulation and rendering.Are these same teams about to start clamouring for new hardware with both GPUs and NPUs(neural processing units)to run the AI assistants designed to help them with inspiration,material choices,pattern development

24、 and so on?Its precisely this line of thinking that places me squarely in the“AI optimist”camp,because of the simple rarity of innovation happening on this kind of scale.Perhaps coincidentally,the introduction of the 昀槨rst GPU happened in the late 1990s-around the time the dot com boom was at its ap

25、ex.And the changes brought about by that introduction of task-speci昀槨c hardware are hard to overstate,from the predictable(huge leaps in real-time computer graphics)to the unintended(cryptocurrency mining).When an entire industry aligns around the idea that a software advance is momentous enough to

26、demand a completely fresh approach to hardware and infrastructure,change tends to happen quickly.4Which is why I personally believe its right to be excited about the potential of AI in general.And which is especially why I believe that AI is going to have a marked impact in fashion,since ours is an

27、industry already deep in the throes of digital transformation,where new technology is meeting tradition and creativity every day-and also an industry where automation and e昀ciency are high on the agenda.This positivity is evidenced by the sheer number of applications and solutions that are now avail

28、able for both domain-speci昀槨c deep learning tasks and more general-purpose analysis and assistance.From language,image,and video models to generative pattern development,material yield optimisation,business intelligence,pricing,planning and competitive analysis solutions,AI in fashion is here to sta

29、y.And if youre able to envision using AI to streamline a task or transform the way a team works,then the odds are good that either a large tech company or a disruptive startup is already building an application to support that vision.Youll 昀槨nd many of these companies pro昀槨led,and their senior execu

30、tives interviewed,later in this report.But while Id pigeonhole myself as an AI optimist(but not an evangelist)even that pragmatically positive position can be an unpopular one when we consider the distaste that parts of the creative community have for generative AI in particular.That sentiment origi

31、nates from the fact that,as deployed in general purpose,cloud-native solutions like ChatGPT,Gemini,Midjourney,and so on,has largely been trained on a very opaque data set that,for all intents and purposes,includes the whole of the publicly available web.Contained within that vast dataset is a huge c

32、orpus of art,music,literature,architecture,design,thought,and,yes,fashion-the majority of which was ingested as training data for these models without the permission of the creators.And for artists,designers,architects,photographers,and other creators,it feels galling to see new work being synthesis

33、ed from unlicensed elements and attributes of their entire oeuvres.This is more than a philosophical complaint,too.Right now the largest AI companies are being litigated against by publishers,authors,comedians and image libraries,on the grounds that the training of models like DALL-E and GPT4(and it

34、s successors)constitutes something closer to theft than fair use.How those cases are resolved will have deep rami昀槨cations for how general purpose AI is trained,especially as more creators and aggregation platforms sign licensing agreements that,by default,include fashion content.5Much of the curren

35、t wave of AI solutions for fashion is focused on sidestepping precisely this issue,with either 昀槨netuning or bespoke training designed to ensure that copyright data does not enter or leave the con昀槨nes of the application,and that the model delivers outputs that align with a particular brands knowled

36、ge graph,heritage and identity.And there is also concurrent work happening to improve the attribution and traceability of AI-generated content,as well as to ensure that the results put out by large models re昀氁ect the values and diverse perspectives of the communities that use them.Overall,when it co

37、mes to AI there is clearly a cultural evolution happening inside a technological revolution,and the answer to a lot of forward-looking questions about the direction of those twin forces is that nobody quite knows what will happen next.Whether were talking about attribution and copyright,capabilities

38、 and use cases,or new and untested business models,unpredictability remains the watchword.And its also important to remember that with transformer models,even the companies that created them have only a limited understanding of how they work,mechanistically speaking.We are,in e昀昁ect,barrelling into

39、the age of AI the same way we barrelled into the web era.There will be uncharted frontiers to push back,some companies will become casualties while others achieve stratospheric valuations compared to where they stand today.Like the march of the internet,fortunes will be made and broken quickly,and t

40、he eventual future will be unrecognisable in ways we cannot predict.Against that backdrop,what matters in the here and now is 昀槨nding AI solutions(or platforms that incorporate meaningful AI capabilities)that will deliver quanti昀槨able value for you,with use cases that can be successfully integrated

41、into existing work昀氁ows,teams,and communities in a way that empowers and creates trust at the same time as challenging entrenched expectations.This report-our 昀槨rst but certainly not our last to tackle AI-is designed to help provide a framework for 昀槨nding those solutions,and to building the foundat

42、ions to turn them to your advantage.Over the next 150+pages youll 昀槨nd editorials and thought leadership pieces from across our contributor and contact network(including a joint technical primer and cultural exploration of AI from yours truly)as well as pro昀槨les and interviews of leading AI technolo

43、gy vendors,and a 昀槨rst-stage analysis of the AI technology market for fashion.This has been,I realise,a very di昀昁erent introduction to what readers of our prior reports on PLM,DPC and Sustainability have come to expect.But AI represents a very di昀昁erent prospect:a radical transformation across softw

44、are,hardware,creativity,culture,and much more,a昀昁ecting fashion itself and the wider world it pulls from and exerts a pull over.I hope this report helps you to understand what AI is,why it has become such a visceral force so quickly,and what you can do,today,to translate possibility into practical r

45、eality.The Interline will be here throughout the journey.6Technology typically takes time to reach everyone.With fashion technology being one of the cornerstones of our live events,weve observed how new innovations-across software,hardware,and material science-steadily spread and scale from initial

46、idea to wider awareness.Arti昀槨cial intelligence(AI)is di昀昁erent.We are now just over eighteen months since the release of ChatGPT,and everyone in fashion is not just talking about AI-a lot of them are actively using it in their personal and professional lives.And the designers,brands,retailers,suppl

47、iers,and service providers that make up the global MMGNET Group audience are all working out how to deploy AI to their advantage.The pace of technological progress here is unprecedented.In the generative AI space,text,image,and video modalities all took huge leaps forward in 2023 and 2024.And across

48、 deep learning,computer vision,analytics and insights,the longer-standing applications of AI and machine learning made quieter but maybe even bigger strides-with data science now touching essentially every part of fashion.But the speed of AI adoption means that vital conversations around its use als

49、o need to be fast-tracked.From creative designers and independent entrepreneurs,to seasoned merchandising,technical,and sourcing teams at major brands,fashion professionals everywhere have a litany of unanswered questions and concerns about AI,as well as their own unique visions for what they want t

50、o be able to do with it in the short and longer term.What could AI mean for creativity?How will it in昀氁uence culture?How is it being regulated?What are the copyright implications of AI-generated works?Which are the best tools to use?How is AI reshaping education and employability?How heavy is the en

51、vironmental footprint of AI use?How much of the potential we hear about it is real?And how much of the fear 昀氁oating around is justi昀槨ed?All these critical conversations surrounding technology and more,have a home at MMGNET-across a new,wider remit that covers our live event brands MAGIC,SOURCING at

52、 MAGIC,COTERIE and PROJECT,and an always-on,year-round stream of trends,insights,inspiration,resources and partnerships like our long-standing one with The Interline.Everywhere fashion meets,MMGNET is committed to creating an ecosystem that not only explores the next waves of enterprise technology i

53、n fashion,but also engages,educates and empowers everyone to move the industry forward.MMGNET Group and The Interline will continue to work together throughout 2024-in established and new ways.The next opportunity to see fashion technology in action,and The Interline on stage in North America,will b

54、e at SOURCING at MAGIC in Las Vegas,which runs from 19th to 21st August 2024.BY KELLY HELFMAN,PRESIDENT,MMGNET GROUPAI IS ON THE FAST-TRACK.ITS TIME FOR FASHION TO TALK ABOUT IT.7The A-Z of AI10CONTENTS192332404856The InterlineThe Evoliving Creative ToolkitThe Interline&Baris Gencel,Artist,and Group

55、 Director,Digital Transformation&Innovation,Lanvin Group Fashion in 2050:Picking Up Where We Left Off Or Starting Anew?Jonathan Brun,Co-Founder&CEO,O昀昁/ScriptChanging RolesMacKenzie Ryan,Investigative JournalistFashion Has Been Tarnished By Inequality.Can Generative AI Help Deliver A More Equitable

56、Future?Aasia DVaz Sterling,Central Saint MartinsIs Fashion Education Ready for AI?Emma Feldner-Busztin,News&Features Editor,The InterlineIf AI Is The Future Of Enterprise Technology,Then Energy,Compute And Infrastructure Are The Currencies Of TomorrowKevin Cochrane,Chief Marketing O昀cer,Vultr8Cont.7

57、0747889179Redefining Corporate Leadership In The Age Of AIEric Huiza,Global Chief Technology O昀cer,Aionic DigitalThe AI Fashion Content Revolution:Is Your Brand Ready?Bryce Quillin,PhD&Jessica Quillin,PhD,Its A Working Title LLCGetting Serious,From Fun And Fear To Foundations And FinanceBen Hanson,E

58、ditor-in-Chief,The InterlineMeet The Key PlayersPro昀槨les and interviews for companies that are working to realise the potential of AI in fashion.AI Market AnalysisBuilding a baseline understanding of the market for AI solutions in fashion.CONTENTS60AI Is Ready To Disrupt Digital TransformationMark H

59、arrop,Founder&CEO,WhichPLMGoor Moshe,Business Development&AI Advisor9FASHIONS NEWEST RESOURCELAS VEGAS August 19-21NEW YORK September 22-24NEW YORK September 22-24LAS VEGAS August 19-21NEW YORK September 22-24LAS VEGAS August 19-212024learn more 10FROM TECHNICAL FUNDAMENTALS TO CULTURAL CONTEXT,STAR

60、T YOUR JOURNEY WITH AN ALPHABET OF KEY AI TERMS EACH ACCOMPANIED BY VALUABLE INSIGHTS INTO WHEN,WHY,AND HOW THEYRE GOING TO MATTER,AND ANYTHING ELSE YOU SHOULD BE AWARE OF.The A-Z of AI11IS FOR ATTENTIONWhile“GPT”(de昀槨ned later in this primer)has become the snappy name for the new wave of AI models

61、and applications,perhaps the biggest unlock behind that acronym was the mechanism of attention.Rather than treating the full scope of input data its given as a homogeneous whole,attention is a form of dynamic weighting that allows AI models to adjust their focus to prioritise the most relevant parts

62、 of the data theyre given(whether it is a text or image prompt).By assigning attention to what it believes is the most relevant part of a question,for example,a model can deliver an answer that feels considered and organic.This is a fundamental part of why large language models feel alive in a way p

63、revious AI chatbots havent,and the seminal“Attention Is All You Need”paper is often cited as the catalysing event of the new AI era.IS FOR BIG DATADespite being a well-worn phrase,“big data”has taken on a new importance over the last couple of years.For a long time,enterprises were told to hoard dat

64、a at all costs,with information being the new oil.Then the pendulum swung towards better-quality data,and many organisations found themselves sitting on huge data lakes 昀槨lled with non-normalised information that required deep e昀昁ort to use.Now that pendulum sits somewhere in the middle:for AI train

65、ing and use,the volume of data available matters a great deal,but the transformative,domain-speci昀槨c applications of AI will demand rigorous categorisation,labelling,organisation,and integration of business data.AI might be everywhere,but we are far from a universal acceptance of what it means,what

66、it should do,where it should act,and how it should interact with culture and society.This report is designed to begin tackling those questions,and to interrogate the people behind the solutions that promise to contextualise AI and make it work for fashion.To help parse and analyse the stories,pro昀槨l

67、es,and interviews that make up the rest of this report,our team put together an A-Z of common AI terms-and asked AI to illustrate them.IS FOR CREATIVITYIrrespective of how you personally feel about it,the wide availability of generative AI applications has already changed the connotations of the ver

68、b“to create”.Prior to the release of generative image models,for instance,someone would have to learn to draw in order to turn an idea into digital art,or a vision into a garment sketch.And prior to large language models,a person would need to learn to write well in order to compose a poem or a repo

69、rt-or would need to take a programming class to enable them to start building an app.None of these things are true any longer.And while the quality of what AI can generate is extremely variable,we have nevertheless seen a fundamental shift in the skills people must learn,and the steps they must take

70、,to translate a thought or a brief into a reality.To deliver these capabilities,though,most large language and image models have been trained-often without permission-on huge volumes of pre-existing creative work made by humans.This has been a very contentious practice,and one that is being actively

71、 litigated today across the creative industries.12IS FOR DIFFUSIONThe key invention behind image,audio,and video generation models,di昀昁usion architectures work by starting with pure visual noise and iteratively removing it to create the desired end result.They are trained to do this forward and in r

72、everse,beginning with a clean original image,before randomly adding noise to it until the original is“lost”-after which the model learns to remove that noise incrementally until the clear image is recovered.When they are asked by an end user to generate a new image or video frame,a language model pa

73、ired to the di昀昁usion model applies attention and other techniques to distinguish intent,and a di昀昁usion model applies weighted de-noising steps to progressively translate a prompt into a 昀槨nal result that comes as close as possible to the users vision,however complex or unprecedented.IS FOR ETHICSA

74、I initiatives roll up a lot of pre-existing ethical,privacy,and data governance concerns.The information that goes into training a model,the output that model creates,and the user-facing application layers built on top of those things,are all subject to the same strictures that apply to other softwa

75、re and services.But AI projects are also uncharted ethical and governance territory since they interact with untested expansions of the de昀槨nition of copyright,are being scrutinised for potential harms and biases that extend to the whole-society level,and are redrawing some of the frontiers of socia

76、l acceptance.On top of these concerns,the speed of advancement of frontier AI models is-in many peoples opinion-not being counterbalanced by investment in AI safety,with OpenAIs dissolution of its“superalignment”team being the most recent lightning rod for criticism.At present,the AI industry is sti

77、ll largely self-policing-a situation that should give our readers in fashion pause,since they will be familiar with how comprehensively an industry can fall under the microscope of external regulation and enforcement.IS FOR FINE TUNINGThe generative AI touchpoint most people are familiar with is a l

78、arge,cloud-hosted language or image generation model which has been trained to be as broadly capable as possible.This adaptability is key to the perception of limitless potential that has driven so much of the adoption of generative AI-with the drawback being that the largest,most broadly-useful mod

79、els are not always well-suited to domain-speci昀槨c tasks.Fine-tuning is the process of beginning with a pre-trained model and then applying additional training with a more focused dataset(a library of all historical eCommerce product photography,for example)to enable the model to creates outputs that

80、 are on-model or more closely aligned with brand heritage and values.This is not the same thing as training a bespoke model,which is a time,cost,and compute-prohibitive task for most organisations,but it does provide the ability to tap into the power of large generative models without being forced t

81、o rely solely on their training data.Alternative approaches include grounding and retrieval-augmented generation,which pair generative models with external data sources to be queried during use.Built into this seemingly-simple process is a huge amount of complexity and power,which is why image gener

82、ation models in particular have made such dramatic progress in the last few years.13IS FOR GPTWhile OpenAI has begun to use the word“GPT”to refer to a tailored and lightly 昀槨ne-tuned chat frontend on one of its large language models(the“GPT Store”is a place users can interact with these“GPTs”,which

83、have di昀昁erent personas,capabilities,and data sources for speci昀槨c purposes like maths tuition,for example)for the wider world GPT refers to the core architecture behind the large language models that kick-started the current AI boom.That acronym stands for Generative Pre-Trained Transformer,which i

84、s a fairly straightforward description.These models generate an output,such as text,after being pre-trained to recognise patterns and structures in a huge quantity of source data-and they achieve this thanks to their transformer neural network architecture.IS FOR HOSTINGFor most of the last eighteen

85、 months,AI models have been large in scope,capability and parameter count,and required a large amount of compute to run inference(see below).As a result,the only viable place for these models to be hosted was on cloud platforms like Azure and AWS.Open-source models such as Metas Llama series,Stable

86、Di昀昁usion and others can be run on local consumer hardware,but much more slowly than all but the most forgiving use cases demanded.IS FOR INFERENCEAfter a model is trained,the process of interacting with it and querying it is referred to as inference-since the model takes the users input and infers

87、the right output,whether thats text,data,or any other modalities.Inference is,for all intents and purposes,the“runtime”of an AI model,and as a result all the typical software interaction and usability criteria apply-especially speed and accuracy.Because large models reside in the cloud and make use

88、of distributed compute at the time they are queried,the word inference is also sometimes used to refer to the cost of running a query.This is beginning to change,with the development of lighter models(SLM refers to small language models,for example)that have much lower parameter counts and that are

89、trained for narrow tasks,and which can be deployed“on the edge”-or on enterprise and consumer hardware.And as we are likely to see in consumer devices later this year,applications and operating systems will be able to dynamically decide what queries to process using dedicated neural processing hardw

90、are,and which to pass on to cloud-hosted models.(The training of large models remains the preserve of massive supercomputer clusters,and has propelled a new hardware race that elevated chipmaker nVidia to a$3 trillion USD valuation in June 2024.)IS FOR JPEGWhile a lot of attention is being paid to t

91、ext and voice-based AI models and applications(which are,after all,how most people will soon interact with a new class of“assistants”and“copilots”),the dominant modality for fashions use cases is likely to be image generation.This may come in the form of generative product and lifestyle photography,

92、generative sketches and inspirations,generative materials and prints,or even generated garment patterns.Many of fashions potential use cases for AI have a visual component front and centre,making a focus on improving the e昀ciency and accuracy of these modalities key to creating a better market 昀槨t f

93、or AI.14IS FOR KNOWLEDGE GRAPHIn addition to the volume and variety of data required to train or 昀槨ne-tune an AI model for fashions purposes,most brands,retailers,and suppliers want to interact with AI that understands the structure,hierarchy,and context of their information and operations.The simpl

94、est way to think about this requirement is as the need to o昀昁er an AI model a metaphorical graph or index that connects the nodes of information that exist within your business with contextual and conceptual links-allowing an AI application to then understand the relationships between data that are

95、not literally linked.IS FOR LABELLINGAs part of the AI pre-training process,models need to be fed not just an avalanche of data,but an avalanche of properly tagged data that allows the training process to pick up on both the broad and the granular attributes it needs to identify patterns and make pr

96、edictions.In order to understand the world,this labelling has to be done by human beings who have the contextual understanding to identify the di昀昁erent components of images,as an example.This labelling process is not something AI companies or data brokers choose to talk about-not least because the

97、work is seemingly sometimes done by underpaid,overseas workers who are also forced to wade through extremely unsavoury data to accomplish the task.And it is easy to see a through-line where garment workers in low-cost labour areas are transferred into data tagging roles for models that must be train

98、ed on vast amounts of apparel and footwear datasets where construction and material knowledge is helpful.IS FOR MULTIMODALIn AI,a modality refers to the type of input and output that a model can process and generate.Single-mode models,for instance,would take in text and output text(ChatGPT was one o

99、f these in its 昀槨rst instantiation)or take in images and output the same.Increasingly,AI models are becoming multi-modal to varying degrees,being capable of taking text prompts as inputs and returning images,video,code,or voice as outputs,or even vice versa.Until very recently,though,these di昀昁erent

100、 modalities were accomplished by quietly bridging disconnected models and systems in the background.If you spoke to ChatGPT it would not receive your voice as a direct input,but rather a translation of your voice into text.Or if you uploaded an image to a large language model like ChatGPT and asked

101、it to generate a variation of it,it would not pass that image directly to the di昀昁usion image generator,but would instead send over a text description of that image that the di昀昁usion model would use as a prompt.The latest generation of chatbot systems(GPT4o,and Googles upcoming versions of Gemini)h

102、ave been rearchitected to be whats referred to as“natively”multi-modal,which gives them the ability to directly accept text,audio,video,photo and other types of inputs,and to respond directly in the form of any of those modalities.A simple example might be uploading a photo of a piece of sheet music

103、 to the upcoming GPT4o voice mode,and having it sing the notes back to you.In fashion,di昀昁erent modalities are a consistent feature of almost every stage of the product lifecycle,so any attempt to realise a product design and development AI assistant would need to be natively designed this way.15IS

104、FOR NEURAL STYLE TRANSFER An image generation technique,neural style transfer is a way for generative models to blend the content of one image with the stylistic elements,colours,or other attributes of another.In applications where experimentation and iteration are common objectives,these abilities

105、have the potential to allow for quick and e昀cient ways of introducing newness and alternative options into the creative process.On the opposite end of the spectrum,there are many fashion use cases where one part of a particular image or other piece of content must remain absolutely unchanged from on

106、e generation to the next.This consistency proved to be a serious challenge for the 昀槨rst wave of generative tools and applications,but the introduction of Low-Rank Adaptation(or LoRA)has allowed users to essentially ringfence an element and then to run generative operations around it.Applications th

107、at allow users(in-house or even at the point of retail)to“stage”products in di昀昁erent settings,lighting conditions,and usage scenarios rely on this technique,and are a prime example of how innovation is helping to overcome the common criticisms that are levelled at generative AI.IS FOR OPEN SOURCE D

108、espite its name,OpenAI does not o昀昁er its models as open source.And the same is true,in fact,of most large,high-parameter-count models:all are proprietary software that users can only access through their parent companies applications,or by making API calls.Meta,notably,is the biggest company o昀昁eri

109、ng its current(but potentially not future)large models to the open source,developer,researcher,and hobbyist communities,although Mistral and other companies have also released open-source versions of their smaller models.For fashions purposes,this debate matters primarily for companies that prioriti

110、se data sovereignty and ownership,and who do not wish their data to leave their premises,and for the global talent pool of researchers and computer scientists who are 昀槨nding themselves pushed and pulled between lucrative proprietary contracts and more widely-applicable open-source work.IS FOR PARAM

111、ETERS In the AI lexicon,parameters are functionally the same thing as model capabilities.In the push towards“scaling,”model creators have pushed parameter counts higher,and there has so far been a strong correlation between the number of parameters and the quality of a models output.Increasingly,tho

112、ugh,developers are building models on divergent tracks,for di昀昁erent purposes.Meta,for instance,is currently training a 400 billion parameter model at the same time as working to release an 8 billion parameter model-both built on the same architecture,but with greater optimisation,streamlining,and f

113、ocus making the smaller models more easily deployable and less resource-intensive.16IS FOR QUERY OPTIMISATION Fashion,and the wider world,is witnessing a transformation in how end users engage with technology.Since the advent of computing,it has been a truism that telling a computer what to do as a

114、user-through a text-only interface,or later through a GUI-will translate into a predictable outcome.Barring bugs and some of the unevenness of display technology and calibration,inputting a CMYK colour reference using a keyboard will always result in the desired colour being shown.Making the same re

115、quest of an AI model-to alter the colour of a material in a 3D design,for example-will mostly result in a predictable action,but its not a guarantee.And this new interface paradigm of asking questions of computers is already presenting a roadblock to users who feel as though they dont know the right

116、 things to ask.Query optimisation can refer to the manual process of getting better at extracting a desired action from an AI model-an upskilling exercise-but it can also refer to the way that multimodal models can re-interpret,embellish,and enrich users prompts behind the scenes to help deliver bet

117、ter end results.Which,in and of itself,represents yet another change in how we think about engaging with software-and further uncertainty for an industry that relies on precision.IS FOR REGULATIONSAt the time of writing,AI regulations operate on a sliding scale,where models,applications,and services

118、 that are perceived to present the greatest risk.For the time being,while the true impact of AI across cultures and communities is still being assessed,this risk is being measured in the potential for harm and misinformation in high-impact scenarios,but over time we expect that regulations and discl

119、osure requirements will begin to demand greater transparency into the training,operations,and outputs of AI models.IS FOR SYNTHETIC DATAThe demand for data thats voluminous enough,speci昀槨c enough,and labelled in su昀cient detail to train the next cohort of AI models is already beginning to outstrip s

120、upply.After scraping or licensing content from publishers and social media platforms,and capturing what content is left on the open web,AI companies will be left with a stark choice:to trawl less desirable data sources,or to begin making use of whats referred to as“synthetic data”.The latter is arti

121、昀槨cial data that is intended to replicate the features of real data,and is procedurally or arti昀槨cially generated according to a clear set of parameters(in the non-AI sense)that ground it in reality.A hypothetical example of this might be an autonomous vehicle project that has been trained on su昀cie

122、nt footage of every road that exists,but that has not yet achieved its aims,so its creators consider training it on rendered or AI-generated footage of roads that do not exist but that otherwise accord with all the laws of reality.This drive to augment the pool of training data available to AI model

123、s in this way could back昀槨re for general AI(where hallucinations could become baked into the training data of future models)but could prove to be a positive thing for fashion.If a model has already been trained on your entire back catalogue,for example,digitally creating some 昀槨ctitious remixes,reco

124、lours,or other ideas could help to increase the creativity of models without stretching the limits of accuracy.17IS FOR TOKENSAs used in AI parlance,a token is a basic unit that an AI model uses to take in or generate language.A single token does not necessarily correspond to a single letter or word

125、,and di昀昁erent models approach the segmentation of tokens,and the relationship between them,di昀昁erently.Tokens are also used as a measure of AI work,with both large language models like ChatGPT and enterprise platforms like Cohere pricing their API usage based on token throughput.IS FOR UNSUPERVISED

126、While many datasets used in AI are explicitly labelled,and many training runs are performed with a target outcome in mind,there are also examples where its desirable to have AI models learn the structure of,and derive insights from,a large volume of unstructured,or fast-changing,data that has not be

127、en labelled and where a human has not set an explicit intention.As an algorithmic approach to discovering patterns,grouping undi昀昁erentiated items,and delivering insights at a speed and scale impossible for humans,unsupervised learning has uses across fashion-from classifying items that have only ex

128、tremely 昀槨nite,or subjective di昀昁erences,to identifying trends and market forces.IS FOR VARIABILITYAs we saw earlier in“Query Optimisation,”the fundamental di昀昁erence between AI models and traditional computer programs is their ability to produce di昀昁erent outputs when fed the same input.Unlike dete

129、rministic interactions like coding,or moving a 昀槨le from one directory to another,interacting with an AI model is an exercise in variability-where hidden and opaque factors can in昀氁uence the outcome.For basic interactions this can create frustration,and in situations where accuracy is paramount,or s

130、uccess is binary and objective,any deviation from it(the fabled“hallucinations”)is considered a failure.But at the same time this variability is also why generative AI models can be so compelling in creative use cases,since even the most minor changes in inputs can deliver drastically di昀昁erent outp

131、uts,and experimentation is a matter of changing a word or two in a prompt.IS FOR WORKFLOW AUTOMATIONOn the aggregate level,fashion does not have great work昀氁ow automation.In The Interlines interaction with brands and retailers,it is rare to see a company that has a complete,end-to-end view of its de

132、sign,development,sourcing,production and retail activities.At least in part,this is because of the complex nature of those processes and the dynamic,distributed shape and scope of the data they generate.We are already beginning to see AI models applied in this capacity-from Cohere(mentioned above)to

133、 Glean in the sector-agnostic space,and several examples contained in the technology vendor section of this report.With so much of fashion brands time and e昀昁ort spent reconciling disconnected work昀氁ows,and attempting to derive real-time insights from incomplete and shifting business data,the promis

134、e of being able to apply an AI model across an entire enterprise-and then to interact with and query it,with the knowledge that its grounded in your data-is perhaps where the biggest potential of AI in the business side of fashion will be found.18IS FOR EXPERTISEAs AI begins to test the boundaries b

135、etween what should remain human work and what is a target for automation,there is still considerable uncertainty in the jobs market and in education(more on this later in the editorial section of this report),and even deeper concern amongst creative,technical,and commercial workers as to where the 昀

136、槨nal balance will be struck between human e昀昁ort and AI generation.At the time of this report,the threatened mass of AI-related job losses has not yet manifested,and there may indeed be room for some optimism.Other disciplines,like programming and software engineering,have found themselves in the 昀槨

137、ring line,but so far AI is being largely used as an aid to productivity rather than a full-blown replacement.This may also be the pattern we see in fashion;while some junior roles,and some senior ones,will undoubtedly be eliminated as a result of AI adoption,the same models and applications may make

138、 for a smoother onramp for new entrants.Despite machines having surpassed human capability in chess,that game is now more popular and accessible than ever,and people and AI working together have been able to begin dissecting its art and its mechanics in more detail.IS FOR YIELDYield can be de昀槨ned i

139、n two ways.First,as a measure of the e昀ciency and optimisation of AI models and systems,quantifying the ability of an AI to produce relevant outputs at speed and scale.And second in its longstanding usage as a yardstick for the e昀ciency of nesting and material cutting and utilisation,and manufacturi

140、ng processes-both of which could be augmented by sensitive,intelligent use of AI.IS FOR ZEROSHOTThe jury may still be out on whether the new batch of generative AI models can be said to be“intelligent,”but the fact remains that we have entered into a new era of both enterprise and consumer technolog

141、y where systems and assistants available to us on a range of devices are capable of performing tasks,generating outputs,or making predictions that they have not been expressly trained to do.This ability,referred to as“Zero-shot”learning,provides some evidence against the criticism that generative is

142、 a kind of“fancy autocomplete”that simply parrots extremely varied permutations of language it has been trained.For these models to instead create meaningful language,accurate insights,or identi昀槨able objects in images in areas they have never encountered before,the more logical conclusion is that g

143、enerally capable models can be capable of delivering genuinely novel results in unseen and unpredictable use cases.And where fashion is concerned,unpredictability is everywhere-making generative AI not just a unique but a potentially extremely useful prospect.19The Interline:Before your current role

144、,you spent a long time spearheading AI design as an art form in its own right-one thats capable of having a tangible positive impact in multiple different areas.How do you feel the fashion industrys recognition of that potential has developed over the last two years?And how much of that positive pos

145、sibility space do you think has been realised?Baris Gencel:Over the past two years,I have not seen a signi昀槨cant shift in the fashion industrys recognition of AI as an art form and its potential for positive impact until recently this year.Initially,there was a degree of skepticism and resistance,wi

146、th many viewing AI primarily as a tool for copying rather than creativity.However,this perception is evolving rapidly.Fashion houses and designers are increasingly acknowledging the innovative capabilities of AI,embracing it not just for optimizing supply chains or enhancing marketing strategies,but

147、 as a creative partner that can inspire and generate new ideas.AIs integration into fashion has opened up numerous possibilities.For instance,it has enabled more sustainable practices through predictive analytics and resource management,reducing waste and promoting eco-friendly production.AI-generat

148、ed designs have also pushed the boundaries of creativity,leading to unique and avant-garde collections that might not have been possible through traditional methods alone.Despite these advancements,we are still only scratching the surface of AIs potential in fashion.Many positive possibilities remai

149、n untapped.As AI technology continues to evolve,I believe we will see even more profound impacts,such as hyper-personalized customer experiences,AI-driven fashion shows,and deeper collaborations between human designers and AI systems.The journey is ongoing,and the future holds immense promise for fu

150、rther integrating AI into the core of fashion innovation.The Interline:On a similar topic,you recently spoke at the United Nations“AI For Good”summit.What were your key takeaways from that conference in terms of how state-level actors are thinking about the positive impacts that AI could have?Baris

151、Gencel:First of all,I am honored to have received the AI for Good Award for my art.This UN initiative means a lot to me.I had the privilege of attending the UNs AI for Good Summit and gave a speech that I feel was very inspiring.The summits focus was on how AI can be used to meet the UNs Sustainable

152、 Development Goals,such as eradicating poverty and hunger,achieving gender equality,promoting clean energy,and taking climate action.The designer of this reports cover art(and the images accompanying our Market Analysis later in this publication),Baris Gencel,was an early advocate for the potential

153、of AI to serve as a creative partner for designers.And as the pace of AI adoption has picked up,his perspective on how to e昀昁ectively incorporate it into the artistic process has become increasingly sought-after.Baris recently won an art award and spoke at the United Nations AI For Good Summit,an ev

154、ent focused on how AI could be used to accelerate the UN Sustainable Development Goals.And in his role as Group Director for Digital Transformation and Innovation at Lanvin Group,Baris also has a unique perspective on how AI is becoming a key pillar of leading brands all-round digital transformation

155、.On the eve of the publication of this report,we sat down with Baris to discuss process,possibility,and why he believes AI will change every element of fashion.INTERVIEWEE:BARIS GENCELARTIST,AND GROUP DIRECTOR,DIGITAL TRANSFORMATION&INNOVATION,LANVIN GROUPTHE EVOL VING CREATIVE TOOLKIT20Leaving the

156、summit,I felt a mix of con昀槨dence and determination.AI has the potential to play a meaningful role in advancing these goals,but we must ensure that all leaders,governments,and citizens of the world push stronger for change.AI should be a tool that decreases our impact,inspires us,and creates works t

157、hat make people care more about our planet and all the species we share it with.The Interline:You designed the artwork for our front cover as a new entry in a thematic series.Walk us through some of the inspirations behind the image,and the tools and processes you used to bring those ideas to life.B

158、aris Gencel:The cover artwork for Interlines AI Report 2024 represents a fusion of technology,human creativity,and the evolving landscape of digital transformation in fashion.Every work we create is a culmination of the experiences we build up in our lives.Our creativity thrives on combining known e

159、lements to form new ones,and this boundary of imagination based on our experiences is what I have been exploring.Having spent 23 years in Asia,I have been profoundly in昀氁uenced by its rich cultural heritage.The tattoos on the 昀槨gures body merge traditional art with futuristic themes,symbolizing the

160、bridge between past and future.This fusion re昀氁ects the idea that while technology advances,it remains deeply rooted in cultural and artistic expressions.The tattoos add a layer of personal identity and storytelling to the otherwise mechanical form.The Interline:Right now,in 2024,how do you believe

161、the roles of the creative designer and creative director have changed as a result of generative AI?And how do you see them changing further in the near future?Baris Gencel:As AI becomes more integral to design,educational institutions will increasingly incorporate AI training into their curricula.Em

162、erging designers will learn to work with AI from the outset,making AI-assisted design a standard practice.This shift will produce a new generation of designers who are adept at leveraging AI to enhance their creative processes.by the integration of generative AI,leading to enhanced creativity,effici

163、ency,and personalization.Looking ahead,these roles will continue to evolve as AI becomes even more embedded in the design process.By embracing AI as a collaborative partner,focusing on sustainability,and committing to continuous learning,designers and directors can navigate this transformation and d

164、rive the future of fashion innovation.The Interline:If you were to speak to an emerging designer,or someone currently in education and hoping to move into a career in fashion,what advice would you give them about developing a creative process and a skillset thats fit for the future?Baris Gencel:Unde

165、rstanding and integrating technology into your creative process is essential.Learn about AI,3D design software,digital pattern-making tools,and virtual reality.These technologies can enhance your creativity and e昀ciency,enabling you to experiment with new ideas and techniques.Stay informed about the

166、 latest advancements and consider how they can be applied to your work.Conceptual art by Baris Gencel21Focus on Sustainability:Sustainability is no longer optional in fashion.Educate yourself on sustainable practices,materials,and processes.Think about how you can incorporate eco-friendly methods in

167、to your designs and advocate for ethical production.Consumers are increasingly valuing sustainability,and being knowledgeable in this area will set you apart.Cultivate a Diverse Skillset:The fashion industry is multifaceted,and having a broad skillset will make you more versatile and valuable.In add

168、ition to design,learn about business,marketing,and technology.Understanding the entire fashion ecosystem will enable you to navigate it more e昀昁ectively and seize diverse opportunities.Develop a Strong Personal Brand:In a crowded industry,having a distinctive personal brand is crucial.Use social med

169、ia and digital platforms to showcase your work and share your design philosophy.Build a portfolio that highlights your unique style and creativity.Networking is also essential;connect with industry professionals and participate in relevant events and competitions.Stay Curious and Open-Minded:The fas

170、hion industry is constantly evolving,and being adaptable is key.Stay curious about new trends,technologies,and cultural shifts.Be open to feedback and willing to experiment with di昀昁erent styles and concepts.Continuous learning and adaptability will help you stay relevant and innovative.The Interlin

171、e:What do you believe is the right approach for brands looking to translate some of the possibilities of AI into lasting institutional transformation?Where should they start,and what do you see as the areas where AI has the potential to add the most value in the shortest timeframe?Baris Gencel:In my

172、 opinion,the right approach for brands looking to translate the possibilities of AI into lasting institutional transformation involves a strategic and phased implementation.By starting with a clear strategy,leveraging data,initiating pilot projects,upskilling employees,and scaling AI initiatives,bra

173、nds can translate AI possibilities into lasting institutional transformation.This approach not only drives immediate value but also sets the foundation for sustained growth and innovation.Encourage a culture of innovation and experimentation.Employees should feel empowered to explore new ideas and u

174、se AI to solve problems creatively.AI can analyze vast amounts of data to identify emerging trends and consumer preferences,guiding the development of new products.AI-powered design tools can assist in creating innovative and efficient product designs.AI can automate repetitive and mundane tasks,fre

175、eing up employees to focus on more strategic activities.The Interline:In your role as Director of Digital Transformation,how do you believe AI fits into the broader ecosystem of technology-enabled and technology-adjacent change?A lot of brands and luxury houses are already several years into transfo

176、rmation journeys that are built on digital workflows,digital assets,digital twins and digital experiences.How do you see AI slotting into those?And how much might it accelerate them?Baris Gencel:In my role as Director of Digital Transformation,I see AI as an indispensable force in the ecosystem of t

177、echnology-enabled and technology-adjacent change.AI seamlessly integrates with existing digital work昀氁ows,assets,twins,and experiences,propelling them to new heights of innovation and e昀ciency.At Lanvin Group,we are committed to embracing new technology,with AI being a key focus area.Our CEO and the

178、 groups vision emphasize leveraging cutting-edge innovations to drive our brands forward.This commitment ensures that AI is not just an addition to our digital transformation toolkit but a catalyst that enhances every aspect of our journey.By integrating AI into digital work昀氁ows,assets,twins,and ex

179、periences,we aim to unlock new levels of e昀ciency,creativity,and customer engagement.The potential for AI to accelerate these transformations is immense,positioning us to lead in the dynamic landscape of luxury and fashion.LANVIN GROUPs brands all have a long history and unique brand DNA.Moving forw

180、ard,we will continue to innovate with AI while maintaining the unique DNA of each brand,allowing more people to understand the profound heritage of our brands.In conclusion,AI is integral to our strategy at Lanvin Group.By harnessing its power,we are poised to achieve signi昀槨cant advancements and de

181、liver exceptional value to our stakeholders.Our vision is clear:to be at the forefront of innovation,setting new benchmarks in the luxury fashion industry.22The Interline:For some readers,the positioning of AI as the next wave of consumer and enterprise technology may feel like a repeat of the hype

182、that was built around the visions for the metaverse,and for Web3,over the last few years.Why do you believe this wave is different?And what would you say to anyone who might find themself fatigued by a series of bold visions for change that do not necessarily pan out?Baris Gencel:The wave of AI is d

183、i昀昁erent because it is backed by a proven track record,rapid advancements,seamless integration capabilities,adaptability,and wide-ranging applications.It o昀昁ers tangible bene昀槨ts that are already being realized across industries.For those wary of new technological promises,AI stands out as a mature

184、and transformative force that is not only here to stay but also poised to drive signi昀槨cant,lasting change.By focusing on real-world applications and measurable outcomes,brands can harness the power of AI to achieve lasting institutional transformation,ensuring that this wave of innovation leads to

185、sustained success and growth.The Interline:What do you see coming next from AI?With the speed at which existing models are evolving,and the pace of development on new models,new modalities,and new applications-what do you see as the future of AI in fashion in the next two years,the next five years,a

186、nd the next ten years?Baris Gencel:The future of AI in fashion is incredibly promising,with rapid advancements expected to transform the industry at every level.From enhancing personalization and sustainability to driving creativity and ethical practices,AI will be a catalyst for innovation and grow

187、th.By embracing these technological advancements,fashion brands can position themselves at the forefront of the industry,delivering exceptional value to customers and stakeholders alike.In the immediate future,AI will further revolutionize personalization in fashion.Brands will leverage AI to create

188、 highly personalized shopping experiences,using advanced algorithms to analyze consumer data and preferences.This will enable real-time customization of product recommendations,styling advice,and even bespoke designs tailored to individual customers.AI will play a signi昀槨cant role in the creative pr

189、ocess.Designers will collaborate with AI tools to generate innovative concepts and explore new aesthetics.AI will analyze vast amounts of design data and trends to inspire new collections,pushing the boundaries of creativity while maintaining brand identity.AI will drive advancements in sustainable

190、fashion by optimizing material usage,reducing waste,and promoting circular economy practices.Machine learning algorithms will help identify eco-friendly materials and processes,enabling brands to produce high-quality,sustainable products at scale.23WHAT MIGHT A RADICALLY REIMAGINED,AINATIVE FASHION

191、VALUE CHAIN LOOK LIKE?AND HOW CLOSE ARE WE TO HAVING AN“ANSWER ENGINE”FOR THE INDUSTRYS RECURRING QUESTIONS?FASHION IN 2050:PICKING UP WHERE WE LEFT OFF OR STARTING ANEW?BY JONATHAN BRUN CO-FOUNDER&CEO OFF/SCRIPTJonathan Brun is the co-founder and CEO of O昀昁/Script.Before launching O昀昁/Script,Jonath

192、an co-founded Lighthouse,an immersive media search engine used globally.He also worked in venture capital,investing in AI and consumer sectors.Jonathan began his career in investment banking,specializing in mergers and acquisitions as well as capital markets transactions.In addition to his professio

193、nal achievements,Jonathan is a father of four and an avid martial arts practitioner.24Technological change comes in waves.This is something we all understand intuitively,because we are accustomed to having it happen to us and around us.In telecommunications,we went from 1G to 5G.In transportation,we

194、 moved from mechanical,to fossil,and now electric energy.Our monetary systems transformed from basic barter to forms like trading cards,gold,paper money,and now,digital currencies.There are thousands of examples like these.Even if the di昀昁erence between how things are and how they were in the past c

195、an feel enormous in retrospect,transformation often happens incrementally,going broadly unnoticed until we zoom out and realize the progress weve made.Yet,every so often,a breakthrough technology emerges that is so revolutionary it catalyzes an entirely new paradigm and leaps us forwards in a way th

196、ats not just perceptible but extremely prominent.These types of new frameworks do not simply build incrementally on what came before.Rather,they introduce a transformative way to rethink old problems,opening doors to solutions previously unimaginable,and they give rise to new opportunities that were

197、 impossible to envision-at least with any degree of practicality-previously.These types of innovations are black swan events that 昀槨t the adage:“there are decades when years happen,and years when decades happen”.I dont think its an exaggeration to say that 2022 and 2023 were examples of the latter.T

198、he rise of generative AI,I think(and many agree),represents such a pivotal shift.Vaulting to the front of peoples minds in late 2022 and continuing to up-end industries,our understanding of“work”and much more over the 18 months that have followed,generative AI models as experienced through user-faci

199、ng applications like ChatGPT and Midjourney are fundamentally changing how we think about not just higher-order tasks but some of the basic pillars of society.And the world of fashion is not exempt.As entrepreneurs,designers,and trendsetters,the disruptive potential of this technology forces us to q

200、uestion the current state of a昀昁airs and wonder what the world could look like if we restarted anew,leveraging the cutting-edge technology available today.How di昀昁erent would our processes,industry dynamics,and economics be?Would our products be better or worse?More human,or less?What would fashion

201、look like if we built it with AI in mind?25Would we,after some initial experimentation,wind up settling for some fairly prosaic use cases and getting incremental innovation,or would the re-imagined industrybe a transformed one,pushing us to rethink the things we considered to be hard truths about th

202、e fashion beforehand?BRANDS MAIN CHALLENGESFirst,lets ask the question:what are the biggest challenges brands face today?The brand value chain is both complex and simple at the same time.Complex in that it involves a large number of di昀昁erent parties-each with their own set of incentives.Simple in t

203、hat this framework has been,by most measures,unchanged for the last 500 years.To an outsider,the way fashion products make the journey from idea to market looks,frankly,messy.To an insider,that complexity and disorder is just the way things work and have always worked.Below is a highly simpli昀槨ed vi

204、ew of a brands value chain as it exists in 2024.I want to walk you through it,step-by-step,starting with the context of the industry itself.LAYER 0:INDUSTRY DYNAMICThe 昀槨rst challenge of a traditional brand value chain lies in its inherent competitive dynamic.Brands all compete for mindspace and sha

205、re of wallet,bidding against each other to the point where unit economics have become broadly unsustainable for most players.At 昀槨rst glance,this doesnt seem like a challenge AI can 昀槨x.Every industry that has a number of players growing faster than its market size is exposed to economic cannibaliza

206、tion and a race-to-the bottom in terms of pro昀槨tability.From a purely commercial perspective:with low(and lowering)barriers to entry,the number of brands in the fashion industry should continue to grow until a market equilibrium is reached where no one is truly making money,and where weaker actors a

207、re competed out of the market.LAYER 1:INSPIRATIONThe inspiration layer for most of us is the fun one.Its where passion lives.At this layer,designers are thinking:What should I create?What message do I want to share with the world?While this layer is obviously informed by the previous one-market real

208、ities increasingly dictate what gets made-this is also where the ine昀昁able“creative”spark sits.This layer also 昀氁ows through many stages,namely:personal experience,idea,and sketch.26Here the challenges are obvious:there is a limited supply of great ideas,getting unbiased market validation on the mer

209、its of an idea is hard,and transforming great ideas into great sketches requires technical skills that not everyone has.The technical skills required to transform an idea into a sketch reduce by several orders of magnitude the number of inspirations we get to ultimately consider to be developed into

210、 new products.LAYER 2:PRODUCT CREATIONThe product creation layer is my personal favorite.Its technical,messy,and requires a lot of coordination with outside actors.This layer is de昀槨ned by the following stages:tech packing/3D design,manufacturer selection,golden sample production(often with many rou

211、nds of iteration),inventory production.Challenges are basically all there is at this part of the process!But if one wants to be speci昀槨c,here are the key pain points most brands face:TRANSPARENCY:it is hard to 昀槨nd good manufacturers and it is even harder to know if you are paying a fair price to wo

212、rk with them.PARALLELIZATION:each brand has to do the same work as the others.Functionally the same tech pack will be designed thousands of times by thousands of di昀昁erent brands,with every one reinventing the wheel multiple times over.SKILLS:Tech packing and 3D design,visualization and simulation r

213、equire expertise,as well as dedicated hardware and software for the latter.QUALITY ASSURANCE:Finding the good manufacturing partners is costly as it involves good through multiple rounds of sampling,knowing what to look for,etc.FINANCIAL RISK:Finally,the biggest challenge of this phase is the negati

214、ve working capital involved in building inventory before knowing if you can sell it.This is the main reason why so many brands die or never see daylight:guessing what the market wants,committing to producing it in volume,and then 昀槨nding that it doesnt sell is an extremely common occurrence.27LAYER

215、3:COMMERCIALIZATIONFinally,the third layer:getting sales.This one has already been transformed by AI in a meaningful way through product recommendation algorithms,better ad targeting,dynamic pricing,etc.Commercialization can be broken down into two phases:(i)marketing-i.e.creating a purchase intent

216、for your product,and(ii)commerce-i.e.handling the mechanical aspects of completing a transaction(payment processing,ERP update,ful昀槨llment,shipping,duties optimization,after-sale support,etc).The commerce side is usually what people identify as the“boring”part of the business.The name of the game he

217、re is efficiency more than anything else.There are best practices,and for most brands,the key is to follow them diligently,not innovate.Other industries,too,have already set fairly cast-iron templates for having the most seamless routes to market.The marketing side gives more space for creativity an

218、d experimentation.Which channel to use?How to frame your messaging?How to design visuals that stand out and stick?Here the biggest challenge is optimizing ROAS(return on ad spend)by creating better promotional material,faster,and distributing it to the right people,at the right time,and in the right

219、 way.WRAPPING UP ON CHALLENGESAs weve seen,there are a bunch of challenges that brands need to overcome at the inspiration,product creation,and commercialization layers.Although there are hundreds of them individually(and any one can easily become a bottleneck that holds back the speed,quality,creat

220、ivity or other outcomes of a 昀槨nal product),we can bundle them into three core categories:FINANCIAL RISK:(i)product development risk,(ii)inventory risk,(iii)high customer acquisition costs SKILLS SCARCITY:shortage of(i)inspiration,(ii)sketches,technical,and 3D design skills(iii)quality insurance exp

221、ertise,(iv)marketing e昀ciency TIME CONSTRAINTS:the inability to move through each layer faster,or even in parallel.THESE TYPES OF INNOVATIONS ARE BLACK SWAN EVENTS THAT FIT THE ADAGE:“THERE ARE DECADES WHEN YEARS HAPPEN,AND YEARS WHEN DECADES HAPPEN”“O昀昁/ScriptO昀昁/Script28LEVERAGING GEN AI TO PROPOS

222、E A NEW FRAMEWORKNow that we better understand the inherent challenges coming with launching a fashion brand or a new collection,lets explore if and how generative AI might help us.From that perspective,the 昀槨rst question to ask is:has it done so already?Did the world truly change over the last year

223、 or so in a way thats delivered meaningful AI applications into the hands of brands and retailers?A few observations:We have already gone from a world where only a handful of trained designers can conceive breathtaking concept mock-ups to one where everyone with taste has that capability.We hear a l

224、ot about generative AIs ability to“democratize”creativity(and there are certainly strong opinions on both sides as to whether that is a desirable thing)and this is the neatest encapsulation of it.If you want to bring an idea you have for a product-no matter whether it 昀槨ts into the apparel,footwear,

225、or accessories categories-to life,you can now do so without needing to learn to draw or model in 3D.We have already moved from a world where creating captivating marketing materials and validating consumer demand could only be done post production with professional assistance,to one where AI can do

226、the job at much earlier stages of the product lifecycle,in a way that requires far fewer professional skills.We have already moved from a world where manufacturers had little visibility and exposure to end-buyers to one where data can both inform and connect them with customers.And while that data i

227、tself is not generated by AI,an increasing number of supply chain connectivity and visibility platforms are making extensive use of AI to provide clarity,accountability,and a new channel for engagement and exposure.We have already moved from a world where creatives might be able to collaborate inste

228、ad of compete,to one where universal accessibility of digital creative tools,digital production methods,and community sourcing platforms has made it easier than ever to work together and to make use of both innovative manufacturing approaches and collective buying power across traditional onesTaking

229、 stock like this,I think its important to recognise how quickly some of the fundamentals of fashion have alreadychanged.But,building on these facts,lets consider how AI could empower us to propose a completely rethought value chain for emerging brands.The framework proposed on the next page obviousl

230、y has its own 昀氁aws and does not claim to be the answer to all of the industrys challenges.However it is a new structure only made available by the rise of generative AI weve witnessed over the last year-and by linking those new possibilities into some of those pre-existing innovations I listed in d

231、esign,marketing,and manufacturing.29I believe this approach is interesting not just because of its innovative approach,but because it also o昀昁ers a realistic way for emerging entrepreneurs to o昀oad risk.For the moment,lets call it the Community-Led Brand Framework,and lets consider how it would di昀昁

232、er from where fashion 昀槨nds itself today.LAYER 0:DYNAMICS The traditional competitive landscape,where brands vie for market share and visibility,is transformed into a collaborative ecosystem.In this community-led dynamic,every new player creator or consumer adds value.Network e昀昁ects,often non-exist

233、ent in the classical approach are now ampli昀槨ed as each participant not only consumes but also potentially contributes ideas,enriching the pool of designs.By crowdsourcing ideas and having all involved actors incentivized in growing the network,the community-led brand transforms the net zero competi

234、tive nature of a multi-brand ecosystem to positive-sum multi-creator ecosystem.This may sound unrealistic in such a cutthroat market,but we only need to glance at the communities that have grown up around online creators and hobbies,or the communities built around AI tools like Midjourney,to see tha

235、t this model is not just feasible but that it also potentially better 昀槨ts the reality of content creation and consumption in 2024.LAYER 1:INSPIRATIONIn a generative AI-infused world,the bottleneck of creativity can be removed.Anyone with taste and patience can generate an amazing product mock-up,ir

236、respective of technical skill.Market validation,previously a stumbling block,is now facilitated by a captive ecosystem of actors voting on the concepts they would like to see move forward.TO PUT IT SIMPLY:AI-NATIVE FASHION COULD GET US TO A STAGE WHERE WE DONT NEED TO ASK SO MANY REPETITIVE,RHETORIC

237、AL QUESTIONS ALONG THE ROAD FROM IDEA TO FINISHED PRODUCT.“30This measurably meritocratic community idea validation process derisks the chances of producing something nobody wants,turns passive buyers into active participants,and more importantly,allows creators to 昀槨netune their approach and design

238、s,signi昀槨cantly accelerating the feedback cycle necessary to their development as designers.There are,of course,social and copyright considerations to take into account here,but broadly speaking the closer fashion gets to culture,and the more the two are allowed to grow in symbiosis,the better they

239、will feed into one another.LAYER 2:COMMERCIALIZATION NOT PRODUCT CREATIONA signi昀槨cant shift in the traditional value chain occurs here,as commercialization leapfrogs product creation,at least up to a certain degree.With generative AI,the market validation phase can inform and precede manufacturing.

240、This is what on-demand producers like Shein are doing,but this can also be accessible to smaller brands if they join an ecosystem where speed of delivery is removed from the equation for consumers.This is a profound change.Moving to a pre-order and on-demand production model,e昀昁ectively mitigates th

241、e 昀槨nancial risks associated with inventory and unsold stock.Commercialization becomes an intelligent,data-driven process that actively shapes the products to be created.And the sustainability bene昀槨ts of e昀昁ectively removing waste could be profound.Obviously,customers will not wait forever to recei

242、ve their products,though.While shoppers are increasingly understanding that fashion cannot be cheap and fast without the planet or another person paying the ultimate cost,fashion is still driven by whims and desires just as much as by practical,long-term planning.For every time someone commissions a

243、 piece of on-demand occasionwear for a pre-planned wedding,there will be a scenario where someone needs something much sooner-just because.So is there anything meaningful that can be done with the now new layer 3,product creation,to accelerate that cycle?Could we reach a point where delivery delay d

244、oesnt exclude such a big market segment that the community-led brand model becomes e昀昁 ectively unviable.31LAYER 3:PRODUCT CREATIONProduct creation in the community-led framework must be agile,proximate,and proactive.In this new model,the generative AI concept generation model of the community-led b

245、rand leverages a visual database of manufacturers design capabilities to arrive at something producible.When creators input their concepts,the AI analyzes this visual history and steers the design towards the nuances and capabilities of the most 昀槨tting manufacturing partner-covertly matching inspir

246、ations with components,materials,and machinery.This AI-native system wouldnt be about pondering variables such as cost or timing,which are where the traditional fashion value chain spends the most time circling back on itself;instead,its a direct conduit between the creative concept and the manufact

247、urers proven expertise,ensuring a smooth transition from idea to physical product.As the AI is exposed to more historical data,patterns,operations,components and so on,its ability to match designs with manufacturers becomes even more precise,accelerating the production process without compromising o

248、n the manufacturers strengths or the creators vision-and without needing anyone to completely re-engineer a basic t-shirt.Just like ChatGPT,Perplexity,and others are opening the era of“answer engines”in opposition to“search engines”,the community-led brands ever evolving network of manufacturers can

249、 create the foundation of a new“answer-machine”when it comes to handling how to transform a given design into a fully 昀氁edged product.To put it simply:AI-native fashion could get us to a stage where we dont need to ask so many repetitive,rhetorical questions along the road from idea to 昀槨nished prod

250、uct.And if you dont*need to*ask them,how many designers,creators,and brands would choose to?CONCLUSIONThrough the lens of generative AI,each layer of a brands value chain could be imbued with new capabilities.Inspiration is no longer con昀槨ned to a select few;commercialization shifts upstream,in昀氁uen

251、cing what is produced;and product creation becomes a dynamic and demand-driven process.A lot needs to be achieved to get there,but I believe the evidence is out there,in both our personal and professional lives,to suggest that this is not a wild idea.I see it as feasible and desirable to recon昀槨gure

252、 the entire value chain around these possibilities,heralding a new era where community,collaboration,and creativity are at the heart of fashion innovation.More importantly,this model could open a new path for aspiring brand owners to tip their toes in the world of entrepreneurship while reducing the

253、ir risk and need to focus on the“boring”parts of having a brand.If this model sparked your curiosity,you might want to consider downloading O昀昁/Script from the app store.My team and I have been working for a long time on turning this vision into a reality,and wed love to hear what The Interlines aud

254、ience thinks.32HOW AND WHERE AI IS TRANSFORMING JOBS,AND WHY OTHERS COULD BE AT RISKBBY MACKENZIE RYAN Y MACKENZIE RYAN INVESTIGATIVE JOURNALISTINVESTIGATIVE JOURNALISTCHANGING ROLESMacKenzie Ryan is an investigative journalist based in Utah.A former investigative producer for the CBS a昀liate in Sal

255、t Lake City,she has also been a contributor to The Guardian since 2022.33At a whole-society level,thought leaders and analysts are still split on whether breakneck improvements in AI,and what seems like its inevitable widespread adoption,are more likely to result in greater productivity,e昀ciency,and

256、 support for existing talent,or in longstanding roles being eroded away by automation.And in both cases,the same commentators are also undecided as to which disciplines will be the most a昀昁ected with the prevailing sentiment being that essentially anyone,from the junior creative to the seasoned CEO,

257、could have their position either augmented or usurped.Where once machine learning-linked job transformations were reserved for clerical and statistical roles,today a massive spectrum of creative and commercial jobs are potentially in AIs sights.What does this mean for fashion?In researching this pie

258、ce,the experts I spoke to con昀槨rmed the expectation that few jobs will be una昀昁ected by AI.In the very near future,manufacturing,marketing,sales forecasting,designing,and even modeling will all necessarily leverage AI as either a creative aid,a productivity boost,or a way to minimize costs and reduc

259、e time to market.And in the slightly longer term,AI does have the potential to subsume some of those roles entirely.In every case,theres also unanimity amongst analysts that fashion industry professionals who become 昀氁uent in these technologies will have the advantage in the job market either as a l

260、ever to safeguard their jobs,or as a way to demonstrate their commitment to embracing new technology.Because,as the CEOs current favourite adage goes,“AI might not take your job,but someone pro昀槨cient in using AI might”.ROUTINE TASKS AND ADMININTENSIVE WORKFLOWS WILL BE FIRST ON THE CHOPPING BLOCK.“

261、AI is most likely to replace roles primarily made up of repetitive tasks,”explained Professor Linwei Xin,from the University of Chicagos Booth School of Business,when we spoke this spring.According to Xin,this is an extrapolation of the existing trend of companies using automation in areas where the

262、y dont have enough humans,or enough interested humans,to 昀槨ll highly routine roles.Until now,that automation has taken a more common and recognizable form:factory robotics,just-in-time supply chains,logistics,warehousing,and distribution.These are what we might characterize as the jobs where technol

263、ogy has been slowly augmenting and replacing human labor for decades.In the United States and China,where relative labor costs have skyrocketed,automation and robotics deployed up and downstream have been key components in reducing companies overall costs.And as Xin pointed out to me,this same drivi

264、ng force is likely to be behind the adoption of AI across the fashion value chain making its probable impacts easier to predict.Some of the fashion industrys manufacturing is already automated,particularly cutting,explained Juan Hinestroza,Cornell University Professor of Fiber Science and Apparel De

265、sign,who I also spoke to in researching this piece although the global distribution of that automation is still uneven.Depending on where garment and textile manufacturing is taking place,it is still currently cheaper for companies to employ humans to make and sew in markets where labor is cheap.But

266、 this 34trend,too,has a clear through-line from the original driver for automation:cost cutting.Fashion has,for several decades,lifted and shifted its sourcing and manufacturing from one region to another as micro and macro-economic forces evolve.In industry parlance,this is whats referred to as“cha

267、sing the cheapest needle,”and it has remained common practice even in the face of coordinated campaigns to“reshore,”or return production to consumption markets like the US,EU,and UK.Hinestroza reminds us that,in the 1970s and 80s,fashion jobs boomed and busted in Eastern Europe,then they moved to Ch

268、ina-and now theyre being relocated to Africa and South America where there is a net bene昀槨t to brands that spans both proximity(shipping is quicker and cheaper from those locations to key markets in Europe and North America)and labor cost.“As a brand you are always racing to minimize costs and conti

269、nue your production,”Hinestroza says.“But what do you do with people who thats the only thing they know how to do?”And this,really,is a critical question.It may be more likely for on-demand,digital production technologies and deployable,composable microfactories to replace o昀昁shore manufacturing,but

270、 the principles and the unanswered queries behind AI adoption are set to be the same.When you have an established system,or a part of one,that operates and sustains itself based on the inputs of people who have been exclusively trained on just that system what happens to them when the commercial jus

271、ti昀槨cation for letting AI do their jobs becomes stronger than the social and moral imperative to keep them in place?This is more than just a theoretical tension,too,since there are entire segments of the fashion value chain where AI already has the capability to replace a swathe of human talent and

272、we are already seeing a pushback from communities(and the general public)that suggests that a delicate balancing act is currently taking place.WILL AI REPLACE REAL LIVE MODELS?The process of making 昀槨nal physical production samples in one part of the world,shipping them to another,identifying the ri

273、ght human beings to wear them,then booking studio space or locations for a professional photographer to shoot them in and then translating the resulting images into marketing and eCommerce materials is a long,35expensive and circuitous one.And this makes it a prime target for AI automation,since a g

274、enerative model could potentially condense many of these steps into one,and many of these people and skillsets into a single piece of software.But is it quite that simple?The commercial argument may be compelling,but the cultural counterargument is equally strong.Dutch AI studio Lalaland has become

275、a prominent name in fashion thanks to its pipeline,which can generate lifelike fashion photoshoots featuring arti昀槨cial models by training an algorithm on actual photos and on a brands speci昀槨c products,as their CEO and Founder Michael Musandu explained to me.At a 昀槨rst glance,their work raises clea

276、r questions about whether fashion brands,which spend thousands to millions of dollars a year to hire models alone without even factoring in the costs of all the other photoshoot elements I mentioned earlier-could cut back on the money they spend on photoshoots,and the models,photographers and other

277、professional skills and materials needed for them,and reinvest that income elsewhere.Instead,Musandu believes their AI models are helping to create equity for brands of all sizes and di昀昁erent budgets,allowing them to scale their photoshoots without those overheads becoming constraints.“By saving 70

278、%of the costs associated with traditional shoots,weve seen brands reinvest these savings into more marketing campaign videos and images featuring real models,”he notes.“Lalalands AI-model supplement does not replace real models,”Musandu pointed out to me.“We believe human models will continue to pla

279、y a vital role in the fashion industry,establishing genuine connections with consumers.Our technology aims to support this.”Musandu argues that the industry will not only need to continue employing models and photographers and lighting engineers and booking studio space,but it will actually need mor

280、e of that talent to come from more diverse and underrepresented groups,and that AI-enabled automation can be a lever to help that happen.The complicated status of AI and diversity and inclusivity is analysed in more depth in another feature contained in this report Editor And this raises an importan

281、t consideration for any organization looking at AI as a way to augment or automate any creative,technical,or commercial role:does that job create a pinch point by simply existing,or could the time and e昀昁ort the person 昀槨lling that role expends create greater value elsewhere?Or,in other words,is the

282、 opportunity a matter of throughput and speed,or of releasing untapped potential?The answer to this question is likely to fall somewhere in the middle,but the template being set by AI-generated product photography does suggest that fashion is going to see an evolution of traditional roles rather tha

283、n a wholesale replacement of them.Although the net result may still be fewer jobs in some areas,including those that have traditionally been considered sacred in fashion.36THE INDUSTRY WILL TREND TOWARD HIRING FEWER,MORE AIFLUENT DESIGNERS.Despite the public perception,technology-enabled or technolo

284、gy-adjacent automation is rarely the proximate cause of a shift in the jobs market.Instead,analysts and thought leaders suggest,the real replacement for an entrenched role is talent with a di昀昁erent,more contemporary skillset.“Its not that the worker gets replaced by a robot or a machine in most cas

285、es,especially for desk jobs,its that some better or more educated worker can do that job because they can be twice as productive or three times as productive,”Code.org Founder and CEO Hadi Partovi said during this years World Economic Forum.“The imperative is to teach how these tools work to every c

286、itizen,and especially to our young people.”And Hinestroza reports that hes been using AI for almost two years himself,as well as teaching generative and non-generative tools in his classes because his students“need to be prepared for the reality of the industry”.Designers in particular need to be 昀氁

287、uent,Hinestroza told me,because the largest improvements in the underlying technology have been observed in generating visuals.And Hinestroza points out,that at the beginning of last fall semester,the output of generative image models went from lackluster quality to experiencing an“incredible leap o

288、f images and technology”.When he attended the biannual Canton Fair,one of the worlds largest trade shows,he noticed nearly every company designing was using text-to-image generators like Midjourney and Dall-E as a way to cope with the demands of creativity in a fast-paced industry.“If you look at a

289、butter昀氁y,and want a collection based on butter昀氁ies in Mexico,AI models will create an entire collection for you,”he says.Does this mean the end of non-AI fashion design as an occupation?Hinestroza reminded me that we simply dont know yet.One potential outcome is that designers and AI become co-pil

290、ots,with generative models providing creative inspiration and assistance and even potentially helping with technical design and patternmaking tasks.On the other hand,companies are signaling that they wont need as many designers,with the hope being that the use of AI tools could help to open up other

291、 roles for those professionals even if the prospects are potentially bleak for junior designers who may no longer get the opportunity to prove themselves before being elevated to mid-weight and senior roles.In an use case that may feel dystopian or unremarkable,depending on your perspective,Hinestro

292、za described a hypothetical example to 37me whereby a brand might fully automate a spring collection based on fruits in Asia,for instance,and then a designer could create a separate portfolio leveraging di昀昁erent tools-and another executive could choose which work昀氁ow and which set of tools did the

293、better job.And while this may not be completely a case of pitting people against machines,it would,in the very least,be an instance of machine-assisted humans being set against machine-native creative pipelines.Crucially,the timeline for this kind of experimentation is short,as Hinestroza told me:“M

294、ost of the companies are using these tools.Some admit it.Some dont.I dont know why they dont admit it.Maybe they want to pretend to be more human-centered,but at the end of the day,you have deadlines to meet and AI can accelerate the process.”AI OPTIMIZES.HUMANS MAKE DECISIONS.Earlier this year,Mass

295、achusetts Institute of Technology labor economist David Autor arguedin Noema magazine that the fear of AI replacing jobs is misplaced and,instead,the adoption of AI could instead serve as a catalysts for what we might term capability elevation-enabling a larger set of workers to perform higher-stake

296、s decision-making.And the experts I interviewed were all clear that AIs impact will not just be felt in narrow roles,but that AI should,at least in theory,enable mid-level to executive-level fashion industry professionals to make betterand fewerdecisions.“Imagine selling 100,000 products.You have to

297、 manage each of them,every week deciding how many of this shirt to order from this supplier.If youre selling thousands and thousands of products,it would be di昀cult and tedious for human buyers,”Xin explained to me.Which is equivalent to asking how can one person make all of the decisions that need

298、to be taken across the extended product lifecycle,and in the compressed timeframe needed to bring new styles to market when they still have a good chance of selling.Xin described to me an example of how AI used past sales data to do a better job of forecasting so that“your human buyer can focus more

299、 on best sellers”and continue the job of communicating with suppliers.A prime instance of something considered an intuitive or higher order task that is,in reality,another matter of e昀ciency,time,and accuracy underlining the extent to which technology can progressively shift the window of what is co

300、nsidered a task thats suitable for automation.38Instead of having people 昀氁y around the world and identify trends by visiting runway shows or conducting 昀槨eld trips to fashion capitals,now algorithms work more e昀昁ectively and e昀ciently.They process social media photos of what people are wearing in S

301、ingapore,London,or Rio and,in real-time,decision-maker are able to“ask the machine”to create a collection based on the next trend.Large brands use mass customization instead of traditional forecasting to reduce the lead time to produce speci昀槨c items,Xin explains.In this process,AI predicts trends a

302、nd optimizes their scheduling from design to delivery.When customers choose preferences for the product theyre purchasing,the brand e昀昁ectively delays customization until the last step in choices such as color and material.This concept is known as postponing.Casting the net wider,Xin could not comme

303、nt on whether or not the fashion sector would cut marketing jobs.Hinestroza,though,told me that he believes most marketing now can be 100 percent automated with AI tools.His students are already learning how to use AI to market clothes to di昀昁erent ages and di昀昁erent markets,and earlier comments fro

304、m Musandu at Lalaland reinforce that the choice between tradition and automation is not always a binary one suggesting that this next generation of talent will be the one to navigate the 昀槨ne line between an evolving technology frontier and cultural attitudes.AI can also optimize shipping routes and

305、 manufacturing,Xin says,even predicting disruptions rather than leaving brands needing to react to them.The push to make supply chains more robust has ramped up after 39COVID and the Russia-Ukraine war.For the 20 years prior,companies wanted to minimize operations,leaving them vulnerable to disrupti

306、ons and limitations in supply chains,he explains.Even with AI taking on a powerful role in planning and monitoring supply chains,though,Xin argues that there would still need to be high-level executives managing the human connections and relationships that keep these supply chain functioning.Xin did

307、 express concern about how automation is replacing some formerly human-powered jobs.However,he repeatedly noted that there will be many jobs AI is less likely to replace,particularly those that involve human relationships and their psychology.Xin used chess moves as an illustration:AI chess engines

308、can tell a player what the best move is,but only a human coach can tell a player why its the best move.And from a disclosure and trust-building perspective,that transparency is likely to matter a great deal.BRANDSAND STOCKHOLDERSBENEFIT.TECH COMPANIES ARE NEXT.With such an unclear picture of how and

309、 where AI is going to supplement or supplant human labour,the question remains:who is 昀槨nancially bene昀槨tting from the proliferation of AI tools in fashion?Hinestroza says tech companies arent making huge amounts of money from it yetbut they will soon enough.This is likely down to the high cost of t

310、raining and running inference for large AI models,but there is already evidence to suggest that these costs are being driven down,and that the sliding scale of customer revenue to cost will soon tip over into pro昀槨tability.Brands,though,are already successfully minimizing their cost,Hinestroza belie

311、ves.In a new work昀氁ow,AI generates designs quickly,the brands choose the best designs for human creatives to either iterate on or approve,and then the manufacturers create prototypes very fast.This also,in e昀昁ect,minimizes production because it reduces the overall amount of prototypes developed and

312、increases the adoption rate of design ideas and samples to 昀槨nished products.And Hinestroza also reminded me that,in his experience,most of the designers for big brands are already contractors.In the future,brands may only need to hire a reduced number of full-time people to take care of AI programs

313、.And while Western brands may hesitate to adopt similar work昀氁ows in their own operations,Hinestroza told me that when he was in China,he saw substantial third-party design companies leveraging AI design tools to create their new product catalogues,and those designs then being sourced by brands and

314、supplying clothes all over the world,knowingly or otherwise.So while obvious AI use by domestic brands is the tip of the iceberg,the reality is that AI has likely already deeply in昀槨ltrated the fashion value chain and is creating a commercial edge for suppliers one that brands themselves will want t

315、o replicate.Quite how this develops culturally,is currently very di昀cult to say.The impact on fashions talent base is clearly imminent,but how that impact will be perceived by the shopping public is hard to predict given the backlash against the use of AI in creative 昀槨elds.“If youre in the business

316、 of fashion,you have to sell clothing to people,not computers.Its a complicated phenomenon,”Hinestroza concludes.40AASIA DVAZ STERLING CENTRAL SAINT MARTINSAasia has a Masters in Fashion Entrepreneurship and Innovation,a degree in Fashion Design Technology,and has previously worked as an Associate L

317、ecturer and Enterprise Consultant-all at London College of Fashion.She now works as a Partnerships Manager at Central Saint Martins and is also an alumna of the University of Cambridge Innovators For Sustainable Fashion accelerator.Copyright Stefan JakubowskiCAN AI BREAK DOWN FASHIONS BARRIERS,OR WI

318、LL IT ENTRENCH EXISTING BIASES?FASHION HAS BEEN TARNISHED BY INEQUALITY.CAN GENERATIVE AI HELP DELIVER A MORE EQUITABLE FUTURE?Equity,diversity,and inclusion although not a specialist in the 昀槨eld,these are words I am faced with every day,and I am sure you are too.At work,in the news,and on social m

319、edia,awareness is everywhere and embedded into our collective consciousness.So why,despite some notable e昀昁ort from certain brands,does a truly equitable,diverse,and inclusive fashion industry feel so far away?Maybe its because these three weighty words come in many forms and guises,not just through

320、 the endless ways in which we organise,structure,and refer to them EDI,DEI,D&I,etc.but more importantly,how we view,prioritise,and interact with them as concepts.Its a complex discourse that can sway between lip service and lively debate,distraction and genuine change as we navigate and establish th

321、eir meaning and importance.For example,should we strive for equality or equity?How do we measure diversity of thought?Can inclusivity truly be all-encompassing?Yet to even begin unpicking these questions requires context,patience,and emotional e昀cacy virtues the fashion industry is not famously know

322、n for.So often within the industry,we speak about EDI as objectives to be achieved under the umbrella of CSR metrics and marketing strategies,but for all of us,EDI needs to exist as concepts that organically live and breathe in both the broader and smaller everyday aspects of our lives and society.A

323、s a result,for a long time,genuine inclusivity within the fashion industry has failed to feel like a feasible reality.It has just been too far away.Yes,it can be said that the industry is making strides forward,but sometimes progress feels akin to treading water.This is particularly acute when we se

324、e examples of individuals at the top of the fashion food chain facing discrimination.Last year Tremaine Emory resigned as creative director of Supreme,alleging that there was systemic racism at play at the brand.And just take a look at representation across the four major fashion weeks,there has bee

325、n a decrease in total size inclusivity,making body positivity feel like a 昀氁eeting trend.Only two years ago,the British Fashion Council reported that just half of fashion businesses had coordinated D&I strategies.So as generative AI speeds ahead in its development and implementation within the fashi

326、on industry,I wonder if arti昀槨cial intelligence can and will help us to do better-or whether it might have the opposite impact,and lead to deeper entrenchment of the current makeup of the industry.Like fashion itself,generative AI is a pandoras box of ethical concerns,philosophical dilemmas,and felt

327、 risks.And it isnt surprising that models trained on our reality,a huge corpus of the worlds publicly available data,re昀氁ects the aggregate biases that are present in that reality.Day to day,our world is hampered by bias,stereotypes,and discrimination,which begs many of us to question if we are cult

328、ivating the right intelligence within our technologies-or if were even capable of doing so.But whether were ready to answer this or not,generative AI is here,and its application in fashion is evolving at lightning speed.Which should bring deep cultural questions that have long gone under-recognised

329、昀槨rmly back into the spotlight.EMBEDDING ETHICS INTO GENERATIVE AI AND ITS FASHION APPLICATIONS Similarly to fashion itself,a lack of diversity in AI has long been a concern for EDI thought leaders,with notable departures from the EDI departments of technology giants illustrating a worrying trend wh

330、ereby the people developing and shaping the AI systems seem to downplay the concerns of people who are worried that those systems are inheriting ingrained biases.So,as these two worlds collide to form innovative solutions,its clear to see why these worries surround their union.But what does EDI mean

331、 in the context of generative AI and fashion?Franki Tabor who works in AI as both an ethics and fashion consultant explains:In the labyrinth of AI,ethics is the compass guiding us towards responsible innovation.Tabor has established 7 pillars on which ethics in AI are upheld.Her thoughts on fairness

332、 and accountability,in particular,speak to the challenges of approaching EDI within a fashion industry thats rushing towards AI adoption.Fairness in AI is not just an option;its the foundation of equitable technology for all.Accountability in AI(is)where innovation meets responsibility,ensuring tech

333、nology serves humanity.These statements begin to answer my question.Simply put,fashions use of AI must be equitable for,and conscientious of,everyone in order to serve everyone,but this is easier said than done.Not just because of fashions own di昀cult history in this regard,but because generative AI,in a wide,industry-agnostic sense,has already proven to be problematic.PULLING BACK THE CURTAIN ON