《前田·約翰：2023科技中的設計報告（英文版）（42頁）.pdf》由會員分享，可在線閱讀，更多相關《前田·約翰：2023科技中的設計報告（英文版）（42頁）.pdf（42頁珍藏版）》請在三個皮匠報告上搜索。

1、2Howdy!Back in 2015,I started sharing everything Id recently learned into a kind of“music album”of knowledge to drop at SXSW.Its a side project that hasnt left my side,thus farI guess it saves me time so I dont have to write another book:+).The typeface used in the 2023 Design in Tech Report is Inte

2、r by Rasmus AnderssonWhat is the 2023 Design in Tech Report?DESIGNINTECH.REPORTjohnmaedaDESIGNINTECH.REPORTAESTHETICS31.Design People 2.DE$IGN And Digital Transformation 3.How To Speak Machine How To Speak Human 4.The AI Ketchup Bottle Moment 5.Resilience Or Renewal 6.AESTHETICS of ETHICSUseful AI S

3、tuff For Right Now0.TABLE OF CONTENTSSorry I couldnt fit the 150 other slides I had planned to go through because I realize the backstory may not matter as much to everyone than the parts about the now.I.Machines Run LOOPS 1.Computers excel at repeating themselves through loops.2.Hard machines are v

4、isible;soft machines are invisible.3.Human computers are the original computing machines.4.Recursion is the most elegant means to repeat oneself.5.Loops are indestructible unless a programmer has made an error.II.Machines Get LARGE 1.Embracing exponential thinking is unnatural at first.2.Loops wrapp

5、ed inside loops open new dimensions.3.Be open to both directions of the powers of ten.4.Losing touch with human scale can make you toxic.5.Computers team up with each other way better than we do.III.Machines Are LIVING 1.Discerning whats alive versus whats not used to be simpler.2.The science of lif

6、elikeness is undergoing a renaissance.3.An artists perspective keeps you humanly curious.4.Life is defined by how we live in relation to each other.5.Computers wont replace us if we remain audacious.How To Speak Machine Computational Thinking For The Rest Of Us2019Source:Raymond Kurzweil https:/ Kur

7、zweils predictions from 2011 seem to trending towards“maybe he was right”DESIGNINTECH.REPORTAESTHETICSHow To Speak Machine Computational Thinking For The Rest Of Us 2019Source:johnmaeda.How To Speak MachineA biologist lives next to a pond.She clears the pond of lily pads.And runs and experiment with

8、 a species of lily pads that doubles over night Simple Moores Law explanation of why ChatGPT feels like it appeared“all of a sudden.”DESIGNINTECH.REPORTAESTHETICSHow To Speak Machine Computational Thinking For The Rest Of Us On Day 3014 15 16 17 18 19 20 21 22 23 24 25 26 27 28 292019Source:johnmaed

9、a.How To Speak MachineOn day 1,theres 1 lily pad.On day 2,theres 4.And so forth.On day 30,the pond is covered with lily pads.Simple Moores Law explanation of why ChatGPT feels like it appeared“all of a sudden.”DESIGNINTECH.REPORTHow To Speak Machine Computational Thinking For The Rest Of Us On Day 3

10、014 15 16 17 18 19 20 21 22 23 24 25 26 27 28 292019Source:johnmaeda.How To Speak MachineOn what day is the pond half full of lily pads?Day 15?Simple Moores Law explanation of why ChatGPT feels like it appeared“all of a sudden.”DESIGNINTECH.REPORTHow To Speak Machine Computational Thinking For The R

11、est Of Us On Day 3014 15 16 17 18 19 20 21 22 23 24 25 26 27 28 292019Source:johnmaeda.How To Speak MachineOn what day is the pond half full of lily pads?Or Day 30?Simple Moores Law explanation of why ChatGPT feels like it appeared“all of a sudden.”DESIGNINTECH.REPORTRESILIENCETECH.REPORT9Design 501

12、:The 4th I.R.introduced a new kind of design.I wrote a book about it in 2019.The trope of,“Speed,price,quality:you only get to pick two”gives way to,“you get all three with computational design!”That means undesirable work can get automated so that you can work on the parts that you want.Finally!Sou

13、rce:johnmaeda TIMEValue creationClassical DesignPHYSICAL WORLDDesign ThinkingORGANIZATIONAL WORLDA GENTLE INTRODUCTION TO A.I./MACHINE LEARNINGComputational DesignDIGITAL WORLDIndustrial Revolution201520162017201820192020202120222023DESIGNINTECH.REPORTAESTHETICS10Computational Design:Artificial Inte

14、lligenceAI isnt good at inclusive design because we arent,too88%of designers surveyed believe that it will be atleast 5 years or more until visual designers arereplaced by AI.AI can already do a lot right now.The history of AI and generating visual art goes backto the 1960s with A.Michael Noll and o

15、ther artists atBell Labs,and stretches back to Marcel Duchamp.AI is extremely proficient at tedious tasks that nohuman should really have to do,like:adjust imagecontrast,correct messy lines,and re-style images.Google is by far the leader in mixing AI with designexperimentation due to the amazing tal

16、ent theyveacquired like Martin Wattenberg and FernandaViegas who at IBM first advanced datavisualization with their landmark Many Eyes.AI is showing us the unintended consequences ofrunning what appear to be“fair”algorithms that feedoff of past activity and practices that are convertedinto training

17、data.But embedded in that trainingdata is our long history of exclusion.2018 Design In Tech Report|Computational Design:AI58/91201520162017201820192020202120222023DESIGNINTECH.REPORTAESTHETICS11Towards conversational designConversation is not a new interface.Its the oldest interface.Conversation is

18、how humans interact with one another,and havefor millennia.We should be able to use the same principles tomake our digital systems easy and intuitive to use by finallygetting the machines to play by our rules.Erika HallSource:mulegirl Conversational Interfaces2018 Design In Tech Report|Computational

19、 Design54/91201520162017201820192020202120222023DESIGNINTECH.REPORTAESTHETICS12Nicholas Negroponte and team sensed the future,and made it.Imagine a machine that can followyour design methodology and at thesame time discern and assimilate yourconversational idiosyncrasy.The samemachine,after observin

20、g yourbehavior,could build a predictivemodel of your conversationalperformance.Such a machine couldthen reinforce the dialogue by using apredictive model to respond to you ina manner that is in rhythm with yourpersonal behavior and conversationalidiosyncrasies.”NicholasNegroponte,The ArchitectureMac

21、hine(1967)“Source:2019 Design In Tech Report|Yaaaaaaaaawn!Boring AI79/98201520162017201820192020202120222023/nollcurve /stepamp exch def/numsteps exch def/cycles exch def /xscale exch def/amp exch def 0 amp moveto/curx 0 def 1 1 cycles /i exch def/radbase 180 i 1 sub mul 90 add def /numincrs numstep

22、s stepamp i mul mul def /dx 180 numincrs div def/x radbase def 1 1 numincrs pop curx amp x sin mul lineto/x x dx add def /curx curx xscale add def for for stroke def /noll90 .7 setlinewidth 0 setgray 1 1 90 gsave 3.4 mul 20 add 0 exch translate 15 1 9 8 1.25 nollcurve grestore for defnoll90A RT A N

23、D C O M P U TAT I O N /1 9 9 0technologydesign1990ICYMIDESIGNINTECH.REPORTAESTHETICSDESIGNINTECH.REPORTAESTHETICS14Useful AI Stuff For Right NowSource:johnmaeda Things that are top of mind for me right now.Everything else in this preso is looking backwards,to help me look forwards.If you want to do

24、that too,then stick around!0DESIGNINTECH.REPORTAESTHETICSEvery year I run a survey related to the Design in Tech Report.But this year Ive been a little busy with all-things GPT-universe.That said,I got a lot of great contributions!As Figma AI plug-ins go, has some really cool ones.My favorite obscur

25、e find was the Pokemon generator.For ready-to-eat prompts go to promptbase or lexica.art which are prompt marketplaces.Source:johnmaeda Annual Design in Tech Survey Respondees|futurepedia.io has amore of these kinds of finds for you to forage through 15Thank you to Henrik,Frederik,Sean-Wood,Kai,Kim,

26、Samuel,Henry,Ryan,Chris,Colin,Agnieszka,Bob,Gary,Agu,Dave,Moiss,Me,Ankita,Darius,Rachel,Aldo,PChia,Heather,Justine,Miguel,Florentin,Juliana,Amanda,Sam,Gaby,Kerry,Chloe,Duncan,Jan,Harshika,Arum,Gael,Raffaela,Ghizlene,Marcel,Narasimha,Eelko,Jason,Debk,Andrea,Ashish,Andr,Bill,Maria,Art,Puneet,Vidhi,Mat

27、t,Lauren,Brian,Sridhar,Gaurang,Julie,Ocean,Axel,Sithembiso,Rodolfo,Sanmitra,Sbastien,Bruce,Jason,Maria,Carlo,Bs,Angelina,Silvia,Filippo,Manisha,Roland-Philippe,Andrs,Ame,Lucas,Marin,Mark,Cristin,Quang,Juhan,Pat,Luiza,Zyo,Chrys,Mea,Rupert,Ellie,Richard,Paolo,Sean,Guy,Kaladhar,Sebastian,Alessio,Surya,

28、Richard,Andre,Dani,Vivek,Christina,Jesper,Ranjeet,Kshipra,Betsongeorge,Jodean,Marc,Paul,Akshay,Roman,Dylan,Naomi,George,Paul for contributing to the 2023 survey!How likely will the design profession need to change given the emergence of Large Language Model AI systems like ChatGPT?Average=8 10=“high

29、ly likely”What prominent Design AI tool out there is top of mind for you?Adobe Tools Figma+AI Plugins DALL-E Midjourney Runway D-ID Copy.ai Elevenlabs Stable Diffusion Hugging Face*Anything*ChatGPT/ChatGDP/Chat GBTWhat*obscure*Design AI tool out there is top of mind for you?Pinecone|Weaviate|Qdrant:

30、Vector DB Uizard:Instant UI NVIDIA Gaugan:Universe Painter Vizcom:3D From Prompts Autodraw:Early Google Experiment Wombo:Lip Sync Synesthesia:Avatar Spokesperson Luma AI:Instant 3D Deepl:Language translator Tome:Instant Presos Lyric Studio:Instant Music Playground:IRL Observability Photofeeler:Judge

31、 Your Photo Edward Teaching BotSomeone I know has been directly affected by the recent spate of layoffs in tech.65%Good luck to everyone out there who will find their next step in life.I believe in you!3 JohnCONTEXTNSWENSWETO DA I J I T E M P L E|N A R A,J A PA N 2 0 0 11992DESIGNINTECH.REPORTAESTHE

32、TICSD E S I G N“”I S D E F I N E D B YM AT E R I A L S S E L E C T I O Nmaterials1992DESIGNINTECH.REPORTAESTHETICSDESIGNINTECH.REPORTAESTHETICS“A Comprehensive Survey on Pretrained Foundation Models:A History from BERT to ChatGPT”Learn about many of the models out there at https:/huggingface.co/docs

33、/transformers/model_doc/flavaSource:https:/arxiv.org/pdf/2302.09419.pdf18CV Context Exemplar-CNNCV BiGAN NAT CountingCV BigBiGAN DeepCluster RotNetCV MoCo SimCLR BEiT MAEGraph DeepWalk LINEGraph VGAE node2vec GraphSageGraph DGIGraph GCC GraphCL SUGAR MoCLHISTORY OF PRETRAINED FOUNDATION MODELS201420

34、16201820202022NLP Skip-GramNLP GloVe LM-LSTMNLP CoVE Shared LSTM FastTextNLP RoBERTa GPT-2 BERTNLP GPT-3 BART PET SimCSE ChatGPT=Large Language ModelUnified PFMs UNIFIED-IO OFA FLAVA Gato BEiT-3=Unified PFMUnified IO https:/ OFA(Once For All)https:/ FLAVA(A Foundational Language And Vision Alignment

35、 Model)https:/ Gato https:/ BEiT(Bidirectional Encoder representation from Image Transformers)https:/ Ball)https:/en.wikipedia.org/wiki/BB-8 Just checking if youre still awake DESIGNINTECH.REPORTAESTHETICSThe end of the year seemed to go really fast for me.Same for you,too?Source:Sonya Huang https:/

36、 17OCT 24SEP 19DESIGNINTECH.REPORTAESTHETICSIt helps to squint a little to understand a few things.Because theres a lot to take in rn.Source:Sequoia https:/ OF MODELSpre-20202030?What you build will change as there are more models available to use.KINDS OF MODELSThe app you build depends upon the mo

37、del(s)you have available.Green means“look to be surprised”MODEL LAYERAPP LAYERDESIGNINTECH.REPORTAESTHETICSSimons Scissors:Context and CognitionScissors Photo by Matt Artz on UnsplashSource:Dan Lockton,“Simons Scissors and Ecological Psychology in Design for Behavior Change”in SSRN Electronic Journa

38、l,Aug 2012.21CognitionHuman rational behavior is shaped by a scissors whose blades are the structure of task environments and the computational capabilities of the actor.”Herbert Simon,“Invariants of Human Behavior”in Annual Review of Psychology,41,1990,p.1-19“THE PRETRAINED FOUNDATIONAL MODELTHE PR

39、OMPT YOU FEED THE MODELDESIGNINTECH.REPORTAESTHETICSCognitionTo build apps with models takes a little time to settle into your mind.Its not as easy as saying to yourself as a dev,“Hey.This is similar to XYZ.”Because its not.New models are being built with plenty of determination across the world.The

40、 blade that matters to the non-model builders is where a lot is happening.OSS langchain and llamaindex are great for design-devs to try out!Source:Unusual VC https:/www.unusual.vc/post/devtools-for-language-modelsLLM APIPrompt DesignVector DB+Other DataYour ProductEASYEASYEASY “DevTools for language

41、 models predicting the future”unusualvcMachine Learning Data WarehouseHire ML ExpertBuild ModelWrap EndpointHARDYour ProductDESIGNINTECH.REPORTAESTHETICSPrompt Engineering 101Source:https:/ the core promptMake your wish.Be specific about the outcome,length,format,and style.Use precision in how you p

42、rompt.When referring to text to act upon,delimit it with“”(3 double quote marks)or#(3 hash sign marks).3.Add in useful contextGive one example of how youd like your prompt to behave.Give multiple examples of how you want your prompt to behave.When you dont give guidance with examples,thats“zero shot

43、.”The other cases are called“one shot”and“few shot.”4.Tune the models settingsSet the temperature high to make it more random;lower the temperature to make it more deterministic.Alternatively,adjust the“p”parameter to broaden or narrow the range of words that will be generated in response.Give it mo

44、re tokens when you need to work with longer prompt or longer responses.OLDER($)NEWER($)1.Select a large language model*Davinci Curie Babbage Ada The OpenAI universe also includes Codex(for generating computer code)and the new ChatGPT model DESIGNINTECH.REPORTAESTHETICSPrompt Engineering 201Playing w

45、ith the OAI tokenizer https:/ gives you valuable insights on how it“eats”text.And learn more about how“chain of thought”prompting works via https:/arxiv.org/abs/2201.11903 asap.Source:https:/arxiv.org/abs/2201.1190324C.Use“chain of thought”promptingGetting a model to use our own human tendency to“th

46、ink out loud”has been shown to produce more accurate outcomes.When prompting,give an explicit“thinking out loud”explanation for how a problem is to be solved with your related reasoning.A.Shorten an often-used prompt You can reduce*the amount of tokens that get used by finding a similar but shorter

47、version of a prompt.This way if youre using the prompt a lot,youre not using up as many tokens in a cumulative sense.*One DiT survey responder suggested using Quillbot.B.Go for a two-fer or three-ferSometimes with one prompt you can get multiple things done.DESIGNINTECH.REPORTAESTHETICSWhen I mentio

48、ned the Herbert Simon scissors metaphor to Sam Schillace,I recall him quickly saying,“And what did our parents teach us?Dont run with scissors.”Scissors Photo by Matt Artz on UnsplashSource:Sam Schillace https:/ HAI“On the Opportunities and Risks of Foundation Model”(2022)Its the most comprehensive

49、take on the pros and cons of pretrained foundation models as a 160 page“weapon of understanding”with 30 pages of references if youre new to this space,its a godsend.Source:https:/arxiv.org/pdf/2108.07258.pdf26MACHINE LEARNINGDEEP LEARNINGFOUNDATION MODELSTIMEEmergence ofHomogenization of“how”feature

50、sfunctionalitieslearning algorithmsarchitecturesmodels1.While new capabilities emerge,our reliance on a more homogenized approach creates a single point of failure.2.Ultimately,all of this modeling eventually gets deployed to real people.And the implications at the scale of millions are huge.Data Cr

51、eationData CurationTrainingAdaptationDeploymentLarge uncharted data sets are a source of robustness and are an increased risk of poisoningFoundational models are a security choke point and also an increased attack surfaceDownstream apps easier to develop but“dual use”(use for unintended purposes)ris

52、k S E C U R I T YDESIGNINTECH.REPORTAESTHETICSFor my introductory AI course,I had Professor Joseph Weizenbaum as the instructor.His significance was lost on me until many years later I realized that he was a GOAT.Theres a fab article on Dr.Weizenbaum on 99%Invisible https:/99percentinvisible.org/epi

53、sode/the-eliza-effect/Source:Computer Power and Human Reason(1976)27I knew of course that people form all sorts of emotional bonds to machines,for example,to musical instruments,motorcycles,and cars.And I knew from long experience that the strong emotional ties many programmers have to their compute

54、rs are often formed after only short exposures to their machines.What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.This insight led me to attach new importance to questions of the relatio

55、nship between the individual and the computer,and hence to resolve to think about them.”Joseph Weizenbaum(1976)“DESIGNINTECH.REPORTAESTHETICSJudgment Call kicked off a nuanced approach to talking about AI in product companies.And Professor Helen Armstrongs 2021 book on AI for designers is certainly

56、timely.Judgment Call(2019)was authored by Stephanie Ballard,Karen M.Chappell,Kristin Kennedy to use value sensitive design and design fiction to surface ethical concerns related to technology.Source:Mira Lane/Helen Armstrong/Sarah Gold28 More card toolkit recommendations courtesy of data and trust g

57、uru CEO/Founder Sarah Gold of (UK)DESIGNINTECH.REPORTAESTHETICSIts easy to get GPT-bots to“pretend youre a”for everything ranging from a JS console to a math teacher.Critical thinking is key right now;David Karam has“Culture Club”for you.Other approaches recommended by DiT survey responders are for

58、simulate audience reaction, webcam reaction aggregation,and synthetic health which is an open source patient generator.Source:https:/ Karam https:/ give us the opportunity to explore concepts from multiple perspectives in simulation.”David Karam“DESIGNINTECH.REPORTAESTHETICSMorris was not entirely a

59、gainst the use of machines,but felt that the division of labour a system designed to increase efficiency,in which the manufacture of an object was broken into small,separate tasks,meaning individuals had a very weak relationship with the results of their labour was a move in the wrong direction.”V&A

60、“Ruskin&Morris famously led the Arts&Crafts Movement in the late 1800s as a reaction to the Industrial Revolutions introduction of new fabrication technologies.A useful take on the plight of the Luddites in relationship to whats happening now with AI fears is https:/ Morris,made by Ada Phoebe Godman

61、(1877)30DESIGNINTECH.REPORTAESTHETICSJust like in the Arts&Crafts Movement,we should expect to see*critical making*happen.Like this brand new animation just released by design technologist Masashi KawamuraSource:masakawa https:/ Will Foundation Models Impact Design Thinking?Source:https:/web.stanfor

62、d.edu/class/me113/d_thinking.html32Summarize customer support or field logs1Categorize common problems2Generate possible solutions3Visualize solution journeys and examples4Simulate customer reaction5EmpathizeDefineIdeatePrototypeTestThis is one hypothetical way:DESIGNINTECH.REPORTAESTHETICSThe Creat

63、ive ProcessExplore Experiment PracticeCreateReflect Assess ReviseShareImagine Examine PerceiveThis is one hypothetical way:How Will Foundation Models Impact Classical Design?Source:https:/www.kennedy-center.org/education/resources-for-educators/classroom-resources/articles-and-how-tos/articles/colle

64、ctions/arts-integration-resources/arts-integration-and-universal-design-for-learning/33Test options with AIUse AI audienceChat with AI criticGenerate options with AISketch with AIDESIGNINTECH.REPORTAESTHETICSIn this new age of partnering with PFMs,we need to keep in mind what we dont enjoy doing ver

65、sus what we really love to do.Remove the drudgery while increasing your joy.I run a poll regularly in large groups of what people prefer not to do at work versus what they really love doing at work.It turns out theres tons of things that dont spark joy in the workplace.Ya think?Source:johnmaeda34Cal

66、endar Management Making Powerpoint Decks Cleaning Data Low-priority E-mails Competitive Research Summarizing Monitoring/Alerts “DOES NOT SPARK JOY”Decision Making Talking With People Learning With Others Mentoring/Coaching Relationship Building Delivering Presentations Crafting Strategy “DOES SPARK

67、JOY”Simplifying the Laws of Simplicity.Quiz:Which cookie will a child choose?2005DESIGNINTECH.REPORTAESTHETICSQuiz:Which cookie will a child choose?2005DESIGNINTECH.REPORTAESTHETICSWhich pile of laundry will a child choose?2005DESIGNINTECH.REPORTAESTHETICSwant more(enjoy/consume)want less(work/trans

68、form)Simplicity is about living life with more enjoyment and less pain.2005LAUNDRYCOOKIEDESIGNINTECH.REPORTAESTHETICS2023Think of how to use AI to get rid of things you dont really want to do.So that you have more time to do more of what you love to do.LAUNDRYCOOKIE“DOES NOT SPARK JOY”“DOES SPARK JO

69、Y”DESIGNINTECH.REPORTAESTHETICSDESIGNINTECH.REPORTAESTHETICSSource:johnmaeda IMDB40 Ive used two sci-fi movie metaphors when struggling to understand this stuffUseful AI Stuff For Right Now1.Arrival(2016)Directed by Denis Villeneuve Story by Ted Chiang Its about communicating with an alien race with

70、in a set of strict limitations in how long you can talk with the alien.The mystery is grounded in discovering the language of the alien through trial and error.Spacetime Language=Prompts2.Black Panther(2018)Directed by Ryan Coogler Story by Ryan Coogler+Joe Cole Its about the king of an African nati

71、on called Wakanda which is the most technologically advanced country on earth.Their early access to an alien material with special properties that defy physics shaped their history.Vibranium=Cognition BladeDESIGNINTECH.REPORTAESTHETICSSource:johnmaeda41 Designers tend to care about people as people.

72、Instead of just as prospects or customers.Design as a field has evolved to speak more machine,more scale,and more$s.A lot of work has been done in the design world to birth a new“aesthetics of ethics.”Speaking human really well will matter more than speaking machine in this next chapter.But its not mainstage because it doesnt sound like“time or$s saved”or“new$s.”Next years report will center around the business value of this new kind of design.Good luck to us all!SummarySource:Tony Ruth lunchbreath in the 2020 CX ReportA MESSAGE FROM OUR SPONSOR THE EARTH DESIGNINTECH.REPORTAESTHETICS