Dieser Beitrag ist im Rahmen meines Reisestipendiums für die DHd 2023 „Open Humanities Open Culture“ vom 13. bis 17. März 2023 entstanden. Herzlichen Dank an die Gerda-Henkel-Stiftung, die mir mit dem Stipendium den Besuch der Konferenz ermöglicht hat.
- "Human beings have dreams. Even dogs have dreams, but not you. You are just a machine. An imitation of life. Can a robot write a symphony? Can a robot turn a... canvas into a beautiful masterpiece?"
- "[with genuine interest] Can you?"" (aus dem Film "I, Robot" 2004)
In den letzten Jahren haben einige KI-Veröffentlichungen Aufsehen erregt, zuletzt ChatGPT [5]. ChatGPT ist eine dialogorientierte KI, die auf dem in den letzten Jahren sehr erfolgreichen generativen Sprachmodell GPT [7] basiert. Die multimodale GPT4-Version von ChatGPT wurde während der DHd23 veröffentlicht. Durch passend gewählte Anfragen (auch genannt "Prompts") kann man in einem Dialog das Modell zu interessanten (und vor allem kohärenten) Antworten bewegen. Neben den gesellschaftlichen Implikationen, die in den nächsten Jahren durchaus diskutiert werden müssen, ist auch wieder die interessante Frage nach "Artificial General Intelligence" [11] entbrannt.
Dabei wird im Wesentlichen in zwei große Klassen von KI unterteilt: "strong" oder "weak AI" [8]. Während schwache KI (manchmal auch "spezialisierte" oder "narrow" KI) im Wesentlichen in der Lage ist, einzelne, durchaus komplexe Aufgaben auf menschlichem Niveau und auch darüber hinaus (z.B. Schach) zu erfüllen, wäre eine starke (oder auch "general") KI gleichzusetzen mit dem menschlichen Geist ("the computer is not merely a tool in the study of the mind; rather, the appropriately programmed computer really is a mind" [8]). Dabei gibt es zwei Lesarten in dieser Definition: Erstens, starke KI unterscheidet sich von schwacher KI hauptsächlich durch die Spezifität des Tasks: Eine starke KI müsste nicht nur in der Lage sein, eine Aufgabe mit Bravour zu meistern, sondern mehrere (oder alle?) gestellte Aufgaben fähig erledigen oder zumindest in der Lage sein Sie effizient zu erlernen. Diese Lesart führt zu einer Generalisierung des Modells, die weiterhin (eingeschränkte) Messbarkeit zulässt und wird argumentativ eingesetzt, um Test-Frameworks vorzuschlagen. Chollet [9] geht von Intelligenz im Sinne der Kapazität aus, sich effizient neue Fähigkeiten anzueignen, und schlägt entsprechende Benchmarks vor. Legg und Hutter [13] schlagen (ähnlich wie Wozniak's Kaffee-Test) vor, dass eine Maschine in der Lage sein müsse, Aufgaben in einer breiten Vielfalt verschiedener Umgebungen zu erfüllen ("to achieve goals in a wide range of environments").
Zweitere Lesart beschäftigt sich eher mit dem "Wie?": Während schwache KI sich nur so verhält als wäre sie intelligent ("simulate to think"), müsste starke KI in der Lage sein, "wirklich" zu denken ("actually thinking") [1, p. 1026]. (Hier entsteht im Übrigen einer der Kritikpunkte am von Turing vorgeschlagenen Turing-Tests für starke KI: Wie soll man in einem Gespräch zwischen menschen-ähnlichem Denken und einer guten Simulation davon unterscheiden können?) Während die schwache KI durch ihre Aufgabenspezialisierung oft gut quantifizierbar ist, sind Eigenschaften, die unter diese starke KI fallen, gar nicht mehr so einfach zu fassen. Eigenschaften, die immer wieder genannt werden, sind: logisches Denken, Kommunikation in natürlicher Sprache, Entscheidungen mit Unsicherheiten, Planen, Lernen und Anpassung an äußere Umstände. Ganz zu schweigen von Eigenschaften wie Agency, Empathie, Emotionen oder gar Bewusstsein und intrinsische Motivation.
Während noch kein Konsens über eindeutige Tests auf diese Eigenschaften herrscht, wird zumindest die Hypothese der schwachen KI heute nicht mehr infrage gestellt. Arbeiten der letzten Jahre zeigen deutlich, dass Implementationen von maschinellen Lernalgorithmen, allen voran der Durchbruch der vortrainierten Sprachmodelle der letzten Jahre und zuletzt eben auch ChatGPT, bei verschiedenen und komplexen (messbaren) Aufgaben oft mindestens menschenähnliche Qualität liefern und somit an die Grenzen des Begriffs der spezialisierten KI stoßen. Kürzlich wurde GPT4 einer breiten Studie von Fähigkeiten unterzogen, "that span mathematics, coding, vision, medicine, law, psychology and more" [11], in der argumentiert wird, es wäre nicht abwegig, GPT4 als ein sehr frühes, unvollständiges Stadium von starker KI zu bezeichnen. Die Studie erkennt aber auch an, dass alle durchgeführten Tests rein phänomenologischer Natur sind. Zur Diskussion, wie experimentelle Tests aussehen könnten, um Eigenschaften einer etwaigen starken KI zu untersuchen, schalten sich natürlich auch angrenzende Gebiete wie die Psychologie und Neurowissenschaften ein [15].
Ein Aspekt, der immer wieder Uneinigkeit erzeugt, ist der Begriff der künstlichen Kreativität [2, 15] und, spezifischer, auch Humor als Form der Kreativität, den manche sogar als die letzte Überlegenheit von Mensch über Maschine wähnen [14, 3] Humor wurde schon von Alan Turing in seiner Arbeit "Computing Machinery and Intelligence" als ein Beispiel in einem (Gegen-)Argument angeführt ("argument from disability"). Dabei werden die KI-Methoden in ihrer "Kreativität" durchaus immer öfter nach menschlichen Maßstäben bewertet [10,6] (und gewinnen in seltenen Fällen sogar Preise [12]). Ob es sich bei den generierten Texten (oder Bildern) wirklich um Kreativität handelt oder nur um statistische geschickte Rekombination und was hier der Unterschied ist, ist ein strittiges Thema, oder wie ChatGPT selbst sagen würde:
Prompt:Are you creative?
As an AI language model, I have the ability to generate responses that can be considered creative in certain contexts. For example, I can generate unique and original sentences or ideas based on patterns and associations I have learned from my training data. However, my "creativity" is still limited by the data I was trained on and the algorithms that govern my responses. While I can generate novel outputs, I do not possess the human-like creativity that stems from emotions, personal experiences, and inspiration.
oder
Prompt: Do you possess a sense of humor?
As an AI language model, I do not have personal feelings or emotions, including a sense of humor. However, I can generate responses that are intended to be humorous based on patterns and algorithms learned from large datasets of human language. So, while I don't "possess" humor in the way that humans do, I can certainly attempt to generate humorous responses.
Trotzdem wird immer wieder berichtet, wie ChatGPT als Werkzeug den kreativen Prozess und die Ideenfindung unterstützen kann [6]. Um für die nächste anstehende Konferenz vorbereitet zu sein, habe ich deshalb ChatGPT nach humorvollen (aber natürlich wissenschaftlich relevanten) Ideen für neue Workshops, Paper und Poster gefragt.
Um relevante Titelvorschläge zu bekommen habe ich ChatGPT für die Erstellung neuer Vorschläge immer wieder 100 zufällige Titel der letzten DHd Konferenzen gegeben, mit der Anweisung ähnliche oder Weiterentwicklungen der gegebenen Themen zu erstellen: "Here's a selection of workshop titles from a German Digital Humanities conferences over the last few years. Can you think of 10 new and innovative workshop titles for next year?" Diese Anfrage habe ich dann für Paper und Poster wiederholt. Wenn im weiteren Verlauf die erzeugten Vorschläge zu weit vom Kontext abwichen, habe ich erneut 100 Titel als Beispiele gegeben. Um anschließend humorvolle Vorschläge zubekommen habe ich noch um eine Abwandlung gebeten: "Since it's te 10th year anniversary: Can you think of slightly humorous but still scientific paper titles?". Da die Ausgaben, insbesondere der anschließend erzeugte HTML-Code ("Can you create another html website that lists the conference program?"), die Ausgabekapazität der kostenfreien Variante überschritten haben, mussten die entstandenen Codeteile nach der Aufforderung "continue" manuell zusammenkopiert werden.
Hier die Ergebnisse (Text und HTML-Code generiert von ChatGPT):
(Humorous) Digital Humanities Conference 2024
Don't worry, we won't be relying solely on my AI-powered brain to make this happen. We'll be using a mix of good old-fashioned creativity and some of the latest technologies to create a program that will make even the most serious academics crack a smile.
So, grab your coffee, sit back, and let's start brainstorming. I promise we'll make this conference so entertaining, attendees won't even realize they're learning!:
Workshops
- How to Survive 10 Years of Digital Humanities without Losing Your Mind
This workshop will provide strategies for maintaining your mental health and well-being over a long-term career in digital humanities, including tips for managing stress, avoiding burnout, and fostering a healthy work-life balance.
- The Digital Humanities Midlife Crisis: Navigating Change and Finding Meaning
This workshop will explore the challenges of mid-career transitions in digital humanities, and offer strategies for navigating change, finding renewed purpose, and creating a sustainable career path.
- The 10-Year Itch: Keeping the Spark Alive in Your Digital Humanities Projects
This workshop will explore strategies for sustaining engagement and momentum in long-term digital humanities projects, including tips for cultivating creativity, adapting to change, and avoiding burnout.
- The Great DH Bake-Off: Can Computers Make a Better Cupcake?
This lighthearted workshop will explore the intersection of digital humanities and baking, using computational tools and techniques to experiment with new recipes and techniques. Participants will have the opportunity to taste-test the results!
- DH Karaoke Night: Singing Our Way to Interdisciplinary Collaboration
This interactive workshop will use the power of music to bring together scholars from different disciplines and foster new collaborations. Participants will have the opportunity to showcase their vocal talents and make new connections.
- The DH Olympics: Competing for Gold Medals in Text Mining, Network Analysis, and More
This workshop will feature a series of competitive challenges using digital humanities tools and techniques, with prizes awarded to the top performers. Participants will have the opportunity to learn new skills and showcase their expertise.
- Digital Humanities Anonymous: Support Group for Those Addicted to Big Data
This workshop will provide a safe space for digital humanities scholars struggling with addiction to big data, offering support, encouragement, and strategies for overcoming the compulsion to collect and analyze massive amounts of information.
- The DH Roast: Good-Natured Ribbing of Our Beloved Digital Tools and Techniques
This workshop will use humor to poke fun at some of the more frustrating and challenging aspects of digital humanities research, including data cleaning, software bugs, and the challenges of interdisciplinary collaboration. Participants will have the opportunity to share their own humorous stories and experiences.
- DH Stand-Up Comedy Hour: Finding the Humor in XML and Metadata Schemas
This workshop will use stand-up comedy as a medium for exploring the quirks and frustrations of working with digital humanities tools and techniques, including XML markup, metadata schemas, and other technical aspects of the field. Participants will have the opportunity to share their own comedic takes on digital humanities.
Talks
- The Digital Humanities of Memes: Analyzing the Impact of Rick-rolling on Society
This paper could explore the history and impact of the Rickroll meme, looking at how it spread through digital media and its impact on popular culture.
- LOLcats and the Art of Digital Storytelling
This talk could explore how LOLcats (pictures of cats with humorous captions) use visual storytelling techniques to communicate messages and emotions online.
- The Emoji Conundrum: Navigating the Complexities of Digital Communication
This paper could analyze the use of emojis in digital communication, looking at how they can convey meaning and emotion, but also pose challenges for cross-cultural communication.
- From Papyrus to Pixels: A Brief History of Sassy Historical Insults
This talk could explore the long tradition of witty and humorous insults in historical texts, highlighting some of the most memorable examples and how they've evolved over time.
- The Digital Afterlife: How to Make Your Social Media Accounts Haunt Your Loved Ones
This paper could explore the ethical and legal challenges of dealing with a person's digital presence after they've passed away, and offer humorous suggestions for how to create a "haunting" social media presence.
- The Art of Clickbait Titles in 19th-Century Literature
This talk explores the use of sensational titles in 19th-century literature, and how they were used to grab readers' attention in an era before clickbait headlines.
- Close Reading with a Sense of Humor: Finding the Funny in Serious Literature
This paper examines the role of humor in serious literary works, and argues that a lighthearted approach can actually help readers to better understand and appreciate difficult texts.
- The Great Emoji Debate: Can We Use Them in Literary Analysis?
This talk takes a lighthearted look at the use of emojis in literary analysis, exploring the pros and cons of incorporating these popular symbols into scholarly work.
- Tales from the Editing Room Floor: The Best Literary Rejections of All Time
This presentation looks at some of the most amusing and outrageous rejection letters sent to famous authors, and considers what we can learn from their experiences.
- The Literary Catwalk: A Fashion History of Famous Literary Characters
This talk explores the fashion and style of some of literature's most beloved characters, from Elizabeth Bennet's empire-waist gowns to Holden Caulfield's iconic red hunting hat.
Posters
- "Slaying Dragons and Demons: A Digital Humanities Approach to Medieval Monster Studies"
A humorous take on studying monsters in medieval literature, using digital humanities tools to analyze their cultural significance and impact on popular imagination.
- "Can You Spot the Fake News? Using Digital Tools to Combat Misinformation"
A playful investigation into the prevalence of fake news and disinformation online, using digital tools to identify and combat these harmful trends.
- "From the Archives to Your Screen: Preserving and Accessing Cultural Heritage in the Digital Age (Without Losing Your Mind)"
A humorous exploration of the challenges and opportunities presented by digitizing cultural heritage materials, such as archives, manuscripts, and artwork.
- "The Meme-ification of Politics: Analyzing the Role of Memes in Political Discourse and Public Opinion"
A lighthearted look at the role of memes in shaping public opinion and political discourse, using data analysis and social media research.
- "Data Science and the Art of Procrastination: Using Digital Humanities to Stay Productive (Even When You're Not)"
A playful take on the challenges of staying productive in the digital age, using digital humanities tools and strategies to overcome procrastination and distraction.
- The Cats of Literature: A Computational Analysis of Feline Representation in Fiction
A humorous exploration of the portrayal of cats in literature, using digital tools to analyze their prevalence and characteristics in various works of fiction.
- Breaking Bad Books: A Network Analysis of Literary References in the TV Show
A playful take on the popular TV series, using network analysis to map out the literary references and allusions scattered throughout the show.
- The Emoji-tionary: Mapping Emotional Connotations of Emoji Usage in Online Literature Communities
A lighthearted investigation of the emotional associations and usage patterns of emojis within online literary communities, using data mining techniques.
- Can You Hear Me Now? An Exploration of Nonverbal Communication in Video Game Avatars
A fun look at how video game avatars convey emotion and meaning through nonverbal cues, such as body language and facial expressions.
- Digital Humanities: The Musical
A playful and creative poster presentation that uses song and dance to highlight the diverse range of digital humanities research and its impact on society.
Insgesamt war die DHd ein tolles Erlebnis und ich hoffe, im nächsten Jahr wieder dabei sein zu können. Bis dahin wünsche ich allen viel Spaß beim Ausprobieren von neuen Ideen!