PoEB: The Digital Firm 2024-2025 PDF

PoEB: The Digital Firm 2024-2025 PoEB 2: The Digital Firm 3. AI and the Future of Work Prof. dr. Hans P. Borgman Prof. Dr. Hans Borgman 0 PoEB: The Digital Firm 2024-2025 The Digital...

PoEB: The Digital Firm 2024-2025 PoEB 2: The Digital Firm 3. AI and the Future of Work Prof. dr. Hans P. Borgman Prof. Dr. Hans Borgman 0 PoEB: The Digital Firm 2024-2025 The Digital Firm: Course Outline Introduction 1. IT and Strategy 2. Winner Takes All? 3. AI and the Future of Work 4. What is IT Success? 1 Prof. Dr. Hans Borgman 1 PoEB: The Digital Firm 2024-2025 WEF Future of Jobs (2025) 2 For link see Canvas Prof. Dr. Hans Borgman 2 PoEB: The Digital Firm 2024-2025 WEF Future of Jobs (2025) % 3 Prof. Dr. Hans Borgman 3 PoEB: The Digital Firm 2024-2025 WEF Future of Jobs (2025) # 4 Good to keep in mind that relative changes are very different from absolute ones. Prof. Dr. Hans Borgman 4 PoEB: The Digital Firm 2024-2025 Will AI kill jobs? Lump of Labor Fallacy & Jevons Paradox Lump of labour fallacy: The mistaken belief that here is a fixed amount of work to be done, and efficiency gains imply a reduced need for workers (Schloss, 1891). Jevons Paradox: A decrease in the cost (or price) of a good or service (through efficiency gains) will increase the quantity demanded (Jevons, 1865). Prof. Dr. Hans Borgman 5 Theories of Digital Business 2024-2025 AI on the hype curve 6 In our first session I showed the August 2024 hype curve specifically focused on AI. Here is another one, this time showing all 'emerging technologies' from August 2024. Note how ‘Artificial Intelligence’ (AI, my highlights in yellow) related technologies dominate this chart. Google and others are now so focused on AI that they rebranded their R&D as '[Google] AI'. Prof. Dr. Hans P. Borgman 6 Theories of Digital Business 2024-2025 AI origins “The science and engineering of making intelligent machines, especially intelligent computer programs” ”…every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. … (within 5-500 years)” (McCarthy, 1955) ================================================ EEEEEEEE L IIIIIII ZZZZZZZ AAA E L I Z A A E L I Z A A EEEEE L I Z A A E L I Z AAAAAAA E L I Z A A Interdisciplinary: cognitive science, computer science, neuroscience, EEEEEEEE LLLLLLLL IIIIIII ZZZZZZZ ================================================ Human: Men are all alike A A psychology, mathematics, philosophy, linguistics ELIZA: IN WHAT WAY? Human: They’re always bugging us about something or other ELIZA: CAN YOU THINK OF A SPECIFIC EXAMPLE? Human: Well, my boyfriend made me come here ELIZA: YOUR BOYFRIEND MADE YOU COME HERE Human: He says I’m depressed most of the time ELIZA: I AM SORRY TO HEAR YOU ARE DEPRESSED Human: It’s true. I’m unhappy ELIZA: DO YOU THINK COMING HERE WILL HELP YOU NOT TO BE UNHAPPY Human: I need some help, that much seems certain 7 AI research and applications go back to the middle of the last century. After an initial period of growth (and hype) there was a long 'AI winter' of several decades in which progress was relatively limited. There have been brief 'AI summers', notable with expert systems going through a short-lived hype cycle in the 1980's, but never has AI dominated the headlines as much as in the last few years. The Eliza animation shows a crude dialogue system developed in the 1960s by MIT's Joseph Weizenbaum mimicking a 'Rogerian' therapist taking keywords from its patient to probe deeper. Test persons felt that the system 'understood' them and projected empathy and other character traits on what was essentially a very simple natural language processing system (by today's standards). Prof. Dr. Hans P. Borgman 7 Theories of Digital Business 2024-2025 Early days AI: Can machines think? The imitation game (’Turing test’: Alan Turing, 1950) Computer ChatGPT Human student barrier Computer ? Human Interrogator Professor 8 The original article: A. M. TURING, I.—COMPUTING MACHINERY AND INTELLIGENCE, Mind, Volume LIX, Issue 236, October 1950, Pages 433–460, https://doi.org/10.1093/mind/LIX.236.433 Prof. Dr. Hans P. Borgman 8 Theories of Digital Business 2024-2025 Rule-based vs Machine Learning Rule-based: If-Then (explicitly coded, static) Machine Learning: Input – Output pattern matching (data, continuous improvement) or 9 Stills from https://www.youtube.com/watch?v=cAkklvGE5io Prof. Dr. Hans P. Borgman 9 Theories of Digital Business 2024-2025 2013 Areas of AI Speaking/listening/understanding/translating/ writing: Natural Language Processing Reasoning (abstract/concrete, inductive/deductive) 2018 Sensory-motor intelligence (vision, motor skills, touch, …) Learning: Artificial General Intelligence (AGI) 2025 Creativity (?) 10 'Her' video trailer: https://www.youtube.com/watch?v=dJTU48_yghs Google duplex (2018!): announcement from Google I/O event: https://www.youtube.com/watch?v=D5VN56jQMWM The Google Deepmind Veo 2 compilation is at https://www.youtube.com/watch?v=G9RDHs9nx04, and the full 'The Heist' video (entirely generated by prompts can be found at https://www.youtube.com/watch?v=lFc1jxLHhyM Prof. Dr. Hans P. Borgman 10 Theories of Digital Business 2024-2025 AI and ML: Getting you in the top 5% "a type of AI often broadly defined as software with the ability to learn or improve without being explicitly programmed" "programming computers to recognize patterns from large amounts of examples and applying these patterns to new situations" Hilary Mason explaining AI at multiple levels. We watch the undergrad computer science student segment, but you can start easier or dive deeper and see how much you can follow. Actually, you could well try (and succeed) to follow both the 'grad student' (here: first year PhD) as well as the 'expert' dialogues as they don't add much conceptual complexity. It shows how an understanding of the underlying principles, already explained at the 'college student' level and central to today's class pretty much equip you to be 'in the top 5%' of people understanding AI and machine learning. Prof. Dr. Hans P. Borgman 11 Theories of Digital Business 2024-2025 ML: Learning by doing Supervised learning: learn from example input-output pairs training set example: sorting cucumbers, loan approval Unsupervised learning: find new patterns without pre-existing labels clustering (what makes observations similar/different) example: customer segmentation, robot soccer 12 The availability of more computing power, better algorithms and more data led to a very strong new surge in AI over the last decade or so. AI research is often (more or less) 'open', with companies such as Google and Microsoft collaborating with academia as well as each other, forming a great example of open innovation. This is not without issues, such as those surrounding OpenAI's language model ChatGPT (not so open anymore, charging for access and keeping part of their innovations closed or proprietary), with wonderful applications but also suitable to flood online discussions with fake contributions, cheat on school assignments, and much more. ‘Nat & Lo’ (Google Youtube series started around 2018 and covers a lot of computer science and AI topics): see https://www.youtube.com/watch?v=r4UjBNZSXjo and https://www.youtube.com/watch?v=bHvf7Tagt18 'AI teaching itself to walk' show Google Deepmind at work (in 2017!), having an animated stick figure teaching itself how to walk, inventing some new techniques along the way ('self-learning'): https://www.youtube.com/watch?v=gn4nRCC9TwQ A more complex task involving robots (rather than animated stick figures) playing football (2023): https://www.youtube.com/watch?v=RbyQcCT6890&t=1s Prof. Dr. Hans P. Borgman 12 Theories of Digital Business 2024-2025 Intelligence: Chess as metaphor How do chess players think? (A.D. de Groot, PhD thesis UvA, 1946) Role of pattern recognition Chess and chunking (Chase & Simon, 1973) 13 References: A.D. de Groot (1946). Het denken van den schaker. Proefschrift Universiteit van Amsterdam, NV Noord-Hollandsche uitgeversmaatschappij, Amsterdam W.G. Chase & H. Simon (1973), Perception in Chess, Cognitive Psychology 4, 55-81 Expert chess players as well as good amateurs all rely to a great degree on their pattern recognition abilities (and therefore also study earlier matches). AI chess computers use this as well, plus they think several moves ahead to calculate the likely value of a particular move. The combination of these ‘brute force’ approaches leads researchers to call this ‘weak AI’ which leads to special-purpose intelligence Prof. Dr. Hans P. Borgman 13 Theories of Digital Business 2024-2025 2x Image page Ins Special vs General purpose AI: Who is teaching who? ‘Brute force’, special-purpose, narrow or ‘weak ‘self-learning’, general-purpose, AGI or ‘strong AI’ AI’ 2017: Deepmind beats Lee Sedol 1997: Deep Blue beats Kasparov AGI: emerging, competent, expert, virtuoso, and superhuman 14 The 90-minute 'AlphaGo - The Movie': https://youtu.be/WXuK6gekU1Y Prof. Dr. Hans P. Borgman 14 Theories of Digital Business 2024-2025 Big Data & machine learning hidden layer(s) is there an eye in the input layer top-left? ouput layer (image pixels) ‘this is probably a face’ is there an eye in the top-right? (with context: 'probably Angela Merkel) is there a nose in the middle? is there a mouth at the bottom? is there hair on top? + context 15 This is a (hypothetical) deep learning model working on image recognition. Note that this is not a simple pattern recognition task comparing a picture to a database and requiring a 100% match but instead a probability model involving multiple layers and following a 'divide and conquer' strategy to break down the task into manageable parts. We humans are extremely good at this: we see a large airplane with airstairs with a 'red carpet' and a single person getting out, so we deduce that this person must be important. We read the country name, maybe recognise the physique of the person, etc. Many people will also 'recognise' Angela Merkel and place a picture in time by combining a partial face with a mask. Computers require an elaborate deep learning 'neural network' to accomplish the same task. Prof. Dr. Hans P. Borgman 15 Theories of Digital Business 2024-2025 2x Image page Inse Big Data & ‘deep’ machine learning 16 Cucumber sorting and many other 'deep learning' AI tasks explained in this excellent Google video: https://youtu.be/_rdINNHLYaQ Video from 2016, illustrating how the underlying mechanisms of machine learning haven't dramatically changed, although the availability of more data and more accessible and better algorithms (and more computing power) have dramatically changed the effectiveness of today's AI solutions. Prof. Dr. Hans P. Borgman 16 Theories of Digital Business 2024-2025 Top 5%: are we there yet? Think back of the 'Google Duplex' video, and then look at the above figure from a paper describing the inner workings of the underlying AI system. Can you follow this with some explanation (you should)? The system essentially consists of an ASR (= automated speech recognition) module feeding the audio (depicted as the green wave) with context information about the task into a 'deep learning' neural network that outputs text that is then in turn translated to speech by the TTS module (= text to speech). Prof. Dr. Hans P. Borgman 17 Theories of Digital Business 2024-2025 Understanding LLMs such as ChatGPT LLM = Large Language Model Chat = memory of earlier part of conversation Generative = creates new data Pre-trained = model is already prepared using data Transformer = 1.5 bn 1.7 T GPT 4 o3 Taking our knowledge about the 'underlying principles' of AI and machine learning to a currently hot issue, LLMs such as ChatGPT. Animation taken from https://jalammar.github.io/visualizing-neural-machine-translation-mechanics-of- seq2seq-models-with-attention/, a page offering a thorough explanation (well beyond what we cover here) of sequence-to-sequence models such as ChatGPT that take sentences as input and produce sentences as output. 'Who has it bigger?': benchmarks for LLMs and tests for achieving 'AGI' are highly debatable. Some LLMs perform good on specific tasks and less on other tasks. Still, rankings exist, such as https://llm-stats.com/ Prof. Dr. Hans P. Borgman 18 Theories of Digital Business 2024-2025 Are we there yet? Our 5% target Prof. Dr. Hans P. Borgman 19 PoEB: The Digital Firm 2024-2025 The Digital Firm: Course Outline Introduction 1. IT and Strategy 2. Winner Takes All? 3. AI and the Future of Work 4. What is IT Success? 2 0 Prof. Dr. Hans Borgman 20