{ "version": "7-en", "exhibits": [ { "room_id": "room-0", "room_title": "Greek precursors", "room_context": "This room shows that the desire for self-moving machines, logical rules and artificial beings is far older than the electronic computer.", "catalogue": "0.1", "title": "Hephaestus’ golden assistants", "year": "~700BCE", "description": "In the Iliad, the craftsman god has made golden assistants who resemble living young women. Homer says they have intelligence, voice and strength. It may be one of the earliest literary descriptions of “intelligent machines”.", "image": "img/0-1.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary literary source", "source": "Homer, Iliad 18.417–420.", "supports": "The description of Hephaestus’ golden assistants as having intelligence, voice and strength.", "status": "Historical text; the connection with artificial life is curatorial.", "url": "https://www.perseus.tufts.edu/hopper/text?doc=Hom.+Il.+18&fromdoc=Perseus%3Atext%3A1999.01.0134" }, "slug": "hephaestus-golden-assistants", "url": "/en/exhibits/hephaestus-golden-assistants/" }, { "room_id": "room-0", "room_title": "Greek precursors", "room_context": "This room shows that the desire for self-moving machines, logical rules and artificial beings is far older than the electronic computer.", "catalogue": "0.2", "title": "Aristotle and logic", "year": "~340BCE", "description": "Aristotle systematises reasoning: rules that lead from valid premises to valid conclusions. Many centuries later, rule-based AI would build on this idea. In the Politics he also imagines tools that could perform their work by themselves, without human hands.", "image": "img/0-2.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary philosophical source", "source": "Aristotle, Prior Analytics; Politics I, 1253b.", "supports": "The systematisation of reasoning and the passage imagining tools able to carry out their own work.", "status": "Historical text; the connection with later symbolic AI is curatorial.", "url": "https://www.perseus.tufts.edu/hopper/text?doc=Perseus%3Atext%3A1999.01.0058%3Abook%3D1%3Asection%3D1253b" }, "slug": "aristotle-and-logic", "url": "/en/exhibits/aristotle-and-logic/" }, { "room_id": "room-0", "room_title": "Greek precursors", "room_context": "This room shows that the desire for self-moving machines, logical rules and artificial beings is far older than the electronic computer.", "catalogue": "0.3", "title": "Ctesibius and self-regulation", "year": "~270BCE", "description": "In Alexandria, Ctesibius improves the water clock with a float that regulates the flow of water automatically. It is among the earliest known examples of automatic control through feedback: the mechanism corrects its own operation. Centuries later, feedback would become central to cybernetics and AI. He also developed pumps and the water organ.", "image": "img/0-ktesibios.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Ancient technical testimony", "source": "Vitruvius, De Architectura IX.8, on Ctesibius and water clocks.", "supports": "The use of floats and regulating mechanisms in water clocks and other automatic devices associated with Ctesibius.", "status": "Well-attested ancient engineering; the term “feedback” is a modern interpretation.", "url": "https://www.perseus.tufts.edu/hopper/text?doc=Vitr.+9.8&lang=original" }, "slug": "ctesibius-and-self-regulation", "url": "/en/exhibits/ctesibius-and-self-regulation/" }, { "room_id": "room-0", "room_title": "Greek precursors", "room_context": "This room shows that the desire for self-moving machines, logical rules and artificial beings is far older than the electronic computer.", "catalogue": "0.4", "title": "Talos", "year": "~250BCE", "description": "In Apollonius of Rhodes’ Argonautica, a bronze giant made by Hephaestus guards Crete by circling the island. A mythical robot-like guardian, with a single vulnerable point.", "image": "img/0-3.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary literary source", "source": "Apollonius of Rhodes, Argonautica, Book IV.", "supports": "The story of the bronze Talos as guardian of Crete.", "status": "Mythological and literary evidence, not a historical machine.", "url": "https://beta.perseus.tufts.edu/urn:cts:greekLit:tlg0001.tlg001.seatonrevised2026-eng2/card:3.1246/" }, "slug": "talos", "url": "/en/exhibits/talos/" }, { "room_id": "room-0", "room_title": "Greek precursors", "room_context": "This room shows that the desire for self-moving machines, logical rules and artificial beings is far older than the electronic computer.", "catalogue": "0.5", "title": "The Antikythera Mechanism", "year": "~100BCE", "description": "Recovered from a shipwreck in 1901 and dated to roughly the second to first century BCE, the mechanism used dozens of gears to model the positions of the Sun and Moon and predict eclipses. It is often described as the first analogue computer. The original is held by the National Archaeological Museum in Athens.", "image": "img/0-4.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Archaeological object and scientific study", "source": "Freeth, T. et al. (2006), Nature 444, 587–591.", "supports": "The function, dating and mechanical complexity of the Antikythera Mechanism.", "status": "Strongly documented archaeological object.", "url": "https://doi.org/10.1038/nature05357" }, "slug": "antikythera-mechanism", "url": "/en/exhibits/antikythera-mechanism/" }, { "room_id": "room-0", "room_title": "Greek precursors", "room_context": "This room shows that the desire for self-moving machines, logical rules and artificial beings is far older than the electronic computer.", "catalogue": "0.6", "title": "Hero of Alexandria", "year": "~60CE", "description": "Hero describes automatic spectacles, including birds made to sing with air and water, and a cart that moves by itself. A rope wound around its axle, with pegs placed at selected points, determines when it travels straight and when it turns. In modern terms, it resembles a program stored in rope.", "image": "img/0-5.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary technical text", "source": "Hero of Alexandria, Automata; Pneumatica.", "supports": "Descriptions of automata and programmable mechanical sequences.", "status": "Historical engineering, not artificial intelligence in the modern sense.", "url": "" }, "slug": "hero-of-alexandria", "url": "/en/exhibits/hero-of-alexandria/" }, { "room_id": "room-1", "room_title": "The beginnings", "room_context": "Here the basic questions of modern AI take shape: how can thought be described, how can a machine learn, and what does it mean for a machine to hold a convincing conversation?", "catalogue": "1.1", "title": "The artificial neuron", "year": "1943", "description": "McCulloch and Pitts describe a brain neuron as a simple mathematical switch: it either fires or it does not. Their abstraction is an ancestor of today’s artificial neural networks.", "image": "img/1-1.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "McCulloch, W. & Pitts, W. (1943), Bulletin of Mathematical Biophysics.", "supports": "A mathematical model of neurons as logical units.", "status": "Primary scientific source.", "url": "https://doi.org/10.1007/BF02478259" }, "slug": "artificial-neuron", "url": "/en/exhibits/artificial-neuron/" }, { "room_id": "room-1", "room_title": "The beginnings", "room_context": "Here the basic questions of modern AI take shape: how can thought be described, how can a machine learn, and what does it mean for a machine to hold a convincing conversation?", "catalogue": "1.2", "title": "Turing’s imitation game", "year": "1950", "description": "Alan Turing proposes a test: if, during a written conversation, you cannot tell whether you are speaking to a person or a machine, should the machine count as intelligent?", "image": "https://commons.wikimedia.org/wiki/Special:FilePath/Alan_Turing_Aged_16.jpg?width=300", "image_kind": "Historical photograph or digital object", "credit": "Ο Alan Turing στα 16 του, περίπου το 1928. Κοινό κτήμα, μέσω Wikimedia Commons .", "documentation": { "type": "Primary scientific and philosophical paper", "source": "Turing, A. M. (1950), “Computing Machinery and Intelligence”, Mind 59(236).", "supports": "The formulation of the imitation game as a practical way to discuss machine intelligence.", "status": "Primary source.", "url": "https://doi.org/10.1093/mind/LIX.236.433" }, "slug": "turing-imitation-game", "url": "/en/exhibits/turing-imitation-game/" }, { "room_id": "room-1", "room_title": "The beginnings", "room_context": "Here the basic questions of modern AI take shape: how can thought be described, how can a machine learn, and what does it mean for a machine to hold a convincing conversation?", "catalogue": "1.3", "title": "The Dartmouth workshop", "year": "1956", "description": "A group of researchers meets for a summer in the United States. The phrase “artificial intelligence”, proposed the previous year, becomes the name of a new scientific field.", "image": "img/1-3.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary archival document", "source": "McCarthy, Minsky, Rochester & Shannon (1955), Dartmouth proposal.", "supports": "The use of the term artificial intelligence and the plan for the 1956 summer research project.", "status": "Primary historical document.", "url": "https://www-formal.stanford.edu/jmc/history/dartmouth.pdf" }, "slug": "dartmouth-workshop", "url": "/en/exhibits/dartmouth-workshop/" }, { "room_id": "room-1", "room_title": "The beginnings", "room_context": "Here the basic questions of modern AI take shape: how can thought be described, how can a machine learn, and what does it mean for a machine to hold a convincing conversation?", "catalogue": "1.4", "title": "The perceptron", "year": "1958", "description": "Frank Rosenblatt builds a machine that learns to recognise simple patterns from examples instead of having every rule written for it.", "image": "img/1-4.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "Rosenblatt, F. (1958), Psychological Review 65(6), 386–408.", "supports": "The perceptron as a model that adjusts its parameters from examples.", "status": "Primary source.", "url": "https://doi.org/10.1037/h0042519" }, "slug": "perceptron", "url": "/en/exhibits/perceptron/" }, { "room_id": "room-1", "room_title": "The beginnings", "room_context": "Here the basic questions of modern AI take shape: how can thought be described, how can a machine learn, and what does it mean for a machine to hold a convincing conversation?", "catalogue": "1.5", "title": "ELIZA, the first chatbot", "year": "1966", "description": "Joseph Weizenbaum at MIT builds a program that imitates a psychotherapist. It understands nothing, yet many users feel that it understands them.", "image": "img/1-5.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "Weizenbaum, J. (1966), “ELIZA”, Communications of the ACM 9(1), 36–45.", "supports": "How ELIZA operates through keywords and decomposition/reassembly rules.", "status": "Primary source.", "url": "https://doi.org/10.1145/365153.365168" }, "slug": "eliza-first-chatbot", "url": "/en/exhibits/eliza-first-chatbot/" }, { "room_id": "room-2", "room_title": "The AI winters", "room_context": "Progress was not continuous. Periods of great expectations were followed by technical limits, disappointment and cuts in funding.", "catalogue": "2.1", "title": "The limits of the perceptron", "year": "1969", "description": "Minsky and Papert show mathematically that a simple perceptron cannot solve some basic classes of problems. Research on neural networks then loses momentum for years.", "image": "img/2-1.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary technical monograph", "source": "Minsky, M. & Papert, S. (1969), Perceptrons, MIT Press.", "supports": "The mathematical analysis of limitations of single-layer perceptrons.", "status": "Primary technical source; its connection with later funding decline is a historical interpretation.", "url": "https://mitpress.mit.edu/9780262130431/perceptrons/" }, "slug": "limits-of-the-perceptron", "url": "/en/exhibits/limits-of-the-perceptron/" }, { "room_id": "room-2", "room_title": "The AI winters", "room_context": "Progress was not continuous. Periods of great expectations were followed by technical limits, disappointment and cuts in funding.", "catalogue": "2.2", "title": "The Lighthill Report", "year": "1973", "description": "A report commissioned by the British government argues that AI research has failed to meet many of its goals. Funding is cut and the first AI winter begins in Britain.", "image": "img/2-2.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Government report", "source": "Lighthill, J. (1973), Artificial Intelligence: A General Survey, Science Research Council.", "supports": "The critical assessment of British AI research that influenced funding decisions.", "status": "Primary institutional document.", "url": "https://www.aiai.ed.ac.uk/events/lighthill1973/lighthill.pdf" }, "slug": "lighthill-report", "url": "/en/exhibits/lighthill-report/" }, { "room_id": "room-2", "room_title": "The AI winters", "room_context": "Progress was not continuous. Periods of great expectations were followed by technical limits, disappointment and cuts in funding.", "catalogue": "2.3", "title": "Expert systems", "year": "1980", "description": "Programs containing hundreds or thousands of “if… then…” rules enter business use. Digital Equipment’s XCON configures computer orders and is credited with major savings.", "image": "img/2-3.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "McDermott, J. (1982), “R1: A Rule-Based Configurer of Computer Systems”, Artificial Intelligence.", "supports": "The operation and business use of XCON/R1 as an expert system.", "status": "Primary source.", "url": "https://doi.org/10.1016/0004-3702(82)90021-2" }, "slug": "expert-systems", "url": "/en/exhibits/expert-systems/" }, { "room_id": "room-2", "room_title": "The AI winters", "room_context": "Progress was not continuous. Periods of great expectations were followed by technical limits, disappointment and cuts in funding.", "catalogue": "2.4", "title": "The second AI winter", "year": "1987", "description": "Expert systems prove expensive to maintain and brittle when faced with exceptions. At the same time, the market for specialised AI computers collapses.", "image": "img/2-4.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Historical synthesis", "source": "Historical literature on expert systems and the Lisp machine market.", "supports": "The commercial downturn of the late 1980s and the maintenance problems of rule-based systems.", "status": "A synthesis of multiple sources rather than a single event.", "url": "" }, "slug": "second-ai-winter", "url": "/en/exhibits/second-ai-winter/" }, { "room_id": "room-2", "room_title": "The AI winters", "room_context": "Progress was not continuous. Periods of great expectations were followed by technical limits, disappointment and cuts in funding.", "catalogue": "2.5", "title": "Deep Blue defeats Kasparov", "year": "1997", "description": "IBM’s computer defeats the reigning world chess champion. It does not “think” like a human player: it searches enormous numbers of positions at high speed. AI returns to front-page news.", "image": "https://commons.wikimedia.org/wiki/Special:FilePath/Deep_Blue.jpg?width=800", "image_kind": "Historical photograph or digital object", "credit": "Φωτογραφία: James the photographer, CC BY 2.0 , μέσω Wikimedia Commons", "documentation": { "type": "Corporate history and technical paper", "source": "IBM, Deep Blue history; Campbell, Hoane & Hsu (2002), Artificial Intelligence.", "supports": "Deep Blue’s 1997 match victory over Garry Kasparov and its computational approach.", "status": "Well-documented historical event.", "url": "https://www.ibm.com/history/deep-blue" }, "slug": "deep-blue-beats-kasparov", "url": "/en/exhibits/deep-blue-beats-kasparov/" }, { "room_id": "room-3", "room_title": "How a machine learns", "room_context": "Machine learning changes the problem: instead of writing every rule, we provide data and examples. Their gaps and biases can pass into the model too.", "catalogue": "3.1", "title": "The term “machine learning”", "year": "1959", "description": "IBM researcher Arthur Samuel develops a checkers program that improves by playing repeatedly and uses the term “machine learning”.", "image": "img/3-1.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "Samuel, A. L. (1959), IBM Journal of Research and Development.", "supports": "The checkers program that improves through experience and the use of the term machine learning.", "status": "Primary source.", "url": "https://doi.org/10.1147/rd.33.0210" }, "slug": "machine-learning-term", "url": "/en/exhibits/machine-learning-term/" }, { "room_id": "room-3", "room_title": "How a machine learns", "room_context": "Machine learning changes the problem: instead of writing every rule, we provide data and examples. Their gaps and biases can pass into the model too.", "catalogue": "3.2", "title": "ImageNet", "year": "2009", "description": "Fei-Fei Li’s team creates a huge database of millions of labelled images. It becomes clear that the quantity and quality of data can matter as much as the algorithm.", "image": "img/3-2.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary conference paper", "source": "Deng, J. et al. (2009), CVPR, “ImageNet: A Large-Scale Hierarchical Image Database”.", "supports": "The creation of ImageNet as a very large hierarchical database of labelled images.", "status": "Primary source.", "url": "https://doi.org/10.1109/CVPR.2009.5206848" }, "slug": "imagenet", "url": "/en/exhibits/imagenet/" }, { "room_id": "room-3", "room_title": "How a machine learns", "room_context": "Machine learning changes the problem: instead of writing every rule, we provide data and examples. Their gaps and biases can pass into the model too.", "catalogue": "3.3", "title": "Gender Shades", "year": "2018", "description": "Joy Buolamwini and Timnit Gebru show that commercial face-analysis systems make far more errors for darker-skinned women than for lighter-skinned men. The result exposes the consequences of unrepresentative data and uneven system performance.", "image": "img/3-3.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary research paper", "source": "Buolamwini, J. & Gebru, T. (2018), Gender Shades, PMLR 81.", "supports": "Large differences in error rates across demographic groups in commercial gender-classification systems.", "status": "Primary evaluation of deployed systems.", "url": "https://proceedings.mlr.press/v81/buolamwini18a.html" }, "slug": "gender-shades", "url": "/en/exhibits/gender-shades/" }, { "room_id": "room-3", "room_title": "How a machine learns", "room_context": "Machine learning changes the problem: instead of writing every rule, we provide data and examples. Their gaps and biases can pass into the model too.", "catalogue": "3.4", "title": "A hiring tool that was abandoned", "year": "2018", "description": "Reports reveal that Amazon dropped an experimental résumé-ranking system after it had learned from historical data to disadvantage women applicants.", "image": "img/3-4.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Investigative journalism", "source": "Reuters (2018), reporting on Amazon’s experimental recruiting tool.", "supports": "That the internal system was abandoned after gender bias was identified in its outputs.", "status": "Reliable journalistic source, not a published technical study.", "url": "" }, "slug": "amazon-hiring-tool", "url": "/en/exhibits/amazon-hiring-tool/" }, { "room_id": "room-4", "room_title": "Neural networks", "room_context": "Neural networks learn by adjusting large numbers of numerical weights. The idea is old, but the scale of data and computing made it far more powerful.", "catalogue": "4.1", "title": "Backpropagation", "year": "1986", "description": "Rumelhart, Hinton and Williams show how a multi-layer network can correct its errors by propagating the error signal backwards. This helps overcome the limitation highlighted by Minsky and Papert.", "image": "img/4-1.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "Rumelhart, Hinton & Williams (1986), Nature 323, 533–536.", "supports": "Training multi-layer networks with back-propagation.", "status": "Primary source.", "url": "https://doi.org/10.1038/323533a0" }, "slug": "backpropagation", "url": "/en/exhibits/backpropagation/" }, { "room_id": "room-4", "room_title": "Neural networks", "room_context": "Neural networks learn by adjusting large numbers of numerical weights. The idea is old, but the scale of data and computing made it far more powerful.", "catalogue": "4.2", "title": "A network that reads", "year": "1989", "description": "Yann LeCun trains a network to recognise handwritten digits in postal ZIP codes. Related systems were later used for tasks such as reading bank cheques.", "image": "img/4-2.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "LeCun et al. (1989), Neural Computation 1(4), 541–551.", "supports": "The use of backpropagation in a network for recognising handwritten postal codes.", "status": "Primary source.", "url": "https://doi.org/10.1162/neco.1989.1.4.541" }, "slug": "a-network-that-reads", "url": "/en/exhibits/a-network-that-reads/" }, { "room_id": "room-4", "room_title": "Neural networks", "room_context": "Neural networks learn by adjusting large numbers of numerical weights. The idea is old, but the scale of data and computing made it far more powerful.", "catalogue": "4.3", "title": "AlexNet", "year": "2012", "description": "A deep neural network trained on graphics processors wins the ImageNet image-recognition competition by a large margin. The modern deep-learning era accelerates.", "image": "img/4-3.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary conference paper", "source": "Krizhevsky, Sutskever & Hinton (2012), NeurIPS 25.", "supports": "AlexNet’s ImageNet performance and the importance of GPU training.", "status": "Primary source.", "url": "https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c-Abstract.html" }, "slug": "alexnet", "url": "/en/exhibits/alexnet/" }, { "room_id": "room-4", "room_title": "Neural networks", "room_context": "Neural networks learn by adjusting large numbers of numerical weights. The idea is old, but the scale of data and computing made it far more powerful.", "catalogue": "4.4", "title": "AlphaGo", "year": "2016", "description": "DeepMind’s system defeats Lee Sedol, one of the world’s leading Go players, in a game long considered exceptionally difficult for machines.", "image": "img/4-4.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "Silver, D. et al. (2016), Nature 529, 484–489.", "supports": "The architecture and results of AlphaGo.", "status": "Primary source.", "url": "https://doi.org/10.1038/nature16961" }, "slug": "alphago", "url": "/en/exhibits/alphago/" }, { "room_id": "room-4", "room_title": "Neural networks", "room_context": "Neural networks learn by adjusting large numbers of numerical weights. The idea is old, but the scale of data and computing made it far more powerful.", "catalogue": "4.5", "title": "Nobel Prize in Physics", "year": "2024", "description": "John Hopfield and Geoffrey Hinton are awarded the 2024 Nobel Prize in Physics for foundational discoveries and inventions that enabled machine learning with artificial neural networks.", "image": "img/4-5.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Official institutional source", "source": "The Royal Swedish Academy of Sciences, Nobel Prize in Physics 2024.", "supports": "The 2024 Nobel Prize in Physics awarded to John Hopfield and Geoffrey Hinton for foundational work related to artificial neural networks.", "status": "Official source.", "url": "https://www.nobelprize.org/prizes/physics/2024/summary/" }, "slug": "nobel-prize-in-physics", "url": "/en/exhibits/nobel-prize-in-physics/" }, { "room_id": "room-5", "room_title": "Language and the first LLMs", "room_context": "Language is treated as a prediction problem. From Shannon’s statistical sequences, the story leads to Transformers and large language models.", "catalogue": "5.1", "title": "Shannon measures prediction", "year": "1948", "description": "Claude Shannon shows that text that resembles English can be generated by measuring which symbols or words tend to follow others.", "image": "img/5-1.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary scientific paper", "source": "Shannon, C. E. (1948), “A Mathematical Theory of Communication”.", "supports": "The mathematical foundations of information and experiments involving prediction and generation of language sequences.", "status": "Primary source.", "url": "https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf" }, "slug": "shannon-measures-prediction", "url": "/en/exhibits/shannon-measures-prediction/" }, { "room_id": "room-5", "room_title": "Language and the first LLMs", "room_context": "Language is treated as a prediction problem. From Shannon’s statistical sequences, the story leads to Transformers and large language models.", "catalogue": "5.2", "title": "Words become numbers", "year": "2013", "description": "With word2vec, each word can be represented as a point in a numerical space. Words used in similar contexts tend to end up close together.", "image": "img/5-2.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary research paper", "source": "Mikolov, T. et al. (2013), “Efficient Estimation of Word Representations in Vector Space”.", "supports": "Learning dense vector representations of words from large text corpora.", "status": "Primary source.", "url": "https://arxiv.org/abs/1301.3781" }, "slug": "words-become-numbers", "url": "/en/exhibits/words-become-numbers/" }, { "room_id": "room-5", "room_title": "Language and the first LLMs", "room_context": "Language is treated as a prediction problem. From Shannon’s statistical sequences, the story leads to Transformers and large language models.", "catalogue": "5.3", "title": "The Transformer", "year": "2017", "description": "Google researchers publish “Attention Is All You Need”. The attention mechanism allows the model to relate each token to other tokens in the sequence.", "image": "img/5-3.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary conference paper", "source": "Vaswani, A. et al. (2017), “Attention Is All You Need”, NeurIPS.", "supports": "The Transformer architecture and its use of self-attention.", "status": "Primary source.", "url": "https://arxiv.org/abs/1706.03762" }, "slug": "transformer", "url": "/en/exhibits/transformer/" }, { "room_id": "room-5", "room_title": "Language and the first LLMs", "room_context": "Language is treated as a prediction problem. From Shannon’s statistical sequences, the story leads to Transformers and large language models.", "catalogue": "5.4", "title": "GPT and BERT", "year": "2018", "description": "OpenAI and Google show that a model first pretrained on a very large body of text can later be adapted to many different tasks.", "image": "img/5-4.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary research papers", "source": "Radford et al. (2018), Generative Pre-Training; Devlin et al. (2018), BERT.", "supports": "The generalisation of Transformer pretraining followed by adaptation to many downstream tasks.", "status": "Primary sources.", "url": "https://openai.com/index/language-unsupervised/" }, "slug": "gpt-and-bert", "url": "/en/exhibits/gpt-and-bert/" }, { "room_id": "room-5", "room_title": "Language and the first LLMs", "room_context": "Language is treated as a prediction problem. From Shannon’s statistical sequences, the story leads to Transformers and large language models.", "catalogue": "5.5", "title": "GPT-3", "year": "2020", "description": "With 175 billion parameters, the model can write text, translate and answer questions, often from only a few examples in the prompt.", "image": "img/5-5.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary conference paper", "source": "Brown, T. et al. (2020), “Language Models are Few-Shot Learners”, NeurIPS 33.", "supports": "GPT-3’s 175 billion parameters and its few-shot behaviour.", "status": "Primary source.", "url": "https://papers.neurips.cc/paper/2020/hash/1457c0d6bfcb4967418bfb8ac142f64a-Abstract.html" }, "slug": "gpt-3", "url": "/en/exhibits/gpt-3/" }, { "room_id": "room-5", "room_title": "Language and the first LLMs", "room_context": "Language is treated as a prediction problem. From Shannon’s statistical sequences, the story leads to Transformers and large language models.", "catalogue": "5.6", "title": "ChatGPT", "year": "2022", "description": "On 30 November 2022, a conversational language-model interface is released publicly as a research preview. Generative AI begins to enter everyday life at unprecedented scale.", "image": "img/5-6.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Official product announcement", "source": "OpenAI, “Introducing ChatGPT”, 30 November 2022.", "supports": "The public release date of ChatGPT as a research preview and its conversational format.", "status": "Official primary source.", "url": "https://openai.com/index/chatgpt/" }, "slug": "chatgpt", "url": "/en/exhibits/chatgpt/" }, { "room_id": "room-6", "room_title": "How an LLM is made today", "room_context": "A modern language model does not appear fully formed. It passes through data collection, training, adaptation, evaluation and, finally, answer generation.", "catalogue": "6.1", "title": "Data collection", "year": "1", "description": "Huge quantities of text are gathered from sources such as websites, books and code. Selection and cleaning influence what a model can learn, how well it performs and which biases it may reproduce.", "image": "img/6-1.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Curatorial synthesis of technical literature", "source": "Technical literature on pretraining, data curation and scaling of modern LLMs.", "supports": "That data selection and curation influence model knowledge, quality and bias.", "status": "General process; details vary by model.", "url": "" }, "slug": "data-collection", "url": "/en/exhibits/data-collection/" }, { "room_id": "room-6", "room_title": "How an LLM is made today", "room_context": "A modern language model does not appear fully formed. It passes through data collection, training, adaptation, evaluation and, finally, answer generation.", "catalogue": "6.2", "title": "Tokenization", "year": "2", "description": "Text is split into smaller units called tokens, which may be whole words or pieces of words. The model processes numerical identifiers rather than written characters directly.", "image": "img/6-2.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary research paper", "source": "Sennrich, Haddow & Birch (2016), ACL, subword units.", "supports": "The use of subword units such as BPE to represent rare or previously unseen words.", "status": "Primary source; exact tokenizers differ across models.", "url": "https://aclanthology.org/P16-1162/" }, "slug": "tokenization", "url": "/en/exhibits/tokenization/" }, { "room_id": "room-6", "room_title": "How an LLM is made today", "room_context": "A modern language model does not appear fully formed. It passes through data collection, training, adaptation, evaluation and, finally, answer generation.", "catalogue": "6.3", "title": "Pretraining", "year": "3", "description": "The model repeatedly learns to predict the next token across enormous text corpora and large amounts of computing. This stage gives it broad language ability and general patterns, but not necessarily assistant-like behaviour.", "image": "img/6-3.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Curatorial synthesis of technical literature", "source": "Research literature on autoregressive language modelling and large-scale pretraining.", "supports": "Training with next-token prediction over very large text corpora.", "status": "Generalised description; not every LLM uses exactly the same training procedure.", "url": "" }, "slug": "pretraining", "url": "/en/exhibits/pretraining/" }, { "room_id": "room-6", "room_title": "How an LLM is made today", "room_context": "A modern language model does not appear fully formed. It passes through data collection, training, adaptation, evaluation and, finally, answer generation.", "catalogue": "6.4", "title": "Instruction tuning", "year": "4", "description": "With examples of instructions and high-quality answers, the model is trained to follow requests and behave more like an assistant.", "image": "img/6-4.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary research paper", "source": "Ouyang, L. et al. (2022), InstructGPT.", "supports": "Supervised instruction tuning with human-written demonstrations before RLHF.", "status": "Primary source for one important family of techniques.", "url": "https://arxiv.org/abs/2203.02155" }, "slug": "instruction-tuning", "url": "/en/exhibits/instruction-tuning/" }, { "room_id": "room-6", "room_title": "How an LLM is made today", "room_context": "A modern language model does not appear fully formed. It passes through data collection, training, adaptation, evaluation and, finally, answer generation.", "catalogue": "6.5", "title": "Alignment", "year": "5", "description": "People, and sometimes other models, evaluate answers. Training then pushes the model toward responses judged more helpful, honest or safe. A well-known approach is reinforcement learning from human feedback (RLHF).", "image": "img/6-5.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Primary research paper", "source": "Ouyang, L. et al. (2022), “Training language models to follow instructions with human feedback”.", "supports": "The use of human preferences and reinforcement learning to shape model behaviour.", "status": "Primary source; modern systems also use methods beyond RLHF.", "url": "https://arxiv.org/abs/2203.02155" }, "slug": "alignment", "url": "/en/exhibits/alignment/" }, { "room_id": "room-6", "room_title": "How an LLM is made today", "room_context": "A modern language model does not appear fully formed. It passes through data collection, training, adaptation, evaluation and, finally, answer generation.", "catalogue": "6.6", "title": "Testing and safety", "year": "6", "description": "Before release, models are evaluated for errors, harmful behaviour and misuse. Testing and monitoring can continue after deployment.", "image": "img/6-6.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Curatorial synthesis of safety practice", "source": "System cards, evaluations and technical literature on model safety and red teaming.", "supports": "The need to evaluate model behaviour and risks before and after release.", "status": "Processes differ across organisations and models.", "url": "" }, "slug": "testing-and-safety", "url": "/en/exhibits/testing-and-safety/" }, { "room_id": "room-6", "room_title": "How an LLM is made today", "room_context": "A modern language model does not appear fully formed. It passes through data collection, training, adaptation, evaluation and, finally, answer generation.", "catalogue": "6.7", "title": "Use", "year": "7", "description": "When you type a prompt, a basic language model does not retrieve a ready-made answer from a database. It generates a response token by token, although some modern systems can also call search engines, tools or external knowledge sources.", "image": "img/6-7.svg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Technical operating principle", "source": "Autoregressive decoding in the language-model literature.", "supports": "That generation proceeds token by token unless the surrounding system also uses external tools or retrieval.", "status": "General principle with an important qualification for tool-using and RAG systems.", "url": "" }, "slug": "use", "url": "/en/exhibits/use/" }, { "room_id": "room-7", "room_title": "Society and ethics", "room_context": "When AI leaves the laboratory, its consequences affect work, rights, information, privacy and accountability.", "catalogue": "7.1", "title": "Deepfakes", "year": "2017", "description": "The term becomes associated with synthetic videos in which one person’s face is replaced or generated using AI. Today voice, image and video can all be fabricated convincingly, with consequences for fraud, elections and public information.", "image": "img/7-1.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Scientific and legal analysis", "source": "Chesney, R. & Citron, D. (2019), California Law Review 107.", "supports": "Risks of deepfakes for privacy, democracy and security.", "status": "Secondary scholarly source.", "url": "https://www.californialawreview.org/print/deep-fakes-a-looming-challenge-for-privacy-democracy-and-national-security" }, "slug": "deepfakes", "url": "/en/exhibits/deepfakes/" }, { "room_id": "room-7", "room_title": "Society and ethics", "room_context": "When AI leaves the laboratory, its consequences affect work, rights, information, privacy and accountability.", "catalogue": "7.2", "title": "Work and creativity", "year": "2023", "description": "The 2023 Hollywood writers’ strike includes negotiated rules on the use of AI. The broader question remains: which jobs change, which disappear, and which new roles emerge?", "image": "img/7-2.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Collective agreement and official union source", "source": "Writers Guild of America (2023), Summary of the 2023 WGA MBA.", "supports": "The terms agreed for the use of AI in writing work and literary material.", "status": "Primary labour and institutional document.", "url": "https://www.wgacontract2023.org/the-campaign/summary-of-the-2023-wga-mba" }, "slug": "work-and-creativity", "url": "/en/exhibits/work-and-creativity/" }, { "room_id": "room-7", "room_title": "Society and ethics", "room_context": "When AI leaves the laboratory, its consequences affect work, rights, information, privacy and accountability.", "catalogue": "7.3", "title": "Copyright", "year": "2023", "description": "Authors, artists and news organisations, including The New York Times, take disputes over the use of copyrighted works in model training to court. Many legal questions remain unsettled.", "image": "img/7-3.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Court filings and evolving case law", "source": "The New York Times Co. v. Microsoft Corp. & OpenAI and related cases.", "supports": "That disputes over model training and copyright have led to litigation.", "status": "Cases are still evolving; unresolved questions are not presented as settled law.", "url": "" }, "slug": "copyright", "url": "/en/exhibits/copyright/" }, { "room_id": "room-7", "room_title": "Society and ethics", "room_context": "When AI leaves the laboratory, its consequences affect work, rights, information, privacy and accountability.", "catalogue": "7.4", "title": "The EU AI Act", "year": "2024", "description": "The European Union adopts the first comprehensive horizontal law on artificial intelligence, introduced in stages. It bans certain practices and imposes stricter duties on high-risk uses such as some employment and education systems.", "image": "img/7-4.jpg", "image_kind": "AI-generated illustration", "credit": "", "documentation": { "type": "Official legal text", "source": "Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence.", "supports": "The European regulatory framework covering prohibited practices, high-risk systems and transparency duties.", "status": "Binding primary legal source with phased application.", "url": "https://eur-lex.europa.eu/eli/reg/2024/1689/oj" }, "slug": "eu-ai-act", "url": "/en/exhibits/eu-ai-act/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.1", "title": "EUROTRA: machines translate Greek", "year": "1985", "description": "A small Greek team of linguists and computer scientists joins the European EUROTRA project to build machine translation between Community languages. This work becomes a starting point for organised Greek language technology.", "image": "img/8-1.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "ILSP / Athena Research Center, History.", "supports": "Ότι το 1985 ελληνική ομάδα εργάστηκε στο EUROTRA και ότι από αυτή την πορεία προέκυψε αργότερα το ΙΕΛ.", "status": "Source reviewed for this exhibit", "url": "https://www.athenarc.gr/el/ilsp/history" }, "slug": "eurotra-greek-machine-translation", "url": "/en/exhibits/eurotra-greek-machine-translation/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.2", "title": "SKEL: a Greek AI laboratory", "year": "1989", "description": "SKEL is founded at NCSR Demokritos and grows into one of Greece’s largest AI research labs, working on intelligent information systems, data analysis and human-computer interaction.", "image": "img/8-2.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "NCSR “Demokritos”, SKEL | The AI Lab.", "supports": "Την ίδρυση του SKEL το 1989 και το ερευνητικό του αντικείμενο.", "status": "Source reviewed for this exhibit", "url": "https://www.iit.demokritos.gr/labs/skel/" }, "slug": "skel-greek-ai-lab", "url": "/en/exhibits/skel-greek-ai-lab/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.3", "title": "The Institute for Language and Speech Processing", "year": "1991", "description": "The EUROTRA team gains a permanent institutional home with the creation of ILSP. Its work expands from translation to speech, language resources, culture and human-machine interaction.", "image": "img/8-3.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "ILSP / Athena Research Center, History.", "supports": "Την ίδρυση του ΙΕΛ το 1991 και τη συνέχεια της ομάδας του EUROTRA.", "status": "Source reviewed for this exhibit", "url": "https://www.athenarc.gr/el/ilsp/history" }, "slug": "institute-language-speech-processing", "url": "/en/exhibits/institute-language-speech-processing/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.4", "title": "GreekBERT", "year": "2020", "description": "Researchers release a BERT model trained specifically on Greek text, showing why a language with far fewer digital resources than English benefits from dedicated models and evaluation.", "image": "img/8-4.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Koutsikakis et al. (2020), GreekBERT.", "supports": "Την ανάπτυξη μονογλωσσικού BERT για τα ελληνικά και την αξιολόγησή του σε εργασίες ελληνικού NLP.", "status": "Source reviewed for this exhibit", "url": "https://arxiv.org/abs/2008.12014" }, "slug": "greekbert", "url": "/en/exhibits/greekbert/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.5", "title": "The Archimedes Unit", "year": "2022", "description": "Athena Research Center establishes a unit for basic and applied research in AI, data science and algorithms, linking researchers in Greece and abroad.", "image": "img/8-5.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Athena Research Center, Archimedes Unit.", "supports": "Την ίδρυση της μονάδας τον Ιανουάριο του 2022 και την αποστολή της.", "status": "Source reviewed for this exhibit", "url": "https://www.athenarc.gr/el/archimedes/" }, "slug": "archimedes-unit", "url": "/en/exhibits/archimedes-unit/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.6", "title": "mAigov: generative AI in public services", "year": "2023", "description": "The gov.gr digital assistant brings generative AI into a public service used by citizens at national scale.", "image": "img/8-6.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Ministry of Digital Governance / DigiGov, mAigov.", "supports": "Τη λειτουργία του mAigov στο gov.gr από τον Δεκέμβριο του 2023.", "status": "Source reviewed for this exhibit", "url": "https://digi.gov.gr/o-protos-psifiakos-voithos-maigov/" }, "slug": "maigov", "url": "/en/exhibits/maigov/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.7", "title": "Meltemi: an open LLM for Greek", "year": "2024", "description": "ILSP introduces Meltemi, an open large language model designed for Greek, highlighting the challenge of languages with fewer digital resources.", "image": "img/8-7.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Athena Research Center, Meltemi.", "supports": "Την παρουσίαση του Meltemi ως πρώτου ανοικτού μεγάλου γλωσσικού μοντέλου για τα ελληνικά.", "status": "Source reviewed for this exhibit", "url": "https://www.athenarc.gr/el/news/meltemi-proto-anoihto-megalo-glossiko-montelo-gia-ta-ellinika" }, "slug": "meltemi-greek-llm", "url": "/en/exhibits/meltemi-greek-llm/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.8", "title": "Krikri: the next Greek LLM", "year": "2025", "description": "Llama-Krikri extends the effort with more Greek training data, a larger context window and broader capabilities, while remaining openly available.", "image": "img/8-8.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Athena Research Center, Llama-Krikri.", "supports": "Την παρουσίαση του Krikri ως νεότερου ανοικτού ελληνικού γλωσσικού μοντέλου.", "status": "Source reviewed for this exhibit", "url": "https://www.athenarc.gr/en/node/6573" }, "slug": "krikri", "url": "/en/exhibits/krikri/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.9", "title": "PHAROS: the Greek AI Factory", "year": "2025–26", "description": "Greece joins the European AI Factories network with PHAROS, focusing among other areas on Greek language and culture, health and sustainability.", "image": "img/8-9.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "EuroHPC Joint Undertaking, Greece AI Factory / PHAROS.", "supports": "Την ένταξη του PHAROS στο δίκτυο AI Factories και τους βασικούς τομείς εστίασής του.", "status": "Source reviewed for this exhibit", "url": "https://www.eurohpc-ju.europa.eu/ai-factories/greece_en" }, "slug": "pharos-ai-factory", "url": "/en/exhibits/pharos-ai-factory/" }, { "room_id": "room-8", "room_title": "Artificial intelligence in Greece", "room_context": "Modern Greek AI runs from language technology and research to open Greek models and national computing infrastructure.", "catalogue": "8.10", "title": "DAEDALUS: computing power in Greece", "year": "2026", "description": "The DAEDALUS supercomputer brings new national computing capacity for science and AI. In the June 2026 TOP500 it ranks 31st with measured performance of 85.69 petaflops.", "image": "img/8-10.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "GRNET, DAEDALUS.", "supports": "Τη μετρημένη επίδοση και την κατάταξη του DAEDALUS στο TOP500 του Ιουνίου 2026.", "status": "Source reviewed for this exhibit", "url": "https://grnet.gr/en/business-directory/grant-for-the-development-of-a-new-national-hpc-system-daedalus/" }, "slug": "daedalus-supercomputer", "url": "/en/exhibits/daedalus-supercomputer/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.1", "title": "From chatbot to agent", "year": "2026 →", "description": "AI systems are beginning to use computers and tools to complete sequences of actions. On OSWorld, agent performance rose from roughly 12% to 66.3% in a year, yet agents still fail about one in three structured attempts.", "image": "img/9-1.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Stanford HAI, AI Index Report 2026, Technical Performance.", "supports": "Τη γρήγορη βελτίωση των agents σε πραγματικές εργασίες υπολογιστή και τα σημερινά όριά τους.", "status": "Source reviewed for this exhibit", "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report/technical-performance" }, "slug": "from-chatbot-to-agent", "url": "/en/exhibits/from-chatbot-to-agent/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.2", "title": "AI gets a body", "year": "2026 →", "description": "Robotics is increasingly connected with vision-language models. Yet the gap between a controlled lab and an unpredictable home remains large: the 2026 AI Index reports only 12% success on real household tasks.", "image": "img/9-2.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Stanford HAI, AI Index Report 2026, Technical Performance.", "supports": "Την πρόοδο της ρομποτικής και το μεγάλο χάσμα ανάμεσα σε προσομοίωση/εργαστήριο και πραγματικό σπίτι.", "status": "Source reviewed for this exhibit", "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report/technical-performance" }, "slug": "ai-gets-a-body", "url": "/en/exhibits/ai-gets-a-body/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.3", "title": "Laboratories that choose the next experiment", "year": "2026 →", "description": "Self-driving laboratories combine AI, robotics and automation: a system proposes an experiment, runs it and uses the result to choose the next step. Today they mostly operate in narrow, well-defined domains.", "image": "img/9-3.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Canty & Abolhasani (2026), Nature Reviews Chemistry.", "supports": "Την εξέλιξη των self-driving laboratories από στενή αυτοματοποίηση προς πλατφόρμες ανακάλυψης.", "status": "Source reviewed for this exhibit", "url": "https://doi.org/10.1038/s41570-026-00847-2" }, "slug": "self-driving-labs", "url": "/en/exhibits/self-driving-labs/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.4", "title": "Medicine and digital models of patients", "year": "2026 →", "description": "AI is already used in diagnosis, drug discovery and research on medical digital twins. Early results are promising, but clinical reliability and generalisation require rigorous evaluation.", "image": "img/9-4.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Stanford HAI, AI Index Report 2026, Medicine.", "supports": "Την ανάπτυξη της ιατρικής ΤΝ και το αυξανόμενο ερευνητικό ενδιαφέρον για digital twins.", "status": "Source reviewed for this exhibit", "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report/medicine" }, "slug": "ai-medicine-digital-twins", "url": "/en/exhibits/ai-medicine-digital-twins/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.5", "title": "The classroom with personal AI", "year": "2026 →", "description": "Generative AI is already widespread in study and education. The question is shifting from whether students will use it to what humans should learn when machines can write, solve and explain.", "image": "img/9-5.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Stanford HAI, AI Index Report 2026, Education.", "supports": "Την ευρεία χρήση generative AI από φοιτητές και τη σταδιακή ένταξη AI education σε εθνικά προγράμματα.", "status": "Source reviewed for this exhibit", "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report/education" }, "slug": "classroom-personal-ai", "url": "/en/exhibits/classroom-personal-ai/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.6", "title": "Work changes tasks before it changes occupations", "year": "2025 →", "description": "The ILO estimates that one in four workers is in an occupation with some GenAI exposure. In the near term, task transformation is more likely than the automatic disappearance of whole occupations.", "image": "img/9-6.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "ILO (2025), Generative AI and Jobs: A 2025 update.", "supports": "Την παγκόσμια έκθεση επαγγελμάτων στη generative AI και την εκτίμηση ότι ο μετασχηματισμός είναι πιθανότερος από την πλήρη αντικατάσταση.", "status": "Source reviewed for this exhibit", "url": "https://www.ilo.org/publications/generative-ai-and-jobs-2025-update" }, "slug": "work-and-tasks", "url": "/en/exhibits/work-and-tasks/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.7", "title": "Smaller, local and specialised models", "year": "2026 →", "description": "The future is not only about ever larger models. Smaller systems can run locally, specialise in specific domains and offer lower cost or better privacy.", "image": "img/9-7.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "Stanford HAI, AI Index Report 2026 and contemporary literature on edge and domain-specific AI.", "supports": "Τη μετατόπιση μέρους της έρευνας προς αποδοτικότερα και εξειδικευμένα μοντέλα.", "status": "Source reviewed for this exhibit", "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report" }, "slug": "smaller-local-models", "url": "/en/exhibits/smaller-local-models/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.8", "title": "The invisible factory behind AI", "year": "2030", "description": "AI depends on physical infrastructure: chips, data centres, cooling and electricity. In the IEA base case, global data-centre electricity use roughly doubles to around 945 TWh by 2030.", "image": "img/9-8.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "International Energy Agency, Energy and AI.", "supports": "Το βασικό σενάριο για την αύξηση της κατανάλωσης ηλεκτρισμού των data centres έως το 2030.", "status": "Source reviewed for this exhibit", "url": "https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai" }, "slug": "invisible-ai-factory", "url": "/en/exhibits/invisible-ai-factory/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.9", "title": "More AI, more rules", "year": "2026 →", "description": "As AI enters higher-stakes decisions, requirements for transparency, evaluation, provenance and accountability grow. In Europe the AI Act is being phased in over several years.", "image": "img/9-9.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "European Commission, AI Act implementation timeline.", "supports": "Τη σταδιακή εφαρμογή του AI Act και το ότι οι υποχρεώσεις εισάγονται σε διαφορετικές ημερομηνίες.", "status": "Source reviewed for this exhibit", "url": "https://ai-act-service-desk.ec.europa.eu/en/ai-act/eu-ai-act-implementation-timeline" }, "slug": "more-ai-more-rules", "url": "/en/exhibits/more-ai-more-rules/" }, { "room_id": "room-9", "room_title": "The future of artificial intelligence", "room_context": "This room separates what is already happening, strong trends and genuinely uncertain futures.", "catalogue": "9.10", "title": "AGI: the great uncertainty", "year": "Unknown", "description": "There is no scientific consensus on whether or when artificial general intelligence will exist. Current systems show increasingly autonomous capabilities, but that does not establish an inevitable path to AGI or loss of control.", "image": "img/9-10.svg", "image_kind": "Curatorial illustration", "credit": "Museum of Artificial Intelligence", "documentation": { "type": "Institutional / scientific source", "source": "International AI Safety Report 2026.", "supports": "Την ύπαρξη πρώιμων πιο αυτόνομων ικανοτήτων και τη σημαντική αβεβαιότητα για μελλοντική απώλεια ελέγχου.", "status": "Source reviewed for this exhibit", "url": "https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026" }, "slug": "agi-great-uncertainty", "url": "/en/exhibits/agi-great-uncertainty/" } ] }