- Alan Turing defined computers mathematically in 1936 and proposed the Turing Test in 1950 as a measure of machine intelligence.
- John McCarthy coined the term 'artificial intelligence' in 1955 and invented LISP programming language in 1958.
- Marvin Minsky's 1969 book Perceptrons mathematically proved limitations of single-layer networks, contributing to the first AI winter.
- Yann LeCun developed CNNs at Bell Labs in the late 1980s, enabling the first major commercial neural network deployment for zip code recognition.
- LeCun shared the 2018 Turing Award with Geoffrey Hinton and Yoshua Bengio for their contributions to deep learning.
Alan Turing: The Foundation
Alan Turing (1912 to 1954) is the founding figure of computer science and artificial intelligence. His 1936 paper 'On Computable Numbers, with an Application to the Entscheidungsproblem' defined what a computer is, in abstract mathematical terms, before any electronic computer had been built. His 1950 paper 'Computing Machinery and Intelligence' asked whether machines could think and proposed the imitation game, which we now call the Turing Test, as a way to evaluate the question.
During World War II, Turing led the effort at Bletchley Park to break the German Enigma cipher, work that is estimated to have shortened the war by two to four years. He developed the Bombe, a machine that automated the search for Enigma settings, which processed messages in volumes no human team could have managed. The work was classified for decades, meaning Turing received no public credit during his lifetime.
After the war, Turing worked on the design of early computers at the National Physical Laboratory and then at the University of Manchester. He wrote programs for the Manchester Mark 1, one of the world's first stored-program computers. He proposed the idea of a 'child machine' that would start with limited capabilities and learn to improve through experience, anticipating the field of machine learning by decades.
Turing was prosecuted for homosexuality in 1952, then a criminal offense in Britain. He was chemically castrated as an alternative to imprisonment. He died in June 1954, apparently from cyanide poisoning. The inquest recorded his death as suicide. He was 41 years old. In 2013, Queen Elizabeth II granted Turing a posthumous royal pardon. In 2021, his face appeared on the British 50 pound note.
John McCarthy: The Man Who Named AI
John McCarthy (1927 to 2011) coined the term 'artificial intelligence' in 1955, when he wrote the proposal for the Dartmouth Conference, the workshop that launched AI as a field. He was 27 years old. The term was deliberately chosen over alternatives like 'machine intelligence' or 'automata studies' because McCarthy wanted to distinguish the enterprise from cybernetics and signal processing. The name stuck.
McCarthy's technical contributions were foundational. He invented LISP in 1958, a programming language based on lambda calculus that became the dominant language for AI programming for three decades. It introduced many concepts that are now standard in programming: garbage collection, recursion as a fundamental construct, and treating code as data. LISP was the language of AI research until the 1980s and its influence persists in modern functional programming languages.
He developed the concept of time-sharing, the idea that a single computer could serve multiple users simultaneously by rapidly switching between their tasks. This was the conceptual precursor of modern operating systems. He also coined the term 'computer science' as a distinct academic discipline, separate from mathematics and electrical engineering.
McCarthy was skeptical of many AI claims, including his own earlier optimism. He became a vocal critic of overly broad assertions about what AI systems could do. He believed that genuine AI required formal representation of knowledge and logical reasoning. He remained committed to the symbolic AI tradition even as neural networks overtook it. He died in 2011, just as the deep learning revolution he had not believed in was beginning to transform the field he had named.
Marvin Minsky: The Skeptic Inside the Field
Marvin Minsky (1927 to 2016) was one of the founding fathers of AI and its most sharp-tongued critic. He co-founded the MIT Artificial Intelligence Laboratory in 1959 with John McCarthy and led it for decades. He believed that intelligence could be understood and replicated by understanding the computational processes of the brain. He also believed that most approaches people were pursuing to build AI were fundamentally misguided.
His 1969 book Perceptrons, co-authored with Seymour Papert, was the most influential critical text in the history of AI. The book proved mathematically that single-layer perceptrons could not compute certain functions, including XOR. It implied that multilayer networks would not overcome these limitations. The book was technically correct but strategically catastrophic: it was widely interpreted as demonstrating that neural networks as a whole were dead ends, and it contributed significantly to the first AI winter.
Minsky's later work focused on the Society of Mind, a theory of how intelligence arises from the interaction of many simple, non-intelligent agents. He argued that there is no single seat of intelligence, no central processor that integrates all thought. Instead, intelligence emerges from the collective operation of specialized modules that interact in complex ways. This framework anticipated some aspects of how modern AI systems are understood, even if its specific details have not been directly implemented.
Minsky was deeply skeptical of AI safety concerns in the way they came to be framed. He did not believe that superintelligent machines were a near-term risk. He believed the challenge was getting machines to be intelligent at all. He was critical of researchers who spent time on safety rather than capability. He died in January 2016, just as both superintelligent AI safety concerns and deep learning capabilities were simultaneously gaining mainstream attention.
Yann LeCun: Convolutional Networks and the Meta Vision
Yann LeCun (born 1960) is Chief AI Scientist at Meta and a professor at New York University. He is one of three researchers, along with Geoffrey Hinton and Yoshua Bengio, who shared the 2018 Turing Award for their contributions to deep learning. His specific contribution is the development of convolutional neural networks (CNNs), the architecture that transformed computer vision.
LeCun developed CNNs at Bell Labs in the late 1980s and early 1990s, applying backpropagationBackpropagationThe algorithm that computes loss gradients for every parameter by applying the chain rule backward through the network.Learn more → to train them. His 1989 paper and the famous 1998 LeNet-5 paper demonstrated that CNNs could reliably recognize handwritten digits. AT&T and the US Postal Service deployed his system to automatically read zip codes from mail, processing millions of pieces per day. This was one of the first major commercial deployments of a neural network.
His theoretical insight was that visual data has spatial structure that can be exploited by a model designed to process it: nearby pixels are more related than distant ones, and the same visual pattern can appear anywhere in an image. Convolutional networks encode these assumptions into their architecture through weight sharing and local connectivity, dramatically reducing the number of parametersParametersThe numerical weights inside a neural network that are learned during training — the 'knowledge' of the model, measured in billions for modern LLMs.Learn more → required and improving generalization.
At Meta, LeCun has been an outspoken advocate for open-source AI and a skeptic of claims that current AI systems are close to human-level intelligence. He argues that large language models, despite their impressive capabilities, fundamentally lack the ability to understand the physical world, make reliable plans, and engage in true reasoning. He believes that a different approach, one based on learning world models and self-supervised learningSelf-Supervised LearningA machine learning approach where models learn from unlabeled data by predicting missing parts of the input, such as the next word in a sentence.Learn more → from sensory data, will be needed to achieve artificial general intelligence. These views put him at odds with much of the AI hype cycle, a position he occupies with characteristic directness.
Yoshua Bengio: The Conscience of Deep Learning
Yoshua Bengio (born 1964) is a professor at the Universite de Montreal and scientific director of Mila, the Quebec AI Institute. He shared the 2018 Turing Award with Hinton and LeCun. His contributions to deep learning span several decades and cover some of the most important technical advances in the field: training techniques for deep networks, attention mechanisms, and generative models.
Bengio's 2003 paper introducing neural language models was an early demonstration that neural networks could learn word representations that captured semantic relationships. The word2vec embeddings that became standard in NLP in the 2010s built directly on this work. His lab's work on attention mechanisms, particularly the Bahdanau attention mechanismAttention MechanismA technique that allows each token in a sequence to 'pay attention' to all other tokens, enabling the model to understand context and relationships across long distances.Learn more → introduced in 2015, was a direct precursor to the transformerTransformerThe neural network architecture that underpins virtually all modern LLMs, introduced in 2017, built around self-attention mechanisms that process entire sequences in parallel.Learn more → architecture.
Unlike some of his peers, Bengio has become increasingly public about AI safety concerns. He signed the open letter in 2023 calling for a pause on training AI systems more powerful than GPTGPTGenerative Pre-trained Transformer — the model architecture and family name behind OpenAI's most famous models, from GPT-2 to GPT-5.Learn more →-4. He has testified before governments and international bodies about AI risks. He argues that the pace of AI development has outrun our understanding of the systems being built and our ability to ensure they behave safely.
Bengio occupies an unusual position in the AI discourse. He is one of the founding scientists of deep learning, someone who has contributed directly to the capabilities that are generating concern. His advocacy for safety and slower development is therefore not the voice of an outsider skeptic. It is the voice of someone who understands the technology deeply and has concluded that the current trajectory is concerning. This gives his concerns a weight that purely philosophical AI safety arguments do not have.
Ilya Sutskever: From OpenAI to Safety
Ilya Sutskever (born 1986) was a PhD student of Geoffrey Hinton's at the University of Toronto and one of the co-authors of the AlexNet paper. After the AlexNet triumph, he co-founded OpenAI with Sam Altman, Elon Musk, Greg Brockman, and others in 2015, and served as Chief Scientist until 2024. He is widely regarded as one of the most talented AI researchers of his generation.
At OpenAI, Sutskever led the technical direction that produced GPT-1, GPT-2, GPT-3, ChatGPT, and GPT-4. He advocated for scaling as the primary driver of capability improvement, a bet that proved spectacularly correct. He was also one of the founders of the alignmentAlignmentThe process of ensuring AI models behave according to human values and intentions through training techniques, evaluation methods, and safety measures.Learn more → team at OpenAI, recognizing early that the systems they were building raised genuine safety questions.
Sutskever was one of the board members who voted to fire Sam Altman in November 2023. He subsequently signed the letter calling for Altman's reinstatement, a reversal that he has not publicly explained in detail. The episode remains one of the most mysterious in recent AI history. Sutskever left OpenAI in May 2024, having been with the company since its founding.
In June 2024, Sutskever announced the founding of Safe Superintelligence Inc. (SSI), a new AI safety company co-founded with Daniel Gross and Daniel Levy. The company's stated mission is to build safe superintelligence as its sole focus, without the commercial pressures that Sutskever appeared to believe were affecting alignment-focused work at OpenAI. SSI has raised substantial funding but operates in unusual secrecy, consistent with a focus on long-term research over near-term product development.
Demis Hassabis: DeepMind and the Protein Folding Prize
Demis Hassabis (born 1976) is the co-founder and CEO of DeepMind, now Google DeepMind. He is one of the most versatile figures in AI: a chess prodigy who reached master level at age 13, a game developer who co-created the Theme Park simulation game at age 17, a neuroscientist who earned a PhD from University College London studying how the brain creates memories, and an AI researcher who has led some of the most significant AI achievements of the past decade.
DeepMind, founded in London in 2010 and acquired by Google in 2014, has produced a remarkable series of breakthroughs. AlphaGo defeated world Go champion Lee Sedol in 2016, a milestone that arrived a decade earlier than many experts had predicted. AlphaZero mastered chess, shogi, and Go simultaneously from scratch in 2018. AlphaStar defeated professional StarCraft II players in 2019.
The most consequential achievement was AlphaFold 2, released in 2020. Protein structure prediction, determining the three-dimensional shape of a protein from its amino acid sequence, had been one of biology's central unsolved problems for fifty years. AlphaFold 2 solved it. Its predictions are accurate enough to be used directly in drug discovery, structural biology research, and vaccine development. More than 200 million protein structures have been predicted and made publicly available. The scientific impact is difficult to overstate.
Hassabis received the Nobel Prize in Chemistry in 2024, shared with John Jumper (who led the AlphaFold team) and David Baker (for his work on protein design). He is one of very few people to have received a Nobel Prize in a scientific field based on the application of AI. He represents the vision of AI as a tool for scientific discovery, not just a commercial product, and has been an articulate advocate for responsible development as DeepMind's capabilities have grown.
Andrej Karpathy: Teacher to the Field
Andrej Karpathy (born 1986) is unusual among prominent AI researchers in that his most influential contributions may be pedagogical as much as technical. He has an extraordinary ability to explain complex technical ideas clearly, and he has used this ability to help an entire generation of engineers understand deep learning from first principles.
Karpathy did his PhD at Stanford under Fei-Fei Li, working on convolutional networks for image captioning. He joined OpenAI in 2015 as a founding member and researcher, working on reinforcement learning and robotics. He then joined Tesla in 2017 as Director of AI, leading the Autopilot vision team that was responsible for developing the neural networks used in Tesla's driver assistance system. His approach to autopilot, building a system based entirely on cameras and neural networks rather than radar and lidar, was distinctive and controversial.
Karpathy returned to OpenAI in 2023, then left in 2024. Since then, he has focused largely on educational content. His YouTube series, including 'The spelled-out intro to neural networks and backpropagation: building micrograd' and 'Let's build GPT: from scratch, in code, spelled out,' have become definitive educational resources. His GitHub repository nanoGPT, a clean, minimal implementation of a GPT model, has been studied by thousands of engineers learning to build language models from scratch.
His impact as a teacher is reflected in how many people in the AI field cite his courses and YouTube videos as formative. He has a rare combination of deep technical expertise and genuine care about communication. At a time when AI is becoming important for an increasingly wide range of people to understand, the ability to explain it clearly is itself a significant contribution.
Fei-Fei Li: ImageNet and the Data Revolution
Fei-Fei Li (born 1976) is a professor at Stanford University and the co-director of the Stanford Human-Centered AI Institute. She is best known as the creator of ImageNet, the massive labeled image database that became the training ground for the computer vision revolution and the proving ground for AlexNet in 2012.
ImageNet was not a technical architecture or an algorithm. It was a dataset: 14 million images, labeled with one of more than 22,000 categories from WordNet, the semantic network of English words. Building it required years of work and an innovative use of Amazon Mechanical Turk to crowdsource the labeling task. At the time it was created, it was the largest labeled visual dataset in existence. Its scale proved decisive. The models trained on ImageNet learned visual representations that were qualitatively more powerful than anything trained on smaller datasets.
Li served as Chief Scientist of AI and Machine Learning at Google Cloud from 2017 to 2018, on leave from Stanford, where she helped Google develop its enterprise AI offerings and advocates for responsible AI development. She has been a consistent voice for keeping human considerations at the center of AI research and deployment, arguing that technical capability must be coupled with attention to fairness, accountability, and the needs of diverse users.
She is also a prominent voice for increasing diversity in AI. AI systems trained predominantly on data from and by certain demographic groups will systematically underperform for other groups. The lack of diversity among AI researchers and engineers contributes to blind spots in how systems are designed and deployed. Li has been an advocate for structural changes in how universities recruit and support students from underrepresented groups in AI and computer science.
Andrew Ng: Democratizing Machine Learning
Andrew Ng (born 1976) is one of the most influential figures in making machine learning accessible beyond research labs and major technology companies. He co-founded Google Brain, the deep learning research team at Google, in 2011. He served as Chief Scientist at Baidu from 2014 to 2017, where he led one of the largest applied AI research teams in China. And he co-founded Coursera and deeplearning.ai, which have brought machine learning education to millions of people worldwide.
The Coursera Machine Learning course that Ng launched in 2011 became one of the most successful online courses ever created. By some estimates, more than four million people have taken it. The course translated academic machine learning into a form accessible to working engineers and programmers without graduate training in the subject. It played a significant role in creating the large pool of AI-capable engineers that the tech industry required as AI applications proliferated.
Ng's deeplearning.ai specializations, launched after his departure from Baidu, have extended this mission. The deep learning specialization has been taken by millions of students. It covers neural networks, convolutional networks, sequence models, and natural language processing in a pedagogical sequence that takes someone with programming experience but limited mathematics background to a functional understanding of modern deep learning.
Ng has been a consistent advocate for the view that AI is a general-purpose technology, comparable to electricity, that will transform every industry. He has been optimistic about the pace of progress and relatively skeptical of claims that near-term AI systems pose existential risks. He has argued that the most immediate AI risks are economic displacement, not superintelligent takeover, and that the appropriate response is education, safety nets, and retraining. His optimism about AI's benefits and his focus on democratizing access to AI education have made him one of the field's most effective ambassadors.