Construction & InfrastructureBreakthroughs
Who Really Invented AI? The Engineers Behind It, From 1958 to Now
AI was not one invention. From a 1958 learning machine to a bedroom in Toronto and the 2017 transformer, meet the engineers who built it step by step.
Key points
- AI was not one invention. It was built in stages by many engineers and scientists over almost 70 years.
- Frank Rosenblatt's perceptron learned to sort cards in 1958, and his Mark I Perceptron was built to recognise images.
- Bernard Widrow and Ted Hoff invented the Least Mean Squares algorithm at Stanford; Hoff later helped design the Intel 4004 microprocessor.
- Backpropagation (1986), ImageNet, GPU computing and AlexNet (2012) made deep neural networks practical.
- The 2017 transformer paper and the 2018 GPT paper led to today's chatbots, and the same ideas power AlphaFold.
- The 2025 Queen Elizabeth Prize for Engineering went to seven people for algorithms, hardware and data behind modern machine learning.
Ask who invented AI and you’ll probably hear the names of the tech bros or big multinational companies. The people who actually built it don’t get credit and the simple truth is AI wasn’t invented by a single person.
It’s more of a relay race, starting in 1958 with the baton passed from team to team. A surprisingly high number of the runners were engineers who had built real machines. Here are some of them, with a brief profile of each individual at the end.
1958: Frank Rosenblatt and a machine that learned from its mistakes
In July 1958 the US Office of Naval Research demonstrated a room-sized IBM 704 that had taught itself something.[1] After 50 trials it could distinguish between cards marked on the left and cards marked on the right.[2] The program was the perceptron, developed by Frank Rosenblatt, a research psychologist and project engineer at the Cornell Aeronautical Laboratory in Buffalo, New York.[3]
Rosenblatt then built it as hardware. The Mark I Perceptron was an electronic neural network designed to recognise images, such as letters of the alphabet.[4] The fundamental idea behind it is still in use today: make a guess, check it, and tweak the connections when it’s wrong. The Mark I is now in the Smithsonian and the IEEE has an award named after him.ow, Ted Hoff and a neuron in a shoebox
At Stanford Widrow (then a professor of electrical engineering) and his first grad student Ted Hoff came up with the Least Mean Squares algorithm.[5][6] That allowed a machine to measure its own error and correct it step by step. A year later Widrow had applied it to train ADALINE, a black shoebox of knobs and switches which was in effect a single artificial neuron.[7] The algorithm got out of the lab and tuned filters in telephone modems, mobile phones, and seismic monitors.[8] Hoff’s career took a further turn. Having joined Intel in 1968, he worked with Stan Mazor on the architecture of the Intel 4004, the world’s first commercial microprocessor.[9] Not a bad CV for one engineer.er.
1986: teaching networks with more than one layer
Single layer networks had their limits and by the early 1970s most AI researchers had given up on them.[10] In 1986 Hinton, Rumelhart, and Williams published work demonstrating that backpropagation could train very deep networks. Backprop works by going backwards from the wrong answer and adjusting each layer in turn. It’s still the engine of deep learning. Hinton went on to share the 2024 Nobel Prize in Physics with Hopfield for their pioneering work on artificial neural networks.[11]
Yann LeCun (postdoc in Hinton’s group in Toronto at the time) then demonstrated that backprop through a convolutional neural network was excellent at reading handwritten digits.[12][13]
2009 to 2012: the data, the chips and a bedroom in Toronto
Clever algorithms still required two things: massive amounts of data and fast, cheap computing. Fei-Fei Li provided the first. From 2006 she began building a collection of images for every noun in the English language and had the labelling done via crowd-sourcing. The result was ImageNet, a huge collection of images with labels for every noun in the English language.[14]
The chips were gaming chips. The graphics processors were designed to do a lot of arithmetic at once and NVIDIA’s CUDA platform, launched in 2007, made them programmable for other purposes.[15] In 2011 Ilya Sutskever convinced fellow Toronto student Alex Krizhevsky to train a network on ImageNet with Hinton as project lead. Krizhevsky trained it on a computer with two NVIDIA graphics cards in his bedroom at his parents’ house.[16][17]
The network AlexNet changed computer vision overnight. Before it, virtually no leading CV papers used neural networks. After it, almost all did.[18] Hinton’s own summary is the best: “Ilya thought we should do it, Alex made it work and I got the Nobel Prize.”[19]
2017: eight names and an idea called attention
In June 2017 eight researchers (Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan Gomez, Łukasz Kaiser and Illia Polosukhin) published a paper called Attention Is All You Need. The paper proposed the transformer, a network which is entirely based on the attention mechanism.[20]
Rather than process a sentence word by word, the transformer looks at every word in relation to every other word. In the example “I arrived at the bank after crossing the river” it can immediately link “bank” to “river” and work out it means a river bank rather than a financial institution. It was quick to train. The best translation model in the paper took 3.5 days to train on 8 GPUs.[21] Google announced it in a blog post by Uszkoreit, who was a software engineer.[22]
2018: the first GPT
A year later Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever put the transformer to work on language.[23] It was a two-stage process. The model first read a massive amount of plain text and learned to predict what would come next. It was then fine-tuned on smaller sets of examples for various specific tasks.[24] The training text was drawn from over 7000 books that had never been published. That first stage of generative pre-training gives GPT its name.
Not just chatbots
The same family of ideas is also being applied in serious science. In 2024 Demis Hassabis and John Jumper from Google DeepMind were awarded the Nobel Prize in Chemistry for AlphaFold, which can predict the 3D structure of a protein given just its amino acid sequence.[25] The predictions are completely free to use and have already assisted over 2 million researchers in 190 countries.[26]
So who invented AI?
Everyone above and thousands more whose names didn’t get on the paper. That’s the verdict of the UK’s most prestigious engineering award. The Queen Elizabeth Prize for Engineering in 2025 was awarded to seven individuals for their work in modern machine learning: Yoshua Bengio, Bill Dally, Geoffrey Hinton, John Hopfield, Jensen Huang, Yann LeCun, and Fei-Fei Li.[27] The judges cited the combined impact of algorithms, hardware, and data, with Huang and Dally specifically recognised for turning graphics processors into the engines of AI.[28] The prize money of £500,000 was shared by the seven and announced at the Science Museum in London.[29]
The common thread is people who built things, tested them, and kept going when everyone else had moved on. Rosenblatt wired up circuits. Widrow and Hoff built a box with knobs. Krizhevsky got every last drop of performance out of two graphics cards. If that’s the sort of thing you’re after, our guide to studying and working in electronic engineering and our look at how a Computer Science A level can get you into engineering are worth a read. Which engineering discipline is right for you? will help you work out which one. For a taster of hardware design, have a look at how students at Bristol designed their own computer chips in one workshop. The QE Prize this year went to the engineers who wired machines to the nervous system.
Engineering deep dive
Meet the engineers behind AI
Here is everyone named in the story, in roughly the order they appear. Each profile links to the sources we checked.
Frank Rosenblatt
Born in 1928, Rosenblatt grew up in New Rochelle, New York, went to the Bronx High School of Science and studied social psychology at Cornell, where he also earned his PhD. After building the Mark I Perceptron he moved to Cornell's Ithaca campus in 1959 to run its Cognitive Systems Research Program. His team's next machine, the Tobermory, was a perceptron designed to recognise speech.
Source: Cornell Chronicle
Bernard Widrow
Widrow spent three years on the faculty at MIT and half a century at Stanford, modelling machines on the way people think. His work led to 21 patents and sits at the core of AI, telephone modems and heart monitors. Two of his ADALINE machines are now looked after by the Smithsonian.
Source: Stanford Magazine
Ted Hoff
Marcian "Ted" Hoff was born in Rochester, New York, in 1937. He studied electrical engineering at Rensselaer Polytechnic Institute, then took a master's and a PhD at Stanford. He joined Intel in 1968 and came up with the idea of replacing custom-designed circuits with one universal processor. In 1980 he became the first Intel Fellow, and in 2009 he shared the US National Medal of Technology with Federico Faggin and Stan Mazor.
Source: Computer History Museum
Stan Mazor
Mazor was born in Chicago in 1941 and studied mathematics at San Francisco State University. In 1971, as part of a team that included Federico Faggin and Masatoshi Shima, he developed the Intel 4004. He also wrote software for the new chip.
Source: Computer History Museum
David Rumelhart
Rumelhart was a cognitive scientist who began his teaching career at the University of California, San Diego, and moved to Stanford in 1987. With James McClelland he wrote Parallel Distributed Processing, a book on connectionism, the idea that no single neuron in the brain does its job alone.
Source: Stanford Report
Ronald Williams
Williams studied for his degree and PhD in his native Southern California and worked for a defence contractor before joining Northeastern University in 1986. He co-wrote the 1986 backpropagation paper, "Learning representations by back-propagating errors", with Rumelhart and Hinton, and went on researching neural networks and reinforcement learning.
Source: Khoury College, Northeastern University
Geoffrey Hinton
Hinton was born in London in 1947. He studied experimental psychology at Cambridge, graduating in 1970, earned his PhD in artificial intelligence at the University of Edinburgh in 1978 and is a professor at the University of Toronto. From 2013 to 2023 he also worked half-time for Google, where he became a VP Engineering Fellow. As well as backpropagation, his work includes the Boltzmann machine.
Sources: Royal Swedish Academy of Sciences, Queen Elizabeth Prize for Engineering
John Hopfield
Hopfield was born in Chicago in 1933, earned his PhD at Cornell in 1958 and became a professor at Princeton. He created a network that can store patterns, such as images, and rebuild them from incomplete copies. He taught physics at Princeton for 16 years and helped set up the Princeton Neuroscience Institute.
Sources: Royal Swedish Academy of Sciences, Princeton University
Yann LeCun
LeCun trained as an engineer, with an Engineering Diploma from ESIEE in Paris, and earned his PhD at Sorbonne Université. After his postdoc in Toronto he joined AT&T Bell Labs in 1988. His ideas ended up in a system used by AT&T and NCR to read handwritten cheques.
Sources: Queen Elizabeth Prize for Engineering, ACM via EurekAlert!
Yoshua Bengio
Bengio is a full professor at the Université de Montréal and the founder and scientific director of Mila, the Quebec AI Institute. He shared the 2018 Turing Award with Hinton and LeCun for breakthroughs that made deep neural networks a central part of computing.
Sources: Queen Elizabeth Prize for Engineering, ACM
Fei-Fei Li
Li graduated from Princeton in 1999 and was back there on the faculty when she started building ImageNet. She is now the Sequoia Capital Professor in Computer Science at Stanford and a founding co-director of the Stanford Institute for Human-Centered Artificial Intelligence.
Sources: Princeton University, Queen Elizabeth Prize for Engineering
Jensen Huang
Huang studied electrical engineering, with a bachelor's degree from Oregon State University and a master's from Stanford. He founded NVIDIA in 1993 and has led it ever since. The company's GPU, invented in 1999, first took off in PC gaming.
Source: Queen Elizabeth Prize for Engineering
Bill Dally
Dally has spent his career building computers that work in parallel. At Caltech he designed the Torus Routing chip. At MIT his team built the J-Machine and M-Machine, two experimental parallel computers. At Stanford he chaired the computer science department. He joined NVIDIA as chief scientist in January 2009.
Source: Queen Elizabeth Prize for Engineering
Alex Krizhevsky
Krizhevsky was a graduate student at the University of Toronto with a knack for getting the most out of GPUs. Before AlexNet he had already written cuda-convnet, a GPU-powered network trained on a much smaller image dataset. AlexNet is named after him.
Source: Computer History Museum
Ilya Sutskever
Sutskever was a Toronto graduate student who believed neural networks would keep getting better as they were given more data. He co-wrote both the AlexNet paper and the first GPT paper, and co-founded OpenAI.
Sources: Computer History Museum, OpenAI
The transformer team
The eight authors of Attention Is All You Need listed their names in random order and set out who did what.
- Jakob Uszkoreit (Google Research) proposed replacing the older recurrent networks with self-attention.
- Ashish Vaswani (Google Brain) and Illia Polosukhin (Google Research) designed and built the first transformer models.
- Noam Shazeer (Google Brain) proposed key parts of the design, including multi-head attention.
- Niki Parmar (Google Research) designed, built, tuned and tested countless versions of the model.
- Llion Jones (Google Research) was responsible for the team's first codebase.
- Łukasz Kaiser (Google Brain) and Aidan Gomez (University of Toronto, working at Google Brain) designed and built tensor2tensor, the software library the project ran on.
Source: Attention Is All You Need, arXiv
The GPT team
Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever were all at OpenAI when they wrote the 2018 paper, with Radford named first.
Source: OpenAI
Demis Hassabis
Hassabis studied computer science at the University of Cambridge from 1994 to 1997, then took a PhD in cognitive neuroscience at University College London. He co-founded DeepMind in 2010 and is chief executive of Google DeepMind. He was knighted in 2024 for services to artificial intelligence.
Sources: University of Cambridge, Google DeepMind
John Jumper
Jumper took an MPhil in theoretical condensed matter physics at Cambridge's Cavendish Laboratory in 2011, then a PhD in chemistry at the University of Chicago. He is a director at Google DeepMind.
Sources: University of Cambridge, Google DeepMind
Links to the curriculum: A level, GCSE
Sources
29 sources
Professor’s perceptron paved the way for AI-60 years too soon
In July 1958, the U.S. Office of Naval Research unveiled a remarkable invention.
Checked 31 October 2026
Professor’s perceptron paved the way for AI-60 years too soon
After 50 trials, the computer taught itself to distinguish cards marked on the left from cards marked on the right.
Checked 31 October 2026
Professor’s perceptron paved the way for AI-60 years too soon
was a research psychologist and project engineer at the Cornell Aeronautical Laboratory in Buffalo, New York.
Checked 31 October 2026
CHM Releases AlexNet Source Code
Cornell University psychologist Frank Rosenblatt developed the Perceptron Mark I, an electronic neural network designed to recognize images—like letters of the alphabet. He introduced it to the public in 1958.
Checked 31 October 2026
Trailblazer of Computers that Learn
Soon after Widrow arrived at Stanford in 1959, he and his first graduate student, Marcian “Ted” Hoff, MS ’59, PhD ’62, invented an algorithm that would become one of the most widely used learning tools in engineering.
Checked 31 October 2026
Oral history of Bernard Widrow
in the autumn of 1959, together with Marcian "Ted" Hoff Jr. his first doctoral student at Stanford, he invented the LMS algorithm, which underpins much of modern digital communications.
Checked 31 October 2026
Trailblazer of Computers that Learn
Within a year, Widrow had used LMS to train a black, shoebox-sized device with an array of knobs and switches called ADALINE, for adaptive linear neuron—essentially a single neuron of a neural network
Checked 31 October 2026
Trailblazer of Computers that Learn
From telephone modems to mobile phones to seismic monitors, the approach “tunes” filters that equalize high and low frequencies
Checked 31 October 2026
Hoff joined Intel in 1968 and is credited with the idea of using a universal processor to replace custom-designed circuits. ... Working with Stan Mazor, Hoff defined the instruction set and architectural specifications of the new chip, known as the Intel 4004.
Checked 31 October 2026
CHM Releases AlexNet Source Code
By the early 1970s, neural networks had been largely rejected by AI researchers.
Checked 31 October 2026
Press release: The Nobel Prize in Physics 2024
“for foundational discoveries and inventions that enable machine learning with artificial neural networks”
Checked 31 October 2026
CHM Releases AlexNet Source Code
A postdoctoral student of Hinton’s at this time was Yann LeCun. While working in Toronto he showed that when backpropagation was used in “convolutional” neural networks they became very good at recognizing hand-written numbers.
Checked 31 October 2026
Fathers of the Deep Learning Revolution Receive ACM A.M. Turing Award
Convolutional neural networks: In the 1980s, LeCun developed convolutional neural networks, a foundational principle in the field, which, among other advantages, have been essential in making deep learning more efficient.
Checked 31 October 2026
2025 QEPrize Winners: Modern Machine Learning
By creating ImageNet, a large-scale image database used for object recognition software research, she enabled access to millions of labelled images that have been instrumental in training and evaluating computer vision algorithms.
Checked 31 October 2026
CHM Releases AlexNet Source Code
NVIDIA, headed by cofounder and CEO Jensen Huang, had led the way in the 2000s in making GPUs more generalizable and programmable for applications beyond 3D graphics, especially with the CUDA programming system, released in 2007.
Checked 31 October 2026
CHM Makes AlexNet Source Code Available to the Public
In 2011, Sutskever persuaded Krizhevsky, a fellow graduate student, to train a neural network for ImageNet. With Hinton serving as faculty advisor, Krizhevsky did so on a computer with two NVIDIA cards.
Checked 31 October 2026
CHM Releases AlexNet Source Code
The training was done on a computer with two NVIDIA cards in Krizhevsky’s bedroom at his parents’ house.
Checked 31 October 2026
CHM Releases AlexNet Source Code
Before AlexNet, almost none of the leading computer vision papers used neural nets. After it, almost all of them would.
Checked 31 October 2026
CHM Releases AlexNet Source Code
In describing the AlexNet project, Geoff Hinton summarized for CHM: “Ilya thought we should do it, Alex made it work, and I got the Nobel Prize.”
Checked 31 October 2026
We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely.
Checked 31 October 2026
On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature.
Checked 31 October 2026
Transformer: A Novel Neural Network Architecture for Language Understanding
Posted by Jakob Uszkoreit, Software Engineer, Natural Language Understanding
Checked 31 October 2026
Improving language understanding with unsupervised learning
Our approach is a combination of two existing ideas: transformers and unsupervised pre-training.
Checked 31 October 2026
Improving Language Understanding by Generative Pre-Training
We demonstrate that large gains on these tasks can be realized by generative pre-training of a language model on a diverse corpus of unlabeled text, followed by discriminative fine-tuning on each specific task.
Checked 31 October 2026
Demis Hassabis & John Jumper awarded Nobel Prize in Chemistry
Sir Demis Hassabis, and Google DeepMind Director Dr. John Jumper were co-awarded the 2024 Nobel Prize in Chemistry for their work developing AlphaFold, a groundbreaking AI system that predicts the 3D structure of proteins from their amino acid sequences.
Checked 31 October 2026
Demis Hassabis & John Jumper awarded Nobel Prize in Chemistry
AlphaFold’s predictions, made freely available through the AlphaFold Protein Structure Database, have given more than 2 million scientists and researchers from 190 countries a powerful tool for making new discoveries.
Checked 31 October 2026
2025 QEPrize Winners: Modern Machine Learning
The 2025 Queen Elizabeth Prize for Engineering is awarded to seven engineers who have made seminal contributions to the development of Modern Machine Learning, a core component of artificial intelligence (AI) advancements.
Checked 31 October 2026
2025 QEPrize Winners: Modern Machine Learning
Jensen Huang and Bill Dally have led developments in the hardware platforms that underpin the operation of modern machine learning algorithms. The vision of exploiting Graphics Processing Units and their subsequent architectural advances have enabled the scaling that has been central to their successful application.
Checked 31 October 2026
Modern machine learning innovators win 2025 Queen Elizabeth Prize for Engineering
The 2025 Laureates, who will share the £500,000 prize, were introduced by Lord Vallance, Chair of the Queen Elizabeth Prize for Engineering Foundation, during a reception at the Science Museum
Checked 31 October 2026
Share
Related stories
All Electronics & CommunicationsHealthInnovation
Neural Interface Pioneers Win 2026 Queen Elizabeth Prize
Electronics & Communications
Nobel prize: physicists share prize for insights into the spooky world of quantum mechanics
