Jürgen Schmidhuber

He has said that since about the age of 15 his main goal has been to build a self-improving artificial intelligence smarter than himself, and then retire.⁠[1]

In 1987 he completed his diploma thesis at the Institut für Informatik of TU Munich on self-referential learning and “learning how to learn,” and published his first paper on genetic programming the same year.⁠[2]

He regards the thesis as presenting the first concrete recursive self-improvement (RSI) algorithms and the start of decades of his work on meta-learning.⁠[3]

He has said that in 1990–1991 his lab at TU Munich published many basic ideas of modern deep learning, including the adversarial generative principle for artificial curiosity, the unnormalised linear Transformer, pre-training of deep networks, neural network distillation and recurrent world models, though few people were interested at the time.⁠[4]

From 1990 he built “curious” agents combining a predictive model with a reinforcement learner rewarded by the predictor’s learning progress, the basis of his formal theory of creativity, fun and intrinsic motivation.⁠[5]

He was a postdoc at the University of Colorado at Boulder from July 1991 to December 1992, Oberassistent at TU Munich from October 1993 and Privatdozent from January 1995.⁠[6]

From March 1995 he has been scientific director of the Swiss AI Lab IDSIA in Lugano; he was professor of AI at the University of Lugano (USI) from 2009 to 2021.⁠[6]

In 1997 he and his PhD student Sepp Hochreiter published Long Short-Term Memory in Neural Computation, which he calls the most cited neural network of the 20th century. He recalls that he married and became a father that year.⁠[7]

His wife is Ulrike Krommer, and they have two daughters; his younger brother, Christof Schmidhuber, is a theoretical physicist turned finance expert.⁠[1]

In 2003 he released a preprint on the Gödel machine, a self-referential universal problem solver that rewrites any part of itself once it proves the rewrite will improve its future performance, formalising I. J. Good’s idea of an “intelligence explosion.”⁠[8]

In 2009–2010 his team’s CTC-trained LSTM beat all competing algorithms on important problems such as connected handwriting recognition, which he describes as triggering the supervised deep learning revolution.⁠[9]

In 2011 DanNet, named after his postdoc Dan Cireșan, became the first pure deep convolutional neural network to win computer vision contests, winning four in a row in 2011–2012 on NVIDIANVIDIANVIDIA is an American semiconductor and computing technology company that provides graphics processing units, artificial intelligence infrastructure, networking interconnects, and computing software, and is listed on Nasdaq under the ticker NVDA.Open the full entry GPUs and achieving the first superhuman performance in a vision challenge.⁠[10]

Founded by AI pioneers, NNAISENSE has worked since 2014 with ACATIS Investment and other financial institutions on AI-based stock selection and portfolio construction, launching its first AI fund in 2017.⁠[11]

In May 2015 his team published Highway Networks, the first working very deep gradient-based feedforward networks with hundreds of layers; he notes that ResNet, published seven months later, is equivalent to a Highway Net with all gates open.⁠[9]

He received the IEEE CIS Neural Networks Pioneer Award at the IJCNN 2016 award banquet in Vancouver for “pioneering contributions to deep learning and neural networks”; in his acceptance speech he called the Ukrainian mathematician A. G. Ivakhnenko the father of deep learning.⁠[12]

He argued that some existing AI systems are already conscious, saying: “I would like to claim we had little, rudimentary, conscious learning systems for at least 25 years.”⁠[13]

In a 2017 article he wrote that deep learning methods his team developed since 1991 had reached billions of users through the world’s most valuable public companies, including Apple, Alphabet, MicrosoftMicrosoftMicrosoft is a technology company founded by Bill Gates and Paul Allen in 1975 and headquartered in Redmond, Washington, with businesses spanning Windows, Microsoft 365, Azure, enterprise software, artificial intelligence, gaming and devices.Open the full entry, Facebook and Amazon.⁠[14]

He appeared on episode 11 of the Lex Fridman Podcast, titled “Godel Machines, MetaMetaMeta (Meta Platforms, Inc.) is a publicly listed American technology company, renamed from Facebook in 2021, that operates social communication, digital advertising, virtual reality, smart glasses and artificial intelligence businesses.Open the full entry-Learning, and LSTMs”; the episode description presents him as co-creator of LSTM networks and the proposer of many unconventional AI ideas, including a formal theory of creativity.⁠[15]

He published a long critique of ACMAustralian Community MediaAustralian Community Media (ACM) is an Australian regional and rural news publishing group with more than 100 brands, including The Canberra Times; in 2019 Nine sold it to a company controlled by interests associated with Antony Catalano and Thorney Investment Group.Open the full entry’s citation for the 2018 Turing Award to Yoshua BengioYoshua BengioYoshua Bengio is a professor of computer science at Université de Montréal, founder of Mila and a deep learning pioneer who shared the 2018 Turing Award with Geoffrey Hinton and Yann LeCun; since 2023 he has turned to AI safety research, chairing the International AI Safety Report and founding the non-profit LawZero in 2025. He appeared on the Lex Fridman Podcast in 2018.Open the full entry, Geoffrey HintonGeoffrey HintonMachine learning researcher.Open the full entry and Yann LeCun, arguing that the awardees’ work rested on algorithms and concepts first published by other researchers whom they failed to cite, and listing eight direct priority disputes with Bengio and Hinton.⁠[16]

From October 1, 2021, he has directed the Artificial Intelligence Initiative at King Abdullah University of Science and Technology (KAUST) in Saudi Arabia, while remaining chief scientist of NNAISENSE, scientific director of IDSIA and a part-time adjunct professor at the University of Lugano.⁠[17]

He wrote that Yann LeCun’s 2022 paper on autonomous machine intelligence rehashed essential work from 1990–2015 without referencing it.⁠[18]

He released an IDSIA technical report alleging that the three 2018 Turing laureates often republished methods first described by his team without proper attribution, listing seven direct priority disputes with Bengio, six with Hinton and four with LeCun.⁠[19]

He released a technical report calling the 2024 Nobel Prize in Physics awarded to John Hopfield and Geoffrey Hinton “effectively a prize for plagiarism,” arguing that they republished methods developed by Ivakhnenko, Amari and others in Ukraine and Japan in the 1960s and 1970s without referencing the original papers.⁠[20]

In 2025 longtime partner ACATIS Investment acquired NNAISENSE, which is building a generative-AI-driven quantitative multi-strategy hedge fund.⁠[11]

In an article updated in March 2026 he argued that the most cited neural networks, including LSTM, the Highway Net and its variant ResNet, DanNet, GANs and Transformers, all build on work done in his labs.⁠[21]

Personal website: Public link: https://people.idsia.ch/~juergen/ (opens in a new window)⁠[1]

X: Public account: https://x.com/SchmidhuberAI (opens in a new window)⁠[1]

IDSIA: Official website: https://idsia.usi-supsi.ch/ (opens in a new window)⁠[1]

KAUST: Official website: https://cemse.kaust.edu.sa/ai (opens in a new window)⁠[1]

NNAISENSE: Official website: https://nnaisense.com/ (opens in a new window)⁠[11]