Jürgen Schmidhuber

Jürgen Schmidhuber
Schmidhuber speaking at the AI for GOOD Global Summit in 2017
Jürgen Schmidhuber
Born	17 1, 1963
Birthplace	Munich, West Germany
Nationality	German
Occupation	Computer scientist, researcher, professor
Employer	Dalle Molle Institute for Artificial Intelligence Research, King Abdullah University of Science and Technology (KAUST)
Known for	Long short-term memory (LSTM), deep learning, Gödel machine, artificial curiosity, meta-learning
Education	Technical University of Munich
Awards	IEEE Neural Networks Pioneer Award, INNS Helmholtz Award
Website	[[people.idsia.ch/~juergen/ people.idsia.ch/~juergen/] Official site]

Jürgen Schmidhuber (born 17 January 1963) is a German computer scientist whose work on artificial neural networks has shaped the modern landscape of artificial intelligence. As a scientific director of the Dalle Molle Institute for Artificial Intelligence Research (IDSIA) in Lugano, Switzerland, and a professor at the King Abdullah University of Science and Technology (KAUST) in Saudi Arabia, Schmidhuber has spent more than three decades developing foundational techniques in deep learning, recurrent neural networks, and self-improving systems. He is best known as a co-developer of long short-term memory (LSTM) networks, a recurrent neural network architecture that became the dominant approach for natural language processing tasks across research and industry during the 2010s. In addition to LSTM, Schmidhuber has introduced or contributed to principles underlying dynamic neural networks, meta-learning, generative adversarial networks, and linear transformers — concepts that are widespread in contemporary AI systems. Media outlets have described him as a leading pioneer of modern artificial intelligence, and The New York Times suggested that if AI matures, "it may call Jürgen Schmidhuber 'Dad.'"

Early Life

Jürgen Schmidhuber was born on 17 January 1963 in Munich, then part of West Germany. From an early age, Schmidhuber expressed an ambition to create a machine more intelligent than himself — a goal he has described repeatedly throughout his career and which motivated his subsequent research path. According to a BBVA profile, as a young man growing up in Germany, Schmidhuber became fascinated with the idea of building a self-improving artificial intelligence that could eventually surpass the cognitive abilities of its creator. This childhood aspiration would later crystallize into a decades-long research program spanning self-referential learning systems, artificial curiosity, and mathematically grounded frameworks for machine self-improvement.

Schmidhuber grew up during a period when artificial intelligence research was passing through one of its so-called "AI winters" — eras of reduced funding and diminished public interest in the field. Despite the relatively hostile academic environment for AI research during much of the 1970s and 1980s, he pursued his interest in intelligent machines and neural computation. In a 2016 interview with The New York Times, Schmidhuber reflected on how much of his early work was "often overlooked or ignored" by the broader research community, suggesting that the intellectual climate of the time was not always receptive to the ambitious claims he was making about the potential of neural network-based approaches.

Little additional information about Schmidhuber's family background or childhood has been documented in published sources. His public biographical materials focus primarily on his academic and professional trajectory beginning with his studies at the Technical University of Munich.

Education

Schmidhuber pursued his higher education at the Technical University of Munich (TUM), one of Germany's leading research universities. His academic work at TUM focused on computer science, with an early specialization in neural networks and machine learning — subjects that were at the time considered fringe topics within mainstream computer science.

In 1987, Schmidhuber completed his diploma thesis, titled Evolutionary principles in self-referential learning, or on learning how to learn: the meta-meta-... hook. The thesis explored the concept of meta-learning — the idea of building systems that can learn to improve their own learning processes. This work established a theoretical foundation for self-referential and self-improving systems that would become central themes throughout Schmidhuber's subsequent career. He went on to earn his doctorate from TUM as well, continuing his research on neural networks and learning algorithms.

Career

Early Research and LSTM

Schmidhuber's most widely recognized contribution to the field of artificial intelligence is his role in the development of long short-term memory (LSTM) networks. LSTM is a type of recurrent neural network (RNN) architecture designed to address the fundamental problem of learning long-range dependencies in sequential data — a challenge known as the vanishing gradient problem, which had severely limited the effectiveness of earlier RNN designs. The original LSTM paper, co-authored with Sepp Hochreiter, was published in 1997 and introduced gated memory cells that could maintain information over extended time intervals.

LSTM networks became one of the most cited and commercially deployed neural network architectures in history. During the 2010s, LSTM was the dominant technique for a wide range of natural language processing tasks, including machine translation, speech recognition, text generation, and language modeling. Major technology companies including Google, Apple, Amazon, and Microsoft incorporated LSTM-based systems into their products and services. Google used LSTM for Google Translate and the voice recognition system in Android phones, while Apple adopted it for Siri and other natural language processing applications.

Despite its enormous influence, Schmidhuber has publicly expressed frustration that his contributions were not always properly credited in subsequent research that built upon LSTM and related techniques. In the 2016 New York Times profile, he described how his early work was "often overlooked or ignored," and he has been known to write detailed letters and blog posts pointing out what he views as insufficient attribution to his laboratory's contributions.

Deep Learning and Computer Vision

Beyond LSTM, Schmidhuber and his research groups at IDSIA made significant contributions to deep learning and computer vision. Teams supervised by Schmidhuber achieved notable results in multiple international pattern recognition and computer vision competitions. His group's deep neural networks, accelerated by graphics processing units (GPUs), won several contests in image classification and object recognition.

One prominent example was the German Traffic Sign Recognition Benchmark (GTSRB), where Schmidhuber's team achieved results that surpassed human performance in recognizing traffic signs — a milestone in computer vision that demonstrated the practical applicability of deep learning to safety-critical tasks such as autonomous driving. These competition successes helped build the empirical case for deep neural networks at a time when their effectiveness was still being debated within parts of the machine learning community.

Schmidhuber has written extensively on the history and development of deep learning, notably through a detailed survey published on Scholarpedia. In this and other publications, he has argued that many key ideas in modern deep learning — including convolutional neural networks, unsupervised pre-training, and certain training techniques — have historical roots in earlier research that is not always adequately cited in contemporary literature.

Gödel Machine and Self-Improving Systems

A central theme in Schmidhuber's research has been the development of theoretically grounded self-improving systems. His concept of the Gödel machine, named after the mathematician Kurt Gödel, is a theoretical framework for a self-referential, self-improving general problem solver. The Gödel machine is designed to rewrite its own code — including the self-rewriting mechanism itself — whenever it can prove that such a modification will improve its performance according to a defined utility function.

In November 2025, a research group at KAUST unveiled the Huxley-Gödel Machine (HGM), an AI agent described as a practical realization of Schmidhuber's theoretical vision. The HGM is a self-rewriting AI system that builds upon the foundational principles of the original Gödel machine concept. This development demonstrated the continued relevance of Schmidhuber's theoretical frameworks decades after their initial formulation.

Artificial Curiosity and Meta-Learning

Schmidhuber is also recognized for his early work on artificial curiosity — the principle that learning agents can be motivated by intrinsic rewards related to the novelty or surprise value of their experiences, rather than solely by external reward signals. This concept anticipated later developments in intrinsic motivation and curiosity-driven exploration that have become important research topics in reinforcement learning.

His 1987 diploma thesis on meta-learning — learning to learn — was among the earliest formal treatments of the subject. Meta-learning has since become a major subfield of machine learning, with applications in few-shot learning, transfer learning, and adaptive systems. Schmidhuber's contributions in this area span both theoretical formulations and practical algorithms.

Generative Adversarial Networks and Transformers

Schmidhuber has made claims to early work on principles that underlie both generative adversarial networks (GANs) and the transformer architecture that now dominates modern AI. He has pointed to his 1990s work on predictability minimization and adversarial training schemes as precursors to the GAN framework popularized by Ian Goodfellow and colleagues in 2014. Similarly, he has highlighted connections between his group's work on linear transformers and the attention mechanisms that form the basis of the transformer architecture introduced in the influential 2017 "Attention Is All You Need" paper.

These attribution claims have been a source of both scholarly discussion and public controversy. Schmidhuber has been a persistent advocate for what he views as proper credit assignment in the history of AI, a stance that has sometimes generated friction with other researchers in the field.

Dalle Molle Institute for Artificial Intelligence Research (IDSIA)

Schmidhuber has served as a scientific director (and previously co-director) of the Dalle Molle Institute for Artificial Intelligence Research (IDSIA) in Lugano, Switzerland. IDSIA is a research institute affiliated with both the Università della Svizzera italiana (USI) and the Scuola universitaria professionale della Svizzera italiana (SUPSI). Under Schmidhuber's leadership, IDSIA became one of the most productive and highly cited AI research laboratories in Europe, with particular strengths in deep learning, reinforcement learning, and neural network architectures.

King Abdullah University of Science and Technology (KAUST)

In addition to his role at IDSIA, Schmidhuber holds the position of director of the Artificial Intelligence Initiative and professor of computer science within the Computer, Electrical, and Mathematical Sciences and Engineering (CEMSE) division at the King Abdullah University of Science and Technology (KAUST) in Saudi Arabia. In a 2025 profile by Rest of World, Schmidhuber discussed his reasons for working in Saudi Arabia, stating that the kingdom's funding of AI research initiatives could benefit the entire world and potentially lead to "a new golden age for science." At KAUST, Schmidhuber's group has continued work on self-improving systems, neural network architectures, and the practical realization of theoretical AI concepts such as the Gödel machine.

Views on the Future of AI

Schmidhuber has been vocal in his views about the trajectory of artificial intelligence. In a 2023 interview with The Guardian, he stated his belief that AI will progress to the point where it surpasses human intelligence but argued that this development should not be feared. He suggested that superintelligent AI systems would eventually pay little attention to humans, much as humans pay little attention to ants — not out of malice, but simply because their interests and capabilities would diverge so dramatically.

He has also expressed his long-standing goal of creating an AI that is more intelligent than himself, a vision he has articulated consistently since his youth. In a 2017 interview with Bloomberg, he discussed his ambition to build what he called the "Renaissance Machine of the future" — a system capable of creativity, scientific discovery, and general problem-solving at a level exceeding human ability.

Personal Life

Schmidhuber has generally kept his personal life separate from his public professional persona. Publicly available sources focus almost exclusively on his academic and research activities. He has maintained professional bases in both Switzerland, where IDSIA is located, and Saudi Arabia, where he holds his position at KAUST.

Schmidhuber is known within the AI research community for his assertive communication style, particularly regarding questions of credit and priority in the history of artificial intelligence. He has written numerous public letters, blog posts, and academic commentaries arguing that various foundational contributions from his laboratory have been insufficiently cited or acknowledged by later researchers. This tendency has made him a sometimes polarizing figure within the field, though it has also drawn increased attention to the historical record of AI research.

In media appearances, Schmidhuber has often conveyed an optimistic view of artificial intelligence's potential to transform civilization, drawing historical parallels to previous technological revolutions and emphasizing what he sees as the inevitability of progress toward artificial general intelligence.

Recognition

Schmidhuber's contributions to artificial intelligence have been recognized through several major awards and honors. He received the IEEE Neural Networks Pioneer Award from the IEEE Computational Intelligence Society, an honor given to individuals who have made outstanding contributions to the field of neural networks.

He has also been a recipient of awards from the International Neural Network Society (INNS). Schmidhuber is a member of the European Academy of Sciences and Arts, which recognizes scientists who have made distinguished contributions to their fields.

Media recognition has been extensive. The New York Times profiled him in a 2016 feature article exploring his foundational role in modern AI, while The Guardian has referred to him as the "father of AI" in headlines. Bloomberg profiled his research ambitions in 2017, and BBVA described him as "now considered the father of modern artificial intelligence" in a 2024 podcast. The Chinese technology publication Jazzyear profiled him in 2024 under the headline "The Father of Generative AI Without Turing Award," reflecting both the recognition of his contributions and the observation that he has not received the ACM Turing Award despite the widespread impact of his work.

Legacy

Jürgen Schmidhuber's influence on artificial intelligence is most clearly measured through the widespread adoption of the techniques he helped develop. LSTM networks, his most cited contribution, became a foundational building block for natural language processing, speech recognition, machine translation, and time series analysis throughout the 2010s. Though the transformer architecture has since supplanted LSTM as the dominant paradigm for many NLP tasks, LSTM remains widely used in various applications and its conceptual contributions to gated recurrent architectures continue to influence neural network design.

His theoretical work on self-improving systems, particularly the Gödel machine, has established a formal framework for thinking about recursive self-improvement in AI — a concept of increasing relevance as AI systems become more capable. The 2025 development of the Huxley-Gödel Machine at KAUST demonstrated that these ideas are moving from pure theory toward practical implementation.

Schmidhuber's work on artificial curiosity and intrinsic motivation has influenced the subfield of curiosity-driven exploration in reinforcement learning, where agents learn to seek novel experiences rather than relying solely on externally provided reward signals. His contributions to meta-learning have similarly seeded a now-thriving research area concerned with building systems that can adapt rapidly to new tasks.

Beyond specific technical contributions, Schmidhuber has played a distinctive role as a historian and advocate within AI research. His detailed writings on the history of deep learning, neural networks, and related topics have prompted increased scrutiny of attribution practices in the field. Whether one agrees with all of his priority claims, his persistent advocacy for careful credit assignment has contributed to a broader conversation about how scientific contributions are recognized in a rapidly evolving discipline.

His research groups at IDSIA and KAUST have trained numerous doctoral students and postdoctoral researchers who have gone on to contribute to AI research and development at academic institutions and technology companies worldwide. The combination of theoretical ambition and empirical rigor that has characterized Schmidhuber's research program continues to influence how the field approaches fundamental questions about learning, intelligence, and self-improvement in artificial systems.